Add GR00T-N1.7-LIBERO LIBERO smoke-test tooling, the 20260512_122756 10-rollout run artifacts (videos/frames/actions/plots/reports), setup logs, README/LICENSE/NOTICE. Built on NVIDIA Isaac-GR00T (Apache-2.0); upstream source / weights / gated backbone not included.

.gitattributes

@@ -33,3 +33,237 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00000.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00001.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00002.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00003.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00004.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00005.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00006.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00007.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00008.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00009.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00010.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00011.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00012.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00013.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00014.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00015.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00016.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00017.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00018.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00019.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00020.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00021.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00022.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00023.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy/frames/frame_00024.png filter=lfs diff=lfs merge=lfs -text
+outputs/libero_smoke_tests/20260512_122756/review_montage.mp4 filter=lfs diff=lfs merge=lfs -text

LICENSE

@@ -0,0 +1,190 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to the Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by the Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding any notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

NOTICE

@@ -0,0 +1,30 @@
+vla-sae-libero
+==============
+
+This repository contains tooling and run artifacts that build on, wrap, and run:
+
+  NVIDIA Isaac-GR00T  —  https://github.com/NVIDIA/Isaac-GR00T
+  Copyright (c) NVIDIA CORPORATION & AFFILIATES.
+  Licensed under the Apache License, Version 2.0 (see LICENSE).
+
+The following were authored as additions on top of that codebase and are
+likewise released under the Apache License, Version 2.0:
+
+  examples/LIBERO/smoke_tests/   (scenarios_10.yaml, run_10_smoke_tests.py,
+                                  _libero_rollout_worker.py, visualize_smoke_run.py,
+                                  make_video_montage.py, review_smoke_tests.py,
+                                  ACTIVATION_HOOK_NOTES.md, README.md)
+  tests/test_libero_smoke_tests.py
+
+NOT redistributed in this repository:
+  - the NVIDIA Isaac-GR00T source tree (obtain from the URL above);
+  - the model checkpoint  nvidia/GR00T-N1.7-LIBERO
+      (https://huggingface.co/nvidia/GR00T-N1.7-LIBERO);
+  - the gated VLM backbone  nvidia/Cosmos-Reason2-2B
+      (https://huggingface.co/nvidia/Cosmos-Reason2-2B);
+  - any Python virtual environments or third-party dependencies.
+
+The files under outputs/ are artifacts (rendered videos, frames, recorded
+action chunks, logs, plots, HTML reports, summaries) produced by running the
+nvidia/GR00T-N1.7-LIBERO model in the LIBERO simulator on CPU/GPU; they contain
+no model weights and no credentials.

README.md

@@ -0,0 +1,122 @@
+---
+license: apache-2.0
+language:
+- en
+pipeline_tag: robotics
+tags:
+- robotics
+- vla
+- vision-language-action
+- libero
+- gr00t
+- isaac-gr00t
+- imitation-learning
+- sae
+- mechanistic-interpretability
+- simulation
+pretty_name: "VLA-SAE — GR00T-N1.7-LIBERO LIBERO smoke-test tooling & run artifacts"
+---
+
+# vla-sae-libero — LIBERO smoke-test tooling & run artifacts for `nvidia/GR00T-N1.7-LIBERO`
+
+This repo holds **only** the smoke-test *tooling* and the *artifacts* of one completed run from deploying
+**[`nvidia/GR00T-N1.7-LIBERO`](https://huggingface.co/nvidia/GR00T-N1.7-LIBERO)** (the `libero_10` checkpoint)
+and running **10 short, simulation-only LIBERO rollouts** to verify the deploy/eval plumbing — a stepping
+stone toward activation-capture / SAE work on the GR00T VLA.
+
+> ## Built on NVIDIA Isaac-GR00T — *not redistributed here*
+> The tooling in this repo runs **on top of** [**NVIDIA Isaac-GR00T**](https://github.com/NVIDIA/Isaac-GR00T)
+> (Apache-2.0, © NVIDIA CORPORATION & AFFILIATES). **This repo does NOT contain the upstream Isaac-GR00T
+> source tree, the model weights, or the gated VLM backbone.** To use the tooling: clone Isaac-GR00T, install
+> its `gr00t` package, drop `examples/LIBERO/smoke_tests/` (and `tests/test_libero_smoke_tests.py`) into the
+> checkout, download the checkpoint, and follow the steps below.
+>
+> Specifically, **not included** (get them from upstream):
+> - the Isaac-GR00T source tree (`gr00t/`, `examples/` except `examples/LIBERO/smoke_tests/`, `getting_started/`, `scripts/`, `docker/`, `demo_data/`, `media/`, …) — <https://github.com/NVIDIA/Isaac-GR00T>
+> - the model checkpoint `nvidia/GR00T-N1.7-LIBERO` (~6.5 GB) — <https://huggingface.co/nvidia/GR00T-N1.7-LIBERO>
+> - the **gated** VLM backbone `nvidia/Cosmos-Reason2-2B` — <https://huggingface.co/nvidia/Cosmos-Reason2-2B>
+> - Python virtualenvs (`.venv/`, `gr00t/eval/sim/LIBERO/libero_uv/`) and the LIBERO submodule (`external_dependencies/LIBERO/`)
+
+## Contents
+
+```
+examples/LIBERO/smoke_tests/                 # the smoke-test tooling
+├── scenarios_10.yaml                        #   10-scenario manifest (8 normal + 2 abnormal_probe)
+├── run_10_smoke_tests.py                    #   runner: validates model/server, runs each scenario via the official eval path, writes summaries
+├── _libero_rollout_worker.py                #   single-episode worker (runs in the LIBERO venv): records action chunks, optional sim-only probes
+├── visualize_smoke_run.py                   #   builds visual_report.html + per-scenario action plots + visual_summary.{csv,json}
+├── make_video_montage.py                    #   builds review_montage.mp4 + per-scenario captioned clips + review_playlist.html (needs ffmpeg)
+├── review_smoke_tests.py                    #   CLI: list/play the rollouts of a run
+├── ACTIVATION_HOOK_NOTES.md                 #   where to insert future SAE activation capture (module names, hook points)
+└── README.md                                #   setup & run instructions
+
+tests/test_libero_smoke_tests.py             # lightweight tests for the manifest + summary writer
+
+outputs/libero_smoke_tests/20260512_122756/  # the completed run (10/10 rollouts OK, 0 errors; success=false everywhere — these are 50-step smoke rollouts, not a benchmark)
+├── summary.json, summary.md                 #   run-level results + the printed comparison table
+├── visual_report.html                       #   self-contained report: embedded videos + plots + comparison table
+├── visual_summary.csv, visual_summary.json  #   per-scenario action-stat table
+├── review_montage.mp4                        #   ~22 s: all 10 scenarios back-to-back with captions
+├── review_clips/<scenario_id>.mp4           #   10 short captioned clips
+├── review_playlist.html                     #   the montage + each clip + metadata + links
+├── plots/<scenario_id>/                     #   action_norm / action_mean_per_dof / gripper_over_time / action_delta_norm  (.png)
+└── <scenario_id>/                           #   ×10
+    ├── video.mp4                            #     rendered rollout (agentview | wrist, 512×256)
+    ├── frames/                              #     decoded PNG frames (25 each; 8 for the short-timeout probe)
+    ├── actions.npy                          #     recorded action chunks, shape (n_calls, 1, 16, 7) float32   [DoF column order: gripper, pitch, roll, x, y, yaw, z]
+    ├── metadata.json, rollout_summary.json  #     scenario config + results
+    └── stdout.log, stderr.log               #     the worker subprocess logs
+
+outputs/_setup_logs/                         # deployment record: uv install / LIBERO setup / server / smoke-run logs (no secrets)
+
+LICENSE                                       # Apache-2.0 (same license as upstream Isaac-GR00T)
+NOTICE                                        # attribution
+```
+
+## The run, in one line
+
+10 LIBERO simulation rollouts of `nvidia/GR00T-N1.7-LIBERO` (`libero_10`), `max_episode_steps = 50` (16 for the short-timeout probe), `n_action_steps = 8`, `n_envs = 1`, **no physical hardware**. 8 `normal` long-horizon tasks + 2 `abnormal_probe` (mild Gaussian observation noise; deliberately shortened timeout). All 10 ran cleanly with actions + video saved; `success = false` everywhere — a 50-step rollout cannot complete these long-horizon tasks, and that's the point: this verifies the deploy/serve/client/sim/render/save plumbing, not task performance.
+
+## Reproduce
+
+Full instructions: [`examples/LIBERO/smoke_tests/README.md`](examples/LIBERO/smoke_tests/README.md). Summary:
+
+```bash
+# 0. start from a clone of NVIDIA Isaac-GR00T and copy this repo's files into it
+git clone https://github.com/NVIDIA/Isaac-GR00T.git && cd Isaac-GR00T
+git submodule update --init external_dependencies/LIBERO
+# ... then place examples/LIBERO/smoke_tests/ and tests/test_libero_smoke_tests.py from THIS repo here ...
+
+# install gr00t (uv sync, or the minimal recipe in examples/LIBERO/smoke_tests/README.md), then:
+sudo apt install libegl1-mesa-dev libglu1-mesa cmake && bash gr00t/eval/sim/LIBERO/setup_libero.sh
+uv run hf download nvidia/GR00T-N1.7-LIBERO --include "libero_10/*" --local-dir checkpoints/GR00T-N1.7-LIBERO
+# request access to https://huggingface.co/nvidia/Cosmos-Reason2-2B  and  export HF_TOKEN=hf_...
+
+# 1. server (terminal 1) — this pulls the gated nvidia/Cosmos-Reason2-2B backbone
+uv run python gr00t/eval/run_gr00t_server.py \
+  --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 \
+  --embodiment-tag LIBERO_PANDA --use-sim-policy-wrapper
+
+# 2. smoke tests (terminal 2)
+uv run python examples/LIBERO/smoke_tests/run_10_smoke_tests.py \
+  --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 --host 127.0.0.1 --port 5555 \
+  --manifest examples/LIBERO/smoke_tests/scenarios_10.yaml --output-dir outputs/libero_smoke_tests \
+  --max-episode-steps 50 --save-video --render
+
+# 3. review artifacts (no GPU, no rerun)
+python examples/LIBERO/smoke_tests/visualize_smoke_run.py  --run-dir outputs/libero_smoke_tests/<ts> --open
+python examples/LIBERO/smoke_tests/make_video_montage.py   --run-dir outputs/libero_smoke_tests/<ts> --mode sequential --open
+```
+
+## Next: activation capture / SAE
+
+See [`examples/LIBERO/smoke_tests/ACTIVATION_HOOK_NOTES.md`](examples/LIBERO/smoke_tests/ACTIVATION_HOOK_NOTES.md). Suggested first step: register a `forward_hook` on `policy.model.backbone` (the `Qwen3Backbone` — its `forward` returns `backbone_features`, the last Qwen3-VL hidden state = the fused image+instruction representation that conditions action generation), dump that per `get_action` call alongside the already-saved `actions.npy` + frames, and train an SAE on it.
+
+## License & attribution
+
+This repo is released under **Apache-2.0**, matching upstream **NVIDIA Isaac-GR00T**
+(<https://github.com/NVIDIA/Isaac-GR00T>, © NVIDIA CORPORATION & AFFILIATES) — see `LICENSE` and `NOTICE`.
+The added tooling (`examples/LIBERO/smoke_tests/`, `tests/test_libero_smoke_tests.py`) is Apache-2.0 and
+imports/wraps the upstream `gr00t` package. The artifacts under `outputs/` are outputs of the
+`nvidia/GR00T-N1.7-LIBERO` model run in LIBERO simulation; **no model weights or upstream source files
+(other than this `LICENSE`) are redistributed here.**

examples/LIBERO/smoke_tests/ACTIVATION_HOOK_NOTES.md

@@ -0,0 +1,100 @@
+# Notes for future SAE activation capture (GR00T-N1.7-LIBERO)
+
+> Status: **not implemented yet.** This file just records the hook points found
+> while wiring up the LIBERO smoke tests, so a later change can add SAE
+> activation collection with minimal surface area. Nothing here changes model
+> behaviour.
+
+## Where actions are produced (the model call chain)
+
+1. `gr00t/eval/rollout_policy.py :: run_rollout_gymnasium_policy`
+   calls `policy.get_action(observations)` once per action chunk.
+   - For the smoke tests the `policy` is a `PolicyClient`
+     (`gr00t/policy/server_client.py`) talking over ZMQ to a server, so the
+     *actual* model lives in the server process. The
+     `examples/LIBERO/smoke_tests/_libero_rollout_worker.py` already wraps
+     `PolicyClient.get_action` to record the returned action chunks
+     (`actions.npy`); that wrapper is the cheapest place to record
+     *input/output* of the policy from the client side, but it cannot see
+     internal activations.
+
+2. Server side: `gr00t/eval/run_gr00t_server.py` builds a
+   `Gr00tPolicy` (`gr00t/policy/gr00t_policy.py`), optionally wrapped in
+   `Gr00tSimPolicyWrapper`. The endpoint `get_action` →
+   `Gr00tPolicy._get_action()` (≈ line 371):
+   - builds `collated_inputs` from the observation (images + proprio state +
+     language instruction), casts to bf16, then:
+   - `model_pred = self.model.get_action(**collated_inputs)` ← **the forward pass**
+   - `normalized_action = model_pred["action_pred"]`, then
+     `self.processor.decode_action(...)` → physical-unit action dict returned.
+
+3. The model: `gr00t/model/gr00t_n1d7/gr00t_n1d7.py :: GR00T_N1_7` (registered
+   `Gr00tN1d7`). `GR00T_N1_7.get_action(inputs)` (≈ line 589):
+   ```
+   backbone_inputs, action_inputs = self.prepare_input(inputs)
+   backbone_outputs = self.backbone(backbone_inputs)          # VLM
+   action_outputs   = self.action_head.get_action(backbone_outputs, action_inputs, options)
+   return action_outputs                                       # {"action_pred": ...}
+   ```
+
+## Likely hook points / module names
+
+Top-level model attributes (`policy.model` on the server, an
+`nn.Module` of class `Gr00tN1d7` / `GR00T_N1_7`):
+
+| Attribute | Class / file | What it computes |
+|---|---|---|
+| `model.backbone` | `Qwen3Backbone` (`gr00t/model/modules/qwen3_backbone.py`) | VLM over images + text instruction. Wraps HF `Qwen3VLForConditionalGeneration` at `model.backbone.model`; transformer decoder layers at `model.backbone.model.language_model.layers[...]` (truncated to `select_layer`); vision tower at `model.backbone.model.visual` (name may vary with the HF Qwen3-VL version). `forward()` runs with `output_hidden_states=True`, takes `hidden_states[-1]`, and returns `BatchFeature({"backbone_features": <Tensor [B, seq, hidden]>, "backbone_attention_mask": ...})`. |
+| `model.action_head` | `Gr00tN1d7ActionHead` (`gr00t/model/gr00t_n1d7/gr00t_n1d7.py`, ≈ line 38) | Flow-matching / diffusion action decoder. Inner net `model.action_head.model` is a `DiT` or `AlternateVLDiT` (`gr00t/model/modules/dit.py`, `gr00t/model/modules/flowmatching_modules.py`). Has `get_action(backbone_output, action_input, options)` and `get_action_with_features(...)` (≈ line 312) which is the natural place to also surface intermediate features. |
+| `model.collator` | `Gr00tN1d7DataCollator` (`gr00t/model/gr00t_n1d7/processing_gr00t_n1d7.py`) | builds VLM batch inputs from `vlm_content`. |
+
+**Recommended primary capture site:** the output of `model.backbone` — i.e.
+the tensor stored under key `backbone_features` returned by
+`Qwen3Backbone.forward` (last LLM hidden state, shape `[B, seq_len, hidden]`).
+This is the single fused representation of *image + instruction + (implicit)
+proprioception context* that conditions action generation, so it's the most
+informative single activation to feed an SAE. Secondary sites: per-layer LLM
+hidden states (`outputs.hidden_states[k]` inside `Qwen3Backbone.forward`),
+and the DiT block activations inside `model.action_head.model`.
+
+**Cheapest mechanism:** register `torch.nn.Module.register_forward_hook` on
+`policy.model.backbone` (and/or specific `...language_model.layers[k]`) right
+after the policy is constructed in `run_gr00t_server.py`, before
+`server.run()`. Hooks can append detached CPU/float16 tensors to a buffer that
+is flushed to disk per get_action call. No edits to model code required.
+Alternative: subclass `Gr00tPolicy` and override `_get_action` to also stash
+`self.model`'s intermediates.
+
+## Where observation / instruction / proprioception enter the model
+
+- **LIBERO env → observation dict**: `gr00t/eval/sim/LIBERO/libero_env.py ::
+  LiberoEnv._process_observation` produces:
+  - `video.image` (agentview 256×256×3), `video.wrist_image` (eye-in-hand),
+  - `state.x/y/z/roll/pitch/yaw` (EEF pose from `robot0_eef_pos` + axis-angle
+    of `robot0_eef_quat`), `state.gripper` (2-dim `robot0_gripper_qpos`),
+  - `annotation.human.action.task_description` (the **language instruction**,
+    a fixed string per task).
+- These are batched/temporally-stacked by `MultiStepWrapper`
+  (`gr00t/eval/sim/wrapper/multistep_wrapper.py`) and then, in
+  `Gr00tPolicy._get_action`, turned into `VLAStepData` via
+  `self._to_vla_step_data(obs)` and run through `self.processor(messages)`
+  (`Gr00tN1d7Processor`, `gr00t/model/gr00t_n1d7/processing_gr00t_n1d7.py`):
+  - images + instruction → `vlm_content` → tokenized/pixel-processed by the
+    Qwen3-VL processor → `model.backbone`.
+  - proprio `state.*` → `action_inputs` → `model.action_head` (conditioning).
+- The smoke-test worker can optionally perturb the image observations
+  *before* `policy.get_action` (`--obs-noise-std`); a future activation-capture
+  run could pair clean vs. perturbed activations from the same seeds.
+
+## Suggested next step for activation capture
+
+1. Add an opt-in flag to `run_gr00t_server.py` (e.g. `--capture-activations
+   DIR` / `--capture-layers backbone,llm.20`) that, after building the policy,
+   registers forward hooks on the chosen modules and writes one
+   `.npz`/`.safetensors` per `get_action` call (keyed by an episode/step id
+   passed through `options`).
+2. Have `_libero_rollout_worker.py` thread a stable `(scenario_id, episode,
+   step)` tag into `policy.get_action(obs, options=...)` so captured
+   activations can be joined back to `actions.npy` and the saved video frames.
+3. Train the SAE on `backbone_features` first (one SAE), then expand to
+   per-layer LLM hidden states if needed.

examples/LIBERO/smoke_tests/README.md

@@ -0,0 +1,160 @@
+# LIBERO simulation-only smoke tests for `nvidia/GR00T-N1.7-LIBERO`
+
+These scripts deploy the `libero_10` checkpoint of
+[`nvidia/GR00T-N1.7-LIBERO`](https://huggingface.co/nvidia/GR00T-N1.7-LIBERO)
+and run **10 short, simulation-only LIBERO rollouts** to verify the deploy /
+eval plumbing:
+
+1. the model checkpoint is present and loads,
+2. the GR00T inference server starts,
+3. the LIBERO rollout client connects to it,
+4. short rollouts run with **no physical robot hardware** (LIBERO sim only),
+5. a video / rendered frames are saved per scenario for human review,
+6. basic rollout metadata + recorded action chunks are saved,
+7. activation hooks can be added later — see [`ACTIVATION_HOOK_NOTES.md`](ACTIVATION_HOOK_NOTES.md).
+
+This is **not** a benchmark. `max_episode_steps` is deliberately tiny, so the
+reported success rates are meaningless — only the plumbing matters here. The
+scripts do **not** rewrite the official evaluation path: each scenario runs
+through `gr00t.eval.rollout_policy.run_rollout_gymnasium_policy` via the thin
+`_libero_rollout_worker.py`, which only adds action recording and two optional,
+simulation-only perturbations (Gaussian pixel noise; action-chunk repeat).
+
+## Files
+
+| File | Purpose |
+|---|---|
+| `scenarios_10.yaml` | the 10-scenario manifest (8 `normal`, 2 `abnormal_probe`) |
+| `run_10_smoke_tests.py` | runner: validates paths/server, runs all scenarios, writes summaries |
+| `_libero_rollout_worker.py` | runs **one** rollout against the server (called by the runner; runs in the LIBERO venv) |
+| `review_smoke_tests.py` | lists scenarios + video paths from a run dir, optionally plays videos |
+| `ACTIVATION_HOOK_NOTES.md` | where to insert future SAE activation capture |
+
+Tests live at `tests/test_libero_smoke_tests.py`.
+
+## One-time setup
+
+```bash
+# 1. Install deps (project venv) — uses uv
+uv sync          # or: uv pip install -e .   (see note below)
+
+# 2. LIBERO simulation env (separate venv; only needed once)
+sudo apt update && sudo apt install libegl1-mesa-dev libglu1-mesa
+bash gr00t/eval/sim/LIBERO/setup_libero.sh
+
+# 3. Download the libero_10 checkpoint
+#    (HuggingFace does not support nested repo paths directly)
+uv run hf download nvidia/GR00T-N1.7-LIBERO \
+  --include "libero_10/config.json" \
+            "libero_10/embodiment_id.json" \
+            "libero_10/model-*.safetensors" \
+            "libero_10/model.safetensors.index.json" \
+            "libero_10/processor_config.json" \
+            "libero_10/statistics.json" \
+  --local-dir checkpoints/GR00T-N1.7-LIBERO
+# If the glob skips config.json, fetch it explicitly:
+uv run hf download nvidia/GR00T-N1.7-LIBERO libero_10/config.json \
+  --local-dir checkpoints/GR00T-N1.7-LIBERO
+```
+
+> **Note on `uv sync`:** the project `pyproject.toml` pins GPU-deployment-only
+> packages (`flash-attn`, `deepspeed`, `tensorrt-*`) and the lockfile references
+> aarch64 wheels that are stored via git-LFS. On a plain x86 box without git-LFS
+> / CUDA toolkit, `uv sync` can fail on those. They are **not needed for the
+> smoke test** (the model falls back to `sdpa` attention if `flash_attn` is
+> missing). A working minimal install is:
+> ```bash
+> uv venv .venv --python 3.10
+> uv pip install --python .venv/bin/python -e . --no-deps
+> uv pip install --python .venv/bin/python \
+>   torch==2.7.1 torchvision==0.22.1 transformers==4.57.3 numpy==1.26.4 \
+>   albumentations==1.4.18 av==16.1.0 diffusers==0.35.1 dm-tree lmdb==1.7.5 \
+>   msgpack==1.1.0 msgpack-numpy==0.4.8 pandas==2.2.3 peft==0.17.1 termcolor==3.2.0 \
+>   tyro==0.9.17 click==8.1.8 datasets==3.6.0 cryptography einops==0.8.1 \
+>   gitpython==3.1.46 jsonlines==4.0.0 gymnasium==1.2.2 matplotlib==3.10.1 \
+>   omegaconf==2.3.0 scipy==1.15.3 torchcodec==0.4.0 wandb==0.23.0 pyzmq==27.0.1 \
+>   "huggingface-hub[cli]" "opencv-python-headless>=4.5,<4.13" safetensors accelerate \
+>   sentencepiece protobuf pyyaml tqdm
+> ```
+
+## Running the smoke tests
+
+**Terminal 1 — start the GR00T inference server:**
+
+```bash
+uv run python gr00t/eval/run_gr00t_server.py \
+    --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 \
+    --embodiment-tag LIBERO_PANDA \
+    --use-sim-policy-wrapper
+```
+
+(Server defaults: `--host 0.0.0.0 --port 5555`. This is exactly the command
+documented in `examples/LIBERO/README.md`.)
+
+> **Gated backbone:** GR00T-N1.7 uses `nvidia/Cosmos-Reason2-2B` (a Qwen3-VL
+> model) as its VLM backbone, and that repo is **gated** on HuggingFace.
+> Starting the server pulls the base repo's config/processor, so it fails
+> without HF auth. One-time fix:
+> ```bash
+> # 1. request access (one click): https://huggingface.co/nvidia/Cosmos-Reason2-2B
+> # 2. authenticate:
+> export HF_TOKEN=hf_xxx          # or: uv run hf auth login
+> ```
+> The `run_10_smoke_tests.py` runner detects this failure mode (when launched
+> with `--start-server`) and prints the same hint.
+
+**Terminal 2 — run the 10 smoke tests:**
+
+```bash
+uv run python examples/LIBERO/smoke_tests/run_10_smoke_tests.py \
+    --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 \
+    --host 127.0.0.1 --port 5555 \
+    --manifest examples/LIBERO/smoke_tests/scenarios_10.yaml \
+    --output-dir outputs/libero_smoke_tests \
+    --max-episode-steps 50 \
+    --save-video --render
+```
+
+Other flags:
+
+- `--start-server` — launch (and later kill) the server from the runner itself,
+  so you only need one terminal.
+- `--dry-run` — validate the manifest, model path and (optionally) the server,
+  write `metadata.json` per scenario, but run no simulations.
+- `--resume` + `--run-dir <output-dir>/<timestamp>` — re-enter an existing run
+  dir and skip scenarios that already produced a `rollout_summary.json`.
+- `--libero-python <path>` — override the LIBERO venv python (default:
+  `gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/python`, falls back to the current
+  interpreter with a warning).
+
+If the model path is missing, or the server is unreachable, or the LIBERO sim
+env is not importable, the runner stops with an actionable error — it does
+**not** fake a successful rollout.
+
+## Outputs
+
+```
+outputs/libero_smoke_tests/<timestamp>/
+  summary.json
+  summary.md
+  server.log                       # only if --start-server
+  <scenario_id>/
+    metadata.json                  # the manifest entry + resolved params
+    rollout_summary.json           # env_name, seed, steps, success, reward, ...
+    actions.npy                    # recorded action chunks (if the rollout ran)
+    video.mp4                      # rendered rollout (if recording succeeded)
+    frames/                        # decoded PNG frames (only with --render)
+    stdout.log / stderr.log        # the worker subprocess logs
+```
+
+At the end the runner prints a table:
+
+```
+scenario_id | label | seed | rollout_started | actions_produced | video_saved | success | output_dir | error_if_any
+```
+
+and writes the same to `summary.json` / `summary.md`. Review videos with:
+
+```bash
+python examples/LIBERO/smoke_tests/review_smoke_tests.py --run-dir outputs/libero_smoke_tests/<timestamp> --open
+```

examples/LIBERO/smoke_tests/_libero_rollout_worker.py

@@ -0,0 +1,311 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""Single-episode LIBERO rollout worker for the smoke-test harness.
+
+This is a *thin* wrapper around the official evaluation path
+(``gr00t.eval.rollout_policy.run_rollout_gymnasium_policy``).  It exists only
+so that, for one short rollout, we can additionally:
+
+  * connect to an already-running GR00T policy server (``PolicyClient``),
+  * record the action chunks the policy returns (-> ``actions.npy``),
+  * optionally apply a tiny, simulation-only observation perturbation
+    (Gaussian pixel noise) or repeat the previous action chunk -- these are
+    placeholders for future "abnormal" probes, not real anomaly logic,
+  * collect the recorded video into the per-scenario output directory.
+
+It does **not** modify the official rollout/eval code.  It is meant to be run
+*inside the LIBERO uv venv* created by
+``gr00t/eval/sim/LIBERO/setup_libero.sh``, e.g.::
+
+    gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/python \
+        examples/LIBERO/smoke_tests/_libero_rollout_worker.py --help
+
+Normally you do not call this directly -- ``run_10_smoke_tests.py`` invokes it.
+"""
+
+from __future__ import annotations
+
+import argparse
+import glob
+import json
+import os
+from pathlib import Path
+import shutil
+import sys
+import time
+import traceback
+
+# LIBERO renders headlessly via EGL; set this before importing the env modules.
+os.environ.setdefault("MUJOCO_GL", "egl")
+os.environ.setdefault("PYOPENGL_PLATFORM", "egl")
+
+import numpy as np
+
+RESULT_PREFIX = "SMOKE_RESULT_JSON:"
+
+
+def _emit_result(result: dict) -> None:
+    """Print the machine-readable result line the parent runner parses."""
+    sys.stdout.flush()
+    print(RESULT_PREFIX + " " + json.dumps(result), flush=True)
+
+
+def _action_to_numpy(action):
+    """Best-effort conversion of a policy action (dict of arrays or array) to ndarray."""
+    if isinstance(action, dict):
+        # LIBERO action space: action.x, action.y, ..., action.gripper
+        keys = sorted(action.keys())
+        try:
+            parts = [np.asarray(action[k]) for k in keys]
+            return np.concatenate(parts, axis=-1)
+        except Exception:
+            return {k: np.asarray(v) for k, v in action.items()}
+    return np.asarray(action)
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description="Single LIBERO rollout against a running GR00T server.")
+    ap.add_argument("--env-name", required=True)
+    ap.add_argument("--host", default="127.0.0.1")
+    ap.add_argument("--port", type=int, default=5555)
+    ap.add_argument("--max-episode-steps", type=int, default=50)
+    ap.add_argument("--n-action-steps", type=int, default=8)
+    ap.add_argument("--seed", type=int, default=0)
+    ap.add_argument("--out-dir", required=True)
+    ap.add_argument("--save-video", dest="save_video", action="store_true", default=True)
+    ap.add_argument("--no-save-video", dest="save_video", action="store_false")
+    ap.add_argument("--save-frames", action="store_true", default=False,
+                    help="Also dump decoded video frames into <out-dir>/frames/.")
+    ap.add_argument("--obs-noise-std", type=float, default=0.0,
+                    help="Std-dev of zero-mean Gaussian noise added to image observations "
+                         "before policy inference (0 = disabled). Simulation-only probe.")
+    ap.add_argument("--action-repeat", type=int, default=1,
+                    help="If >1, re-use the previous action chunk instead of querying the "
+                         "policy on (k-1)/k of the steps (1 = disabled). Simulation-only probe.")
+    ap.add_argument("--timeout-ms", type=int, default=120000,
+                    help="ZMQ send/recv timeout for the policy client.")
+    args = ap.parse_args()
+
+    out_dir = Path(args.out_dir)
+    out_dir.mkdir(parents=True, exist_ok=True)
+    video_raw_dir = out_dir / "_video_raw"
+
+    result: dict = {
+        "ok": False,
+        "rollout_started": False,
+        "actions_produced": False,
+        "video_saved": False,
+        "success": None,
+        "error": None,
+        "env_name": args.env_name,
+        "seed": args.seed,
+        "n_action_calls": 0,
+        "episode_length": None,
+        "episode_reward": None,
+        "video_path": None,
+        "actions_path": None,
+    }
+
+    try:
+        from gr00t.eval.rollout_policy import (
+            MultiStepConfig,
+            VideoConfig,
+            WrapperConfigs,
+            run_rollout_gymnasium_policy,
+        )
+        from gr00t.eval.sim.env_utils import get_embodiment_tag_from_env_name  # noqa: F401
+        from gr00t.policy.server_client import PolicyClient
+
+        try:
+            from gr00t.utils.determinism import seed_everything
+
+            seed_everything(args.seed)
+        except Exception:
+            pass
+
+        # --- connect to the running server -----------------------------------
+        # Quick reachability probe with a SHORT timeout (so we fail fast instead
+        # of blocking for `timeout_ms` when nothing is listening).
+        import socket as _socket
+
+        reachable = False
+        try:
+            with _socket.create_connection((args.host, args.port), timeout=4.0):
+                pass
+            probe = PolicyClient(host=args.host, port=args.port, timeout_ms=4000)
+            reachable = bool(probe.ping())
+            del probe
+        except OSError:
+            reachable = False
+        if not reachable:
+            result["error"] = (
+                f"GR00T policy server is not reachable at {args.host}:{args.port}. "
+                "Start it first (see run_10_smoke_tests.py output / smoke_tests/README.md). "
+                "Note: GR00T-N1.7's backbone 'nvidia/Cosmos-Reason2-2B' is a gated HF repo, so "
+                "the server needs HF auth (request access + export HF_TOKEN)."
+            )
+            _emit_result(result)
+            return 3
+
+        policy = PolicyClient(host=args.host, port=args.port, timeout_ms=args.timeout_ms)
+
+        # --- instrument get_action: record actions + optional perturbations ---
+        recorded_actions: list = []
+        rng = np.random.default_rng(args.seed)
+        n_calls = {"n": 0}
+        last_av = {"v": None}
+        orig_get_action = policy.get_action
+
+        def patched_get_action(observation, *a, **k):
+            if args.obs_noise_std and args.obs_noise_std > 0:
+                for key, val in list(observation.items()):
+                    if isinstance(val, np.ndarray) and val.ndim >= 3 and "image" in key.lower():
+                        noisy = val.astype(np.float32) + rng.normal(
+                            0.0, args.obs_noise_std, size=val.shape
+                        )
+                        observation[key] = np.clip(noisy, 0, 255).astype(val.dtype)
+            if (
+                args.action_repeat
+                and args.action_repeat > 1
+                and last_av["v"] is not None
+                and (n_calls["n"] % args.action_repeat) != 0
+            ):
+                action, info = last_av["v"]
+            else:
+                action, info = orig_get_action(observation, *a, **k)
+                last_av["v"] = (action, info)
+            n_calls["n"] += 1
+            recorded_actions.append(_action_to_numpy(action))
+            return action, info
+
+        policy.get_action = patched_get_action  # type: ignore[assignment]
+
+        # --- run one short rollout via the official path ---------------------
+        wrapper_configs = WrapperConfigs(
+            video=VideoConfig(
+                video_dir=str(video_raw_dir) if args.save_video else None,
+                max_episode_steps=args.max_episode_steps,
+            ),
+            multistep=MultiStepConfig(
+                n_action_steps=args.n_action_steps,
+                max_episode_steps=args.max_episode_steps,
+                terminate_on_success=True,
+            ),
+        )
+
+        result["rollout_started"] = True
+        t0 = time.time()
+        env_name, episode_successes, episode_infos = run_rollout_gymnasium_policy(
+            env_name=args.env_name,
+            policy=policy,
+            wrapper_configs=wrapper_configs,
+            n_episodes=1,
+            n_envs=1,
+            seed=args.seed,
+        )
+        elapsed = time.time() - t0
+
+        # --- collect results -------------------------------------------------
+        success = bool(np.any(episode_successes)) if len(episode_successes) else None
+        ep_lengths = list(episode_infos.get("episode_lengths", []) or [])
+        ep_rewards = list(episode_infos.get("episode_rewards", []) or [])
+        episode_length = int(ep_lengths[0]) if ep_lengths else None
+        episode_reward = float(ep_rewards[0]) if ep_rewards else None
+
+        result["n_action_calls"] = n_calls["n"]
+        result["success"] = success
+        result["episode_length"] = episode_length
+        result["episode_reward"] = episode_reward
+
+        # --- save actions ----------------------------------------------------
+        actions_path = None
+        if recorded_actions:
+            actions_path = out_dir / "actions.npy"
+            try:
+                stacked = np.stack(recorded_actions, axis=0)
+                np.save(actions_path, stacked)
+            except Exception:
+                np.save(actions_path, np.array(recorded_actions, dtype=object), allow_pickle=True)
+            result["actions_produced"] = True
+            result["actions_path"] = str(actions_path)
+
+        # --- collect video / frames -----------------------------------------
+        video_path = None
+        if args.save_video and video_raw_dir.exists():
+            mp4s = sorted(
+                glob.glob(str(video_raw_dir / "**" / "*.mp4"), recursive=True),
+                key=lambda p: os.path.getsize(p),
+                reverse=True,
+            )
+            if mp4s:
+                video_path = out_dir / "video.mp4"
+                shutil.copyfile(mp4s[0], video_path)
+                result["video_saved"] = True
+                result["video_path"] = str(video_path)
+                # keep a record of the original (success-suffixed) filename
+                result["video_source_name"] = os.path.basename(mp4s[0])
+                if args.save_frames:
+                    try:
+                        import cv2
+
+                        frames_dir = out_dir / "frames"
+                        frames_dir.mkdir(exist_ok=True)
+                        cap = cv2.VideoCapture(str(video_path))
+                        idx = 0
+                        while True:
+                            ok, frame = cap.read()
+                            if not ok:
+                                break
+                            cv2.imwrite(str(frames_dir / f"frame_{idx:05d}.png"), frame)
+                            idx += 1
+                        cap.release()
+                        result["frames_dir"] = str(frames_dir)
+                        result["n_frames"] = idx
+                    except Exception as e:  # noqa: BLE001
+                        result["frames_error"] = repr(e)
+
+        # --- write per-scenario rollout summary ------------------------------
+        rollout_summary = {
+            "env_name": env_name,
+            "seed": args.seed,
+            "requested_max_episode_steps": args.max_episode_steps,
+            "n_action_steps": args.n_action_steps,
+            "n_get_action_calls": n_calls["n"],
+            "episode_successes": [bool(x) for x in episode_successes],
+            "success": success,
+            "episode_length": episode_length,
+            "episode_reward": episode_reward,
+            "elapsed_sec": elapsed,
+            "obs_noise_std": args.obs_noise_std,
+            "action_repeat": args.action_repeat,
+            "video_path": str(video_path) if video_path else None,
+            "actions_path": str(actions_path) if actions_path else None,
+            "policy_server": {"host": args.host, "port": args.port},
+        }
+        with open(out_dir / "rollout_summary.json", "w") as f:
+            json.dump(rollout_summary, f, indent=2)
+
+        result["ok"] = True
+        _emit_result(result)
+        return 0
+
+    except Exception:  # noqa: BLE001
+        tb = traceback.format_exc()
+        result["error"] = tb.strip().splitlines()[-1] if tb.strip() else "unknown error"
+        result["traceback"] = tb
+        # also write a partial rollout_summary so the runner has something
+        try:
+            with open(out_dir / "rollout_summary.json", "w") as f:
+                json.dump({"error": result["error"], "traceback": tb,
+                           "env_name": args.env_name, "seed": args.seed}, f, indent=2)
+        except Exception:
+            pass
+        print(tb, file=sys.stderr, flush=True)
+        _emit_result(result)
+        return 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

examples/LIBERO/smoke_tests/make_video_montage.py

@@ -0,0 +1,578 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""Build a compact human-review video montage for a completed LIBERO smoke run.
+
+Reads every scenario directory under ``--run-dir``, takes a short clip from each
+``video.mp4`` (default 5 s; full video if shorter), burns in an overlay
+(scenario_id / label / seed / instruction / success), and writes:
+
+    <run-dir>/review_clips/<scenario_id>.mp4   one short, captioned clip per scenario
+    <run-dir>/review_montage.mp4               all scenarios back-to-back (sequential) or in a grid
+    <run-dir>/review_playlist.html             the montage + each clip with metadata and a link to the full video
+
+It does NOT rerun LIBERO and does NOT modify any existing rollout artifact —
+it only reads ``video.mp4`` / ``metadata.json`` / ``rollout_summary.json`` and
+writes the three review artifacts above. Requires ``ffmpeg`` (and ``ffprobe``)
+on PATH; no GPU.
+
+CLI::
+
+    python examples/LIBERO/smoke_tests/make_video_montage.py \
+        --run-dir outputs/libero_smoke_tests/20260512_122756 \
+        --clip-seconds 5 --mode sequential --open
+
+Flags: --run-dir  --clip-seconds  --mode {sequential,grid}  --hold-seconds
+       --instruction-chars  --cols  --open  --dry-run
+"""
+
+from __future__ import annotations
+
+import argparse
+import datetime as _dt
+import html
+import json
+import math
+import os
+from pathlib import Path
+import shutil
+import subprocess
+import sys
+import tempfile
+
+FONT_CANDIDATES = [
+    "/usr/share/fonts/truetype/dejavu/DejaVuSans.ttf",
+    "/usr/share/fonts/truetype/liberation/LiberationSans-Regular.ttf",
+    "/usr/share/fonts/dejavu/DejaVuSans.ttf",
+    "/Library/Fonts/Arial.ttf",
+    "/System/Library/Fonts/Supplemental/Arial.ttf",
+]
+
+
+# --------------------------------------------------------------------------- #
+# small helpers
+# --------------------------------------------------------------------------- #
+def _die(msg: str, code: int = 2) -> None:
+    print(msg, file=sys.stderr)
+    raise SystemExit(code)
+
+
+def _ensure_ffmpeg() -> tuple[str, str]:
+    ff = shutil.which("ffmpeg")
+    fp = shutil.which("ffprobe")
+    if ff and fp:
+        return ff, fp
+    missing = " and ".join(x for x, ok in [("ffmpeg", ff), ("ffprobe", fp)] if not ok)
+    _die(
+        f"ERROR: required tool(s) not found on PATH: {missing}.\n"
+        "Install ffmpeg, e.g.:\n"
+        "  sudo apt-get update && sudo apt-get install -y ffmpeg      # Debian/Ubuntu\n"
+        "  conda install -c conda-forge ffmpeg                        # conda\n"
+        "  brew install ffmpeg                                        # macOS\n"
+        "  uv pip install imageio-ffmpeg && export PATH=... (its bundled ffmpeg)  # last resort\n"
+        "Then re-run this script."
+    )
+    raise AssertionError  # unreachable
+
+
+def _find_font() -> str | None:
+    for f in FONT_CANDIDATES:
+        if os.path.exists(f):
+            return f
+    return None
+
+
+def _load_json(path: Path):
+    try:
+        with open(path) as f:
+            return json.load(f)
+    except Exception:
+        return None
+
+
+def _probe_video(ffprobe: str, path: Path) -> dict | None:
+    try:
+        out = subprocess.run(
+            [ffprobe, "-v", "error", "-select_streams", "v:0",
+             "-show_entries", "stream=width,height,r_frame_rate,duration,nb_frames",
+             "-show_entries", "format=duration", "-of", "json", str(path)],
+            capture_output=True, text=True, check=True,
+        ).stdout
+        info = json.loads(out)
+        st = (info.get("streams") or [{}])[0]
+        w = int(st.get("width") or 0)
+        h = int(st.get("height") or 0)
+        rfr = st.get("r_frame_rate") or "0/1"
+        try:
+            num, den = rfr.split("/")
+            fps = float(num) / float(den) if float(den) else 0.0
+        except Exception:
+            fps = 0.0
+        dur = st.get("duration") or (info.get("format") or {}).get("duration")
+        try:
+            dur = float(dur)
+        except (TypeError, ValueError):
+            dur = 0.0
+        nb = st.get("nb_frames")
+        try:
+            nb = int(nb)
+        except (TypeError, ValueError):
+            nb = None
+        if (not dur or dur <= 0) and nb and fps:
+            dur = nb / fps
+        return {"width": w, "height": h, "fps": fps or 20.0, "duration": dur, "nb_frames": nb}
+    except Exception:
+        return None
+
+
+def _truncate(s: str, n: int) -> str:
+    s = " ".join(str(s).split())
+    return s if len(s) <= n else s[: max(0, n - 1)].rstrip() + "…"
+
+
+def _drawtext_chain(font: str | None, lines: list[str], tmp_dir: Path,
+                    box_h: int, fontsize: int = 15) -> str:
+    """Build a filter chain: one translucent box + one drawtext per line (read from textfiles)."""
+    parts = [f"drawbox=x=0:y=0:w=iw:h={box_h}:color=black@0.55:t=fill"]
+    y = 5
+    for i, line in enumerate(lines):
+        tf = tmp_dir / f"line_{i}.txt"
+        tf.write_text(line if line else " ", encoding="utf-8")
+        # textfile path uses only safe chars (our tmp dir); expansion=none -> fully literal text.
+        fontspec = f"fontfile='{tf.parent / 'FONT'}'"  # placeholder, replaced below
+        if font:
+            fontspec = f"fontfile={font}"
+        else:
+            fontspec = "font=sans"
+        parts.append(
+            f"drawtext={fontspec}:textfile={tf}:expansion=none:reload=0:"
+            f"fontsize={fontsize}:fontcolor=white:shadowcolor=black@0.8:shadowx=1:shadowy=1:"
+            f"x=6:y={y}"
+        )
+        y += fontsize + 7
+    return ",".join(parts)
+
+
+def _run_ffmpeg(ffmpeg: str, args: list[str], desc: str) -> tuple[bool, str]:
+    try:
+        r = subprocess.run([ffmpeg, "-hide_banner", "-y", "-nostdin", *args],
+                           capture_output=True, text=True)
+        if r.returncode != 0:
+            tail = "\n".join((r.stderr or "").strip().splitlines()[-12:])
+            return False, f"{desc}: ffmpeg exited {r.returncode}\n{tail}"
+        return True, ""
+    except Exception as e:  # noqa: BLE001
+        return False, f"{desc}: {e!r}"
+
+
+# --------------------------------------------------------------------------- #
+# planning
+# --------------------------------------------------------------------------- #
+def discover_scenarios(run_dir: Path) -> tuple[list[str], dict]:
+    summary = _load_json(run_dir / "summary.json")
+    order: list[str] = []
+    rows_by_id: dict[str, dict] = {}
+    if summary and isinstance(summary.get("scenarios"), list):
+        for r in summary["scenarios"]:
+            sid = r.get("scenario_id")
+            if sid:
+                order.append(sid)
+                rows_by_id[sid] = r
+    skip = {"plots", "review_clips"}
+    for p in sorted(run_dir.iterdir()):
+        if p.is_dir() and p.name not in skip and p.name not in order:
+            order.append(p.name)
+    return order, {"summary": summary, "rows_by_id": rows_by_id}
+
+
+def build_plan(run_dir: Path, ffprobe: str, clip_seconds: float, instr_chars: int) -> list[dict]:
+    order, ctx = discover_scenarios(run_dir)
+    rows_by_id = ctx["rows_by_id"]
+    plan: list[dict] = []
+    for sid in order:
+        sc_dir = run_dir / sid
+        meta = _load_json(sc_dir / "metadata.json") or {}
+        roll = _load_json(sc_dir / "rollout_summary.json") or {}
+        srow = rows_by_id.get(sid, {})
+        manifest_entry = meta.get("manifest_entry", {}) or {}
+        resolved = meta.get("resolved", {}) or {}
+        label = manifest_entry.get("label") or meta.get("label") or srow.get("label") or "?"
+        seed = manifest_entry.get("seed", resolved.get("seed", srow.get("seed")))
+        instruction = manifest_entry.get("instruction") or roll.get("instruction") or "(unknown)"
+        env_name = resolved.get("env_name") or manifest_entry.get("env_name") or roll.get("env_name")
+        success = roll.get("success", srow.get("success"))
+        video = sc_dir / "video.mp4"
+        info = _probe_video(ffprobe, video) if video.exists() else None
+        src_dur = (info or {}).get("duration") or 0.0
+        clip_dur = src_dur if (src_dur and src_dur < clip_seconds) else clip_seconds
+        warn = None
+        if not video.exists():
+            warn = "video.mp4 not found"
+        elif info is None:
+            warn = "ffprobe could not read video.mp4"
+        elif src_dur <= 0:
+            warn = "video.mp4 has zero/unknown duration"
+        plan.append({
+            "scenario_id": sid, "label": label, "seed": seed,
+            "instruction": instruction, "instruction_short": _truncate(instruction, instr_chars),
+            "env_name": env_name, "success": success,
+            "video": video if video.exists() else None,
+            "video_rel": os.path.relpath(video, run_dir) if video.exists() else None,
+            "src_duration": round(float(src_dur), 3) if src_dur else 0.0,
+            "src_info": info,
+            "clip_duration": round(float(clip_dur), 3),
+            "clip_path": run_dir / "review_clips" / f"{sid}.mp4",
+            "clip_rel": os.path.join("review_clips", f"{sid}.mp4"),
+            "warning": warn,
+            "ok_source": video.exists() and info is not None and src_dur > 0,
+        })
+    return plan
+
+
+def _success_str(s) -> str:
+    if s is True:
+        return "true"
+    if s is False:
+        return "false"
+    return "unknown"
+
+
+# --------------------------------------------------------------------------- #
+# clip / montage generation
+# --------------------------------------------------------------------------- #
+def make_clip(ffmpeg: str, font: str | None, entry: dict, idx: int, n: int,
+              tmp_root: Path, target_w: int) -> tuple[bool, str]:
+    """Trim entry['video'] to clip_duration, burn overlay, write entry['clip_path']."""
+    src = entry["video"]
+    out = entry["clip_path"]
+    out.parent.mkdir(parents=True, exist_ok=True)
+    tdir = tmp_root / f"clip_{idx:02d}"
+    tdir.mkdir(parents=True, exist_ok=True)
+    lines = [
+        f"[{idx + 1}/{n}]  {entry['scenario_id']}",
+        f"label={entry['label']}    seed={entry['seed']}    success={_success_str(entry['success'])}",
+        f"task: {entry['instruction_short']}",
+    ]
+    box_h = 5 + 3 * (15 + 7) + 4
+    overlay = _drawtext_chain(font, lines, tdir, box_h=box_h, fontsize=15)
+    # scale to a consistent width (keep aspect, force even dims), then overlay text.
+    vf = (f"scale={target_w}:-2:flags=bicubic,setsar=1,{overlay},format=yuv420p")
+    args = [
+        "-t", f"{entry['clip_duration']:.3f}", "-i", str(src),
+        "-an", "-vf", vf, "-r", "20",
+        "-c:v", "libx264", "-pix_fmt", "yuv420p", "-preset", "veryfast", "-crf", "23",
+        "-movflags", "+faststart", str(out),
+    ]
+    return _run_ffmpeg(ffmpeg, args, f"clip[{entry['scenario_id']}]")
+
+
+def make_montage_sequential(ffmpeg: str, clips: list[Path], out_path: Path,
+                            hold_seconds: float, target_w: int) -> tuple[bool, str]:
+    if not clips:
+        return False, "no clips to montage"
+    inputs: list[str] = []
+    for c in clips:
+        inputs += ["-i", str(c)]
+    # pad the tail of each clip by hold_seconds (freeze last frame) so the overlay is readable,
+    # normalize fps/size/sar/pixfmt, then concat.
+    n = len(clips)
+    parts = []
+    for i in range(n):
+        parts.append(
+            f"[{i}:v]scale={target_w}:-2:flags=bicubic,setsar=1,fps=20,"
+            f"tpad=stop_mode=clone:stop_duration={hold_seconds:.3f},format=yuv420p[v{i}]"
+        )
+    concat_in = "".join(f"[v{i}]" for i in range(n))
+    parts.append(f"{concat_in}concat=n={n}:v=1:a=0[out]")
+    fc = ";".join(parts)
+    args = [*inputs, "-filter_complex", fc, "-map", "[out]", "-an",
+            "-c:v", "libx264", "-pix_fmt", "yuv420p", "-preset", "veryfast", "-crf", "23",
+            "-movflags", "+faststart", str(out_path)]
+    return _run_ffmpeg(ffmpeg, args, "montage(sequential)")
+
+
+def make_montage_grid(ffmpeg: str, clips: list[Path], out_path: Path, cols: int,
+                      cell_w: int, cell_h: int, ffprobe: str) -> tuple[bool, str]:
+    if not clips:
+        return False, "no clips to montage"
+    n = len(clips)
+    cols = max(1, min(cols, n))
+    rows = math.ceil(n / cols)
+    n_cells = cols * rows
+    # max duration across clips (to pad shorter ones)
+    durs = []
+    for c in clips:
+        info = _probe_video(ffprobe, c)
+        durs.append((info or {}).get("duration") or 0.0)
+    max_dur = max(durs) if durs else 0.0
+    if max_dur <= 0:
+        return False, "could not determine clip durations for grid"
+
+    inputs: list[str] = []
+    for c in clips:
+        inputs += ["-i", str(c)]
+    # extra black inputs to fill the grid
+    n_pad = n_cells - n
+    for _ in range(n_pad):
+        inputs += ["-f", "lavfi", "-t", f"{max_dur:.3f}", "-i",
+                   f"color=c=black:s={cell_w}x{cell_h}:r=20"]
+
+    parts = []
+    for i in range(n):
+        # scale to fit the cell, pad to exact cell size (centered), freeze-pad to max_dur
+        parts.append(
+            f"[{i}:v]scale={cell_w}:{cell_h}:force_original_aspect_ratio=decrease,"
+            f"pad={cell_w}:{cell_h}:(ow-iw)/2:(oh-ih)/2:color=black,setsar=1,fps=20,"
+            f"tpad=stop_mode=clone:stop_duration={max(0.0, max_dur - durs[i]):.3f},format=yuv420p[v{i}]"
+        )
+    for j in range(n_pad):
+        k = n + j
+        parts.append(f"[{k}:v]setsar=1,fps=20,format=yuv420p[v{k}]")
+    layout = "|".join(f"{(i % cols) * cell_w}_{(i // cols) * cell_h}" for i in range(n_cells))
+    stack_in = "".join(f"[v{i}]" for i in range(n_cells))
+    parts.append(f"{stack_in}xstack=inputs={n_cells}:layout={layout}[out]")
+    fc = ";".join(parts)
+    args = [*inputs, "-filter_complex", fc, "-map", "[out]", "-an",
+            "-c:v", "libx264", "-pix_fmt", "yuv420p", "-preset", "veryfast", "-crf", "23",
+            "-movflags", "+faststart", str(out_path)]
+    return _run_ffmpeg(ffmpeg, args, "montage(grid)")
+
+
+# --------------------------------------------------------------------------- #
+# HTML
+# --------------------------------------------------------------------------- #
+_CSS = """
+body{font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif;margin:0;padding:24px;background:#f4f5f7;color:#1a1a1a;}
+h1{margin:0 0 4px 0;} .sub{color:#666;font-size:14px;margin-bottom:18px;}
+.top{background:#fff;border:1px solid #e0e0e0;border-radius:10px;padding:16px;margin-bottom:22px;box-shadow:0 1px 3px rgba(0,0,0,.06);}
+.top video{width:100%;max-width:760px;border-radius:8px;background:#000;display:block;}
+.grid{display:grid;grid-template-columns:repeat(auto-fill,minmax(360px,1fr));gap:16px;}
+.card{background:#fff;border:1px solid #e0e0e0;border-radius:10px;padding:14px;box-shadow:0 1px 3px rgba(0,0,0,.05);}
+.card h3{margin:0 0 6px 0;font-size:15px;}
+.card video{width:100%;border-radius:6px;background:#000;}
+.badge{display:inline-block;padding:1px 8px;border-radius:10px;font-size:11px;font-weight:600;}
+.b-normal{background:#e3f0ff;color:#1a5fb4;} .b-abn{background:#ffe9d6;color:#b35a00;}
+.b-ok{background:#e6f6ea;color:#1a7f37;} .b-fail{background:#fde8e8;color:#b42318;} .b-unk{background:#eee;color:#555;}
+.kv{font-size:12.5px;line-height:1.6;margin:6px 0;} .kv b{color:#444;}
+.warn{background:#fff8e1;border:1px solid #ffe082;border-radius:6px;padding:8px 10px;margin:6px 0;font-size:12.5px;color:#7a5c00;}
+a{color:#1a5fb4;} code{background:#f0f0f0;padding:1px 4px;border-radius:3px;font-size:12px;}
+"""
+
+
+def _badge_label(label):
+    if label == "normal":
+        return '<span class="badge b-normal">normal</span>'
+    if label == "abnormal_probe":
+        return '<span class="badge b-abn">abnormal_probe</span>'
+    return f'<span class="badge b-unk">{html.escape(str(label))}</span>'
+
+
+def _badge_success(s):
+    if s is True:
+        return '<span class="badge b-ok">success: true</span>'
+    if s is False:
+        return '<span class="badge b-fail">success: false</span>'
+    return '<span class="badge b-unk">success: unknown</span>'
+
+
+def render_playlist_html(run_dir: Path, plan: list[dict], montage_rel: str | None,
+                         montage_mode: str, clip_seconds: float, clips_ok: dict) -> str:
+    p = []
+    p.append("<!DOCTYPE html><html><head><meta charset='utf-8'>")
+    p.append(f"<title>LIBERO smoke-test review montage — {html.escape(run_dir.name)}</title>")
+    p.append(f"<style>{_CSS}</style></head><body>")
+    p.append("<h1>LIBERO smoke-test — video review</h1>")
+    p.append(f"<div class='sub'>run dir: <code>{html.escape(str(run_dir))}</code> &nbsp;|&nbsp; "
+             f"generated {html.escape(_dt.datetime.now().isoformat(timespec='seconds'))} &nbsp;|&nbsp; "
+             f"{len(plan)} scenarios &nbsp;|&nbsp; clip length ≤ {clip_seconds:g}s &nbsp;|&nbsp; montage mode: {html.escape(montage_mode)}</div>")
+    p.append("<div class='top'>")
+    if montage_rel:
+        p.append(f"<h3 style='margin-top:0'>review_montage.mp4</h3>")
+        p.append(f"<video controls preload='metadata' src='{html.escape(montage_rel)}'></video>")
+        p.append(f"<div class='kv'><a href='{html.escape(montage_rel)}'>{html.escape(montage_rel)}</a></div>")
+    else:
+        p.append("<div class='warn'>review_montage.mp4 was not produced (no usable clips, or ffmpeg failed — see console output).</div>")
+    p.append("</div>")
+    p.append("<div class='grid'>")
+    for i, e in enumerate(plan):
+        sid = e["scenario_id"]
+        p.append("<div class='card'>")
+        p.append(f"<h3>[{i + 1}/{len(plan)}] {html.escape(sid)}</h3>")
+        p.append("<div class='kv'>" + _badge_label(e["label"]) + " " + _badge_success(e["success"])
+                 + f" &nbsp; seed={html.escape(str(e['seed']))}</div>")
+        clip_ok = clips_ok.get(sid, False)
+        if clip_ok and (run_dir / e["clip_rel"]).exists():
+            p.append(f"<video controls preload='metadata' src='{html.escape(e['clip_rel'])}'></video>")
+        elif e["warning"]:
+            p.append(f"<div class='warn'>⚠ {html.escape(str(e['warning']))} — no review clip for this scenario.</div>")
+        else:
+            p.append("<div class='warn'>⚠ review clip not generated (ffmpeg failed for this scenario — see console).</div>")
+        p.append("<div class='kv'>"
+                 f"<b>instruction:</b> {html.escape(str(e['instruction']))}<br>"
+                 + (f"<b>env:</b> <code>{html.escape(str(e['env_name']))}</code><br>" if e.get("env_name") else "")
+                 + f"<b>source video:</b> {e['src_duration']:g}s"
+                 + (f" &nbsp; <a href='{html.escape(e['video_rel'])}'>full video.mp4</a>" if e.get("video_rel") else " (missing)")
+                 + (f"<br><b>clip:</b> <a href='{html.escape(e['clip_rel'])}'>{html.escape(e['clip_rel'])}</a> ({e['clip_duration']:g}s)" if clip_ok else "")
+                 + "</div>")
+        p.append("</div>")
+    p.append("</div></body></html>")
+    return "\n".join(p)
+
+
+# --------------------------------------------------------------------------- #
+# main
+# --------------------------------------------------------------------------- #
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--run-dir", required=True)
+    ap.add_argument("--clip-seconds", type=float, default=5.0, help="Max clip length per scenario (full video if shorter).")
+    ap.add_argument("--mode", choices=["sequential", "grid"], default="sequential")
+    ap.add_argument("--hold-seconds", type=float, default=1.0,
+                    help="In sequential mode, freeze the last frame of each clip this long so the overlay is readable.")
+    ap.add_argument("--instruction-chars", type=int, default=56, help="Truncate the overlaid instruction to this many chars.")
+    ap.add_argument("--cols", type=int, default=0, help="Grid mode: number of columns (0 = auto = ceil(sqrt(N))).")
+    ap.add_argument("--width", type=int, default=512, help="Clip width in px (height auto, aspect preserved).")
+    ap.add_argument("--open", action="store_true", help="Open review_playlist.html when done.")
+    ap.add_argument("--dry-run", action="store_true", help="Print the planned clips and exit without writing anything.")
+    args = ap.parse_args()
+
+    run_dir = Path(args.run_dir).resolve()
+    if not run_dir.is_dir():
+        _die(f"ERROR: --run-dir does not exist or is not a directory: {run_dir}")
+
+    ffmpeg, ffprobe = _ensure_ffmpeg()
+    font = _find_font()
+
+    plan = build_plan(run_dir, ffprobe, args.clip_seconds, args.instruction_chars)
+    if not plan:
+        _die(f"ERROR: no scenario directories found under {run_dir}")
+
+    n = len(plan)
+    n_with_video = sum(1 for e in plan if e["ok_source"])
+    n_missing = n - n_with_video
+
+    # --- dry run -------------------------------------------------------------
+    if args.dry_run:
+        print(f"[dry-run] run dir: {run_dir}")
+        print(f"[dry-run] {n} scenarios; {n_with_video} with usable video.mp4; {n_missing} missing/unreadable")
+        print(f"[dry-run] mode={args.mode}  clip-seconds={args.clip_seconds:g}  hold-seconds={args.hold_seconds:g}  font={font or '(builtin sans)'}")
+        print(f"[dry-run] would write:")
+        print(f"           {run_dir / 'review_clips'}/<scenario_id>.mp4")
+        print(f"           {run_dir / 'review_montage.mp4'}")
+        print(f"           {run_dir / 'review_playlist.html'}")
+        print()
+        hdr = f"{'#':>2}  {'scenario_id':40s}  {'label':14s}  {'seed':>5s}  {'src_s':>6s}  {'clip_s':>6s}  {'success':8s}  note"
+        print(hdr)
+        print("-" * len(hdr))
+        for i, e in enumerate(plan):
+            print(f"{i + 1:>2}  {e['scenario_id'][:40]:40s}  {str(e['label'])[:14]:14s}  "
+                  f"{str(e['seed']):>5s}  {e['src_duration']:>6.2f}  {e['clip_duration']:>6.2f}  "
+                  f"{_success_str(e['success']):8s}  {e['warning'] or 'ok -> ' + e['clip_rel']}")
+            print(f"      instruction: {e['instruction_short']}")
+        return 0
+
+    if n_with_video == 0:
+        # still write an HTML so the user has something, but no montage.
+        print(f"WARNING: none of the {n} scenarios has a usable video.mp4 — writing an HTML with warnings only.", file=sys.stderr)
+        html_path = run_dir / "review_playlist.html"
+        html_path.write_text(render_playlist_html(run_dir, plan, None, args.mode, args.clip_seconds, {}), encoding="utf-8")
+        print("\n==== SUMMARY ====")
+        print(f"total scenarios : {n}")
+        print(f"clips generated : 0")
+        print(f"missing videos  : {n_missing}  ({', '.join(e['scenario_id'] for e in plan if not e['ok_source'])})")
+        print(f"montage path    : (not produced)")
+        print(f"html path       : {html_path}")
+        return 1
+
+    # --- generate per-scenario clips ----------------------------------------
+    (run_dir / "review_clips").mkdir(parents=True, exist_ok=True)
+    tmp_root = Path(tempfile.mkdtemp(prefix="libero_montage_"))
+    clips_ok: dict[str, bool] = {}
+    generated_clips: list[Path] = []
+    warnings: list[str] = []
+    target_w = args.width - (args.width % 2)  # even
+    for i, e in enumerate(plan):
+        sid = e["scenario_id"]
+        if not e["ok_source"]:
+            clips_ok[sid] = False
+            warnings.append(f"[{sid}] {e['warning'] or 'no usable video'} — skipped")
+            print(f"[{i + 1}/{n}] {sid}: SKIP ({e['warning'] or 'no usable video'})")
+            continue
+        ok, err = make_clip(ffmpeg, font, e, i, n, tmp_root, target_w)
+        clips_ok[sid] = ok
+        if ok and e["clip_path"].exists():
+            generated_clips.append(e["clip_path"])
+            print(f"[{i + 1}/{n}] {sid}: clip -> {e['clip_rel']}  ({e['clip_duration']:g}s)")
+        else:
+            warnings.append(f"[{sid}] clip generation failed: {err}")
+            print(f"[{i + 1}/{n}] {sid}: CLIP FAILED\n{err}", file=sys.stderr)
+
+    # --- montage -------------------------------------------------------------
+    montage_path = run_dir / "review_montage.mp4"
+    montage_rel = None
+    if generated_clips:
+        if args.mode == "grid":
+            cols = args.cols if args.cols and args.cols > 0 else max(1, math.ceil(math.sqrt(len(generated_clips))))
+            # cell size: keep clips' aspect (target_w x ~target_w/2), shrink so total width <= ~1920
+            src_info = next((e["src_info"] for e in plan if e["ok_source"] and e["src_info"]), {"width": 512, "height": 256})
+            ar = (src_info.get("height") or 256) / (src_info.get("width") or 512)
+            cell_w = min(target_w, max(160, (1920 // cols) - ((1920 // cols) % 2)))
+            cell_h = int(round(cell_w * ar))
+            cell_h -= cell_h % 2
+            ok, err = make_montage_grid(ffmpeg, generated_clips, montage_path, cols, cell_w, cell_h, ffprobe)
+            if not ok:
+                print(f"WARNING: grid montage failed ({err}); falling back to sequential montage.", file=sys.stderr)
+                warnings.append(f"grid montage failed: {err} (fell back to sequential)")
+                ok, err = make_montage_sequential(ffmpeg, generated_clips, montage_path, args.hold_seconds, target_w)
+        else:
+            ok, err = make_montage_sequential(ffmpeg, generated_clips, montage_path, args.hold_seconds, target_w)
+        if ok and montage_path.exists():
+            montage_rel = os.path.relpath(montage_path, run_dir)
+            print(f"montage -> {montage_rel}")
+        else:
+            warnings.append(f"montage generation failed: {err}")
+            print(f"WARNING: montage generation failed: {err}", file=sys.stderr)
+    else:
+        warnings.append("no clips were generated, so no montage was produced")
+
+    # --- HTML ----------------------------------------------------------------
+    html_path = run_dir / "review_playlist.html"
+    html_path.write_text(
+        render_playlist_html(run_dir, plan, montage_rel, args.mode, args.clip_seconds, clips_ok),
+        encoding="utf-8",
+    )
+
+    # cleanup temp dir
+    shutil.rmtree(tmp_root, ignore_errors=True)
+
+    # --- summary -------------------------------------------------------------
+    missing_ids = [e["scenario_id"] for e in plan if not e["ok_source"]]
+    failed_ids = [sid for sid, ok in clips_ok.items() if ok is False and sid not in missing_ids]
+    print("\n==== SUMMARY ====")
+    print(f"total scenarios : {n}")
+    print(f"clips generated : {len(generated_clips)}  ({run_dir / 'review_clips'}/)")
+    if failed_ids:
+        print(f"clip failures   : {len(failed_ids)}  ({', '.join(failed_ids)})")
+    print(f"missing videos  : {len(missing_ids)}" + (f"  ({', '.join(missing_ids)})" if missing_ids else ""))
+    print(f"montage path    : {montage_path if montage_rel else '(not produced)'}")
+    print(f"html path       : {html_path}")
+    if warnings:
+        print(f"\n{len(warnings)} warning(s):")
+        for w in warnings:
+            print(f"  - {w}")
+
+    if args.open:
+        import webbrowser
+        try:
+            webbrowser.open(html_path.as_uri())
+            print(f"\nOpened {html_path.as_uri()}")
+        except Exception:
+            print(f"\nCould not auto-open a browser; open manually:\n  {html_path}")
+
+    # exit non-zero only if we produced *nothing* useful
+    return 0 if generated_clips else 1
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

examples/LIBERO/smoke_tests/review_smoke_tests.py

@@ -0,0 +1,139 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""List smoke-test scenarios + their video paths, and optionally play them.
+
+Usage::
+
+    # most recent run under outputs/libero_smoke_tests/
+    python examples/LIBERO/smoke_tests/review_smoke_tests.py
+
+    # a specific run dir
+    python examples/LIBERO/smoke_tests/review_smoke_tests.py --run-dir outputs/libero_smoke_tests/20260512_120000
+
+    # try to open each video one by one (xdg-open / open / ffplay, if available)
+    python examples/LIBERO/smoke_tests/review_smoke_tests.py --open
+"""
+
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+import shutil
+import subprocess
+import sys
+
+REPO_ROOT = Path(__file__).resolve().parents[3]
+DEFAULT_OUTPUT_DIR = REPO_ROOT / "outputs" / "libero_smoke_tests"
+
+
+def _latest_run_dir(output_dir: Path) -> Path | None:
+    if not output_dir.exists():
+        return None
+    runs = sorted((p for p in output_dir.iterdir() if p.is_dir()), key=lambda p: p.name)
+    return runs[-1] if runs else None
+
+
+def _find_video(scenario_dir: Path) -> Path | None:
+    cand = scenario_dir / "video.mp4"
+    if cand.exists():
+        return cand
+    vids = sorted(scenario_dir.glob("**/*.mp4"))
+    return vids[0] if vids else None
+
+
+def _open_video(path: Path) -> None:
+    for opener in ("xdg-open", "open", "ffplay"):
+        exe = shutil.which(opener)
+        if exe:
+            args = [exe, str(path)]
+            if opener == "ffplay":
+                args = [exe, "-autoexit", "-loglevel", "error", str(path)]
+            print(f"  opening with: {' '.join(args)}")
+            try:
+                subprocess.run(args, check=False)
+            except Exception as e:  # noqa: BLE001
+                print(f"  (failed to open: {e})")
+            return
+    print("  no video opener found (xdg-open / open / ffplay). Path printed above.")
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--output-dir", default=str(DEFAULT_OUTPUT_DIR))
+    ap.add_argument("--run-dir", default=None, help="Specific <output-dir>/<timestamp> dir.")
+    ap.add_argument("--open", action="store_true", help="Try to play each video, pausing between.")
+    args = ap.parse_args()
+
+    run_dir = Path(args.run_dir) if args.run_dir else _latest_run_dir(Path(args.output_dir))
+    if run_dir is None or not run_dir.exists():
+        print(f"No run directory found under {args.output_dir}. Run run_10_smoke_tests.py first.",
+              file=sys.stderr)
+        return 1
+
+    print(f"Run directory: {run_dir}\n")
+
+    summary_path = run_dir / "summary.json"
+    rows = []
+    if summary_path.exists():
+        try:
+            rows = json.loads(summary_path.read_text()).get("scenarios", [])
+        except Exception:
+            rows = []
+
+    scenario_dirs = sorted(p for p in run_dir.iterdir() if p.is_dir())
+    by_id = {r.get("scenario_id"): r for r in rows}
+
+    for i, sc_dir in enumerate(scenario_dirs, 1):
+        sid = sc_dir.name
+        row = by_id.get(sid, {})
+        meta = {}
+        mp = sc_dir / "metadata.json"
+        if mp.exists():
+            try:
+                meta = json.loads(mp.read_text())
+            except Exception:
+                meta = {}
+        label = row.get("label") or meta.get("label", "?")
+        success = row.get("success", "unknown")
+        video = _find_video(sc_dir)
+        actions = sc_dir / "actions.npy"
+        print(f"[{i:2d}] {sid}")
+        print(f"     label={label}  success={success}  "
+              f"rollout_started={row.get('rollout_started')}  "
+              f"actions={'yes' if actions.exists() else 'no'}")
+        env_name = (meta.get("resolved") or {}).get("env_name") or (meta.get("manifest_entry") or {}).get("env_name")
+        if env_name:
+            print(f"     env: {env_name}")
+        if row.get("error_if_any"):
+            print(f"     error: {row['error_if_any']}")
+        if video:
+            print(f"     video: {video}")
+        else:
+            frames = sc_dir / "frames"
+            if frames.exists():
+                n = len(list(frames.glob('*.png')))
+                print(f"     frames: {frames}/ ({n} png)")
+            else:
+                print(f"     video: <none saved>")
+        print()
+        if args.open and video:
+            _open_video(video)
+            if i < len(scenario_dirs):
+                try:
+                    input("     [enter] for next, Ctrl-C to stop... ")
+                except (EOFError, KeyboardInterrupt):
+                    print()
+                    break
+
+    print(f"Summary: {summary_path if summary_path.exists() else '(not found)'}")
+    md = run_dir / "summary.md"
+    if md.exists():
+        print(f"Markdown summary: {md}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

examples/LIBERO/smoke_tests/run_10_smoke_tests.py

@@ -0,0 +1,590 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""Run 10 short, simulation-only LIBERO smoke tests against a GR00T policy server.
+
+What this does (and does NOT do):
+  * It does NOT use any physical robot hardware -- LIBERO simulation only.
+  * It does NOT rewrite the official evaluation path. Each scenario is run
+    through ``gr00t.eval.rollout_policy.run_rollout_gymnasium_policy`` via the
+    thin ``_libero_rollout_worker.py`` (which only adds action recording and a
+    couple of tiny, optional simulation-only perturbations).
+  * It does NOT fake successful rollouts. If the model checkpoint is missing,
+    or the server is unreachable, or the LIBERO sim env is not installed, it
+    fails with a clear, actionable error message.
+
+Typical usage (two terminals):
+
+  Terminal 1 - start the GR00T inference server::
+
+      uv run python gr00t/eval/run_gr00t_server.py \
+          --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 \
+          --embodiment-tag LIBERO_PANDA \
+          --use-sim-policy-wrapper
+
+  Terminal 2 - run the smoke tests::
+
+      uv run python examples/LIBERO/smoke_tests/run_10_smoke_tests.py \
+          --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 \
+          --manifest examples/LIBERO/smoke_tests/scenarios_10.yaml \
+          --output-dir outputs/libero_smoke_tests \
+          --max-episode-steps 50 --save-video --render
+
+Pass ``--start-server`` to have this script launch (and later kill) the server
+itself, or ``--dry-run`` to validate the manifest / paths without running sims.
+"""
+
+from __future__ import annotations
+
+import argparse
+import datetime as _dt
+import json
+import os
+from pathlib import Path
+import shutil
+import subprocess
+import sys
+import time
+import traceback
+
+REPO_ROOT = Path(__file__).resolve().parents[3]
+WORKER = Path(__file__).resolve().parent / "_libero_rollout_worker.py"
+DEFAULT_MANIFEST = Path(__file__).resolve().parent / "scenarios_10.yaml"
+RESULT_PREFIX = "SMOKE_RESULT_JSON:"
+
+# Path to the dedicated LIBERO uv venv created by setup_libero.sh.
+LIBERO_VENV_PYTHON = (
+    REPO_ROOT / "gr00t" / "eval" / "sim" / "LIBERO" / "libero_uv" / ".venv" / "bin" / "python"
+)
+
+SERVER_CMD_TEMPLATE = [
+    "uv", "run", "python", "gr00t/eval/run_gr00t_server.py",
+    "--model-path", "{model_path}",
+    "--embodiment-tag", "LIBERO_PANDA",
+    "--use-sim-policy-wrapper",
+]
+
+
+# --------------------------------------------------------------------------- #
+# Helpers
+# --------------------------------------------------------------------------- #
+def _load_manifest(path: Path) -> list[dict]:
+    try:
+        import yaml
+    except ImportError as e:  # pragma: no cover
+        raise SystemExit(
+            "PyYAML is required to read the scenario manifest. Install it with "
+            "`uv pip install pyyaml` (it is a transitive dependency of the gr00t "
+            "package, so this normally just works inside the project venv)."
+        ) from e
+    if not path.exists():
+        raise SystemExit(f"Manifest not found: {path}")
+    with open(path) as f:
+        data = yaml.safe_load(f)
+    scenarios = data.get("scenarios") if isinstance(data, dict) else data
+    if not isinstance(scenarios, list) or not scenarios:
+        raise SystemExit(f"Manifest {path} does not contain a non-empty 'scenarios' list.")
+    return scenarios
+
+
+def _check_model_path(model_path: Path) -> None:
+    if model_path.exists():
+        # Must look like a real checkpoint dir.
+        has_cfg = (model_path / "config.json").exists()
+        has_weights = any(model_path.glob("*.safetensors")) or (model_path / "pytorch_model.bin").exists()
+        if has_cfg and has_weights:
+            return
+        missing = []
+        if not has_cfg:
+            missing.append("config.json")
+        if not has_weights:
+            missing.append("model-*.safetensors / model.safetensors.index.json")
+        raise SystemExit(
+            f"Model path {model_path} exists but is missing: {', '.join(missing)}.\n"
+            "Re-download the checkpoint:\n\n" + _download_hint()
+        )
+    raise SystemExit(
+        f"Model checkpoint not found at: {model_path}\n\n"
+        "Download it first (HuggingFace does not support nested repo paths directly):\n\n"
+        + _download_hint()
+    )
+
+
+def _download_hint() -> str:
+    return (
+        "  uv run hf download nvidia/GR00T-N1.7-LIBERO \\\n"
+        "    --include \"libero_10/config.json\" \\\n"
+        "              \"libero_10/embodiment_id.json\" \\\n"
+        "              \"libero_10/model-*.safetensors\" \\\n"
+        "              \"libero_10/model.safetensors.index.json\" \\\n"
+        "              \"libero_10/processor_config.json\" \\\n"
+        "              \"libero_10/statistics.json\" \\\n"
+        "    --local-dir checkpoints/GR00T-N1.7-LIBERO\n"
+        "  # (also fetch libero_10/config.json explicitly if the glob above skips it)\n"
+    )
+
+
+def _server_command(model_path: Path) -> list[str]:
+    return [tok.format(model_path=str(model_path)) for tok in SERVER_CMD_TEMPLATE]
+
+
+# The GR00T-N1.7 backbone (nvidia/Cosmos-Reason2-2B, a Qwen3-VL model) is a
+# *gated* HuggingFace repo. Loading any GR00T-N1.7 checkpoint pulls that base
+# repo's config/processor, so the server will fail to start without HF auth +
+# granted access. We surface a clear hint when we detect this.
+GATED_BACKBONE_HINT = (
+    "The GR00T-N1.7 backbone 'nvidia/Cosmos-Reason2-2B' is a GATED HuggingFace repo.\n"
+    "To start the server you must:\n"
+    "  1. Request access at https://huggingface.co/nvidia/Cosmos-Reason2-2B (one click, usually instant).\n"
+    "  2. Authenticate, e.g.  export HF_TOKEN=hf_xxx   (or: uv run hf auth login)\n"
+    "  3. Re-run the server / smoke tests.\n"
+)
+
+
+def _scan_log_for_gated_repo(log_path: Path) -> bool:
+    try:
+        text = log_path.read_text(errors="replace")
+    except OSError:
+        return False
+    return ("gated repo" in text) or ("Cosmos-Reason2-2B is restricted" in text) or (
+        "Access to model nvidia/Cosmos-Reason2-2B" in text
+    )
+
+
+def _ping_server(host: str, port: int, timeout_ms: int = 3000) -> bool:
+    """Return True if a GR00T policy server answers on host:port.
+
+    First does a cheap TCP connect (so we don't drag in the heavy torch/gr00t
+    import stack when nothing is listening); only if *something* is listening do
+    we import ``PolicyClient`` and validate the msgpack ``ping`` endpoint.
+    """
+    import socket
+
+    try:
+        with socket.create_connection((host, port), timeout=timeout_ms / 1000.0):
+            pass
+    except OSError:
+        return False
+    # Something is listening -- confirm it actually speaks the GR00T protocol.
+    try:
+        from gr00t.policy.server_client import PolicyClient
+
+        client = PolicyClient(host=host, port=port, timeout_ms=timeout_ms)
+        return bool(client.ping())
+    except Exception:
+        # Reachable on TCP but the client/import failed; treat as "up enough".
+        return True
+
+
+def _resolve_libero_python(explicit: str | None) -> str:
+    if explicit:
+        p = Path(explicit)
+        if not p.exists():
+            raise SystemExit(f"--libero-python {explicit} does not exist.")
+        return str(p)
+    if LIBERO_VENV_PYTHON.exists():
+        return str(LIBERO_VENV_PYTHON)
+    # Fall back to the current interpreter, but warn -- LIBERO needs its own venv.
+    print(
+        "WARNING: the dedicated LIBERO uv venv was not found at\n"
+        f"  {LIBERO_VENV_PYTHON}\n"
+        "Falling back to the current Python interpreter. If LIBERO / robosuite\n"
+        "are not importable there, set up the sim env first:\n"
+        "  sudo apt update && sudo apt install libegl1-mesa-dev libglu1-mesa\n"
+        "  bash gr00t/eval/sim/LIBERO/setup_libero.sh\n",
+        file=sys.stderr,
+    )
+    return sys.executable
+
+
+def _start_server(model_path: Path, host: str, port: int, log_path: Path):
+    cmd = [
+        sys.executable, str(REPO_ROOT / "gr00t" / "eval" / "run_gr00t_server.py"),
+        "--model-path", str(model_path),
+        "--embodiment-tag", "LIBERO_PANDA",
+        "--use-sim-policy-wrapper",
+        "--host", host,
+        "--port", str(port),
+    ]
+    log_f = open(log_path, "w")
+    print(f"Starting GR00T server (logging to {log_path}):\n  {' '.join(cmd)}")
+    proc = subprocess.Popen(cmd, stdout=log_f, stderr=subprocess.STDOUT, cwd=str(REPO_ROOT))
+    return proc, log_f
+
+
+# --------------------------------------------------------------------------- #
+# Main
+# --------------------------------------------------------------------------- #
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--model-path", default="checkpoints/GR00T-N1.7-LIBERO/libero_10")
+    ap.add_argument("--host", default="127.0.0.1")
+    ap.add_argument("--port", type=int, default=5555)
+    ap.add_argument("--manifest", default=str(DEFAULT_MANIFEST))
+    ap.add_argument("--output-dir", default="outputs/libero_smoke_tests")
+    ap.add_argument("--max-episode-steps", type=int, default=50,
+                    help="Upper cap (and default) on max_episode_steps for every scenario.")
+    ap.add_argument("--n-action-steps", type=int, default=8)
+    ap.add_argument("--save-video", action="store_true",
+                    help="Force-enable video recording for every scenario.")
+    ap.add_argument("--render", action="store_true",
+                    help="Also dump decoded frames into <scenario>/frames/.")
+    ap.add_argument("--dry-run", action="store_true",
+                    help="Validate manifest/model path/server, write metadata, run no sims.")
+    ap.add_argument("--resume", action="store_true",
+                    help="Skip scenarios that already have a rollout_summary.json in the run dir.")
+    ap.add_argument("--run-dir", default=None,
+                    help="Existing <output-dir>/<timestamp> dir to resume into "
+                         "(default: create a new timestamped dir).")
+    ap.add_argument("--libero-python", default=None,
+                    help="Path to the LIBERO uv venv python (default: auto-detect).")
+    ap.add_argument("--start-server", action="store_true",
+                    help="Launch the GR00T server in a subprocess and kill it at the end.")
+    ap.add_argument("--per-scenario-timeout", type=int, default=900,
+                    help="Hard timeout (seconds) for each scenario subprocess.")
+    args = ap.parse_args()
+
+    os.chdir(REPO_ROOT)
+
+    model_path = Path(args.model_path)
+    manifest_path = Path(args.manifest)
+    scenarios = _load_manifest(manifest_path)
+
+    print(f"Loaded {len(scenarios)} scenarios from {manifest_path}")
+    n_normal = sum(1 for s in scenarios if s.get("label") == "normal")
+    n_abnormal = sum(1 for s in scenarios if s.get("label") == "abnormal_probe")
+    print(f"  labels: {n_normal} normal, {n_abnormal} abnormal_probe")
+
+    server_cmd_str = " ".join(_server_command(model_path))
+    print("\nExpected GR00T server command (run this in a separate terminal):\n  " + server_cmd_str + "\n")
+    print("NOTE: " + GATED_BACKBONE_HINT)
+
+    # ---- validate model path -------------------------------------------------
+    _check_model_path(model_path)
+    print(f"Model checkpoint OK: {model_path}")
+
+    # ---- output dir ----------------------------------------------------------
+    out_root = Path(args.output_dir)
+    if args.run_dir:
+        run_dir = Path(args.run_dir)
+        run_dir.mkdir(parents=True, exist_ok=True)
+    else:
+        ts = _dt.datetime.now().strftime("%Y%m%d_%H%M%S")
+        run_dir = out_root / ts
+        run_dir.mkdir(parents=True, exist_ok=True)
+    print(f"Run directory: {run_dir}")
+
+    # ---- optionally start server --------------------------------------------
+    server_proc = None
+    server_log_f = None
+    if args.start_server and not args.dry_run:
+        server_proc, server_log_f = _start_server(model_path, args.host, args.port,
+                                                  run_dir / "server.log")
+
+    # ---- check server reachability ------------------------------------------
+    reachable = _ping_server(args.host, args.port)
+    if args.start_server and not args.dry_run and not reachable:
+        # give the server time to load the (multi-GB) model
+        print("Waiting for the GR00T server to come up (loading model can take a few minutes)...")
+        for _ in range(120):
+            time.sleep(5)
+            if server_proc is not None and server_proc.poll() is not None:
+                hint = ""
+                if _scan_log_for_gated_repo(run_dir / "server.log"):
+                    hint = "\n\n" + GATED_BACKBONE_HINT
+                raise SystemExit(
+                    f"The GR00T server process exited (rc={server_proc.returncode}); "
+                    f"see {run_dir / 'server.log'}.{hint}"
+                )
+            if _ping_server(args.host, args.port):
+                reachable = True
+                break
+
+    if not reachable and not args.dry_run:
+        gated_hint = ""
+        if args.start_server and _scan_log_for_gated_repo(run_dir / "server.log"):
+            gated_hint = "\n" + GATED_BACKBONE_HINT
+        msg = (
+            f"\nERROR: no GR00T policy server is reachable at {args.host}:{args.port}.\n\n"
+            "Start it first in a separate terminal:\n\n  " + server_cmd_str + "\n\n"
+            "...then re-run this script. (Or pass --start-server to have this script\n"
+            "launch it for you, or --dry-run to validate everything without running sims.)\n"
+            + gated_hint
+        )
+        print(msg, file=sys.stderr)
+        return 2
+    if reachable:
+        print(f"GR00T server reachable at {args.host}:{args.port}.")
+    else:
+        print("(dry-run) skipping server reachability requirement.")
+
+    libero_python = _resolve_libero_python(args.libero_python)
+    print(f"LIBERO rollout interpreter: {libero_python}")
+
+    # ---- run scenarios -------------------------------------------------------
+    rows: list[dict] = []
+    try:
+        for i, sc in enumerate(scenarios, 1):
+            sid = sc["id"]
+            label = sc.get("label", "normal")
+            seed = int(sc.get("seed", 0))
+            # Keep a scenario's own (smaller) budget, but cap larger ones at --max-episode-steps.
+            mes = min(int(sc.get("max_episode_steps", args.max_episode_steps)), args.max_episode_steps)
+            n_action_steps = int(sc.get("n_action_steps", args.n_action_steps))
+            env_name = sc["env_name"]
+            save_video = bool(sc.get("save_video", True)) or args.save_video
+            obs_noise_std = float(sc.get("obs_noise_std", 0.0) or 0.0)
+            action_repeat = int(sc.get("action_repeat", 1) or 1)
+            instruction_override = sc.get("instruction_override")
+
+            sc_dir = run_dir / sid
+            sc_dir.mkdir(parents=True, exist_ok=True)
+
+            print(f"\n[{i}/{len(scenarios)}] {sid}  ({label})  env={env_name}  "
+                  f"seed={seed}  max_episode_steps={mes}")
+
+            # metadata.json (always written)
+            metadata = {
+                "scenario_id": sid,
+                "label": label,
+                "manifest_entry": sc,
+                "resolved": {
+                    "env_name": env_name,
+                    "seed": seed,
+                    "max_episode_steps": mes,
+                    "n_action_steps": n_action_steps,
+                    "save_video": save_video,
+                    "save_frames": bool(args.render),
+                    "obs_noise_std": obs_noise_std,
+                    "action_repeat": action_repeat,
+                    "instruction_override": instruction_override,
+                },
+                "model_path": str(model_path),
+                "policy_server": {"host": args.host, "port": args.port},
+                "libero_python": libero_python,
+                "timestamp": _dt.datetime.now().isoformat(timespec="seconds"),
+            }
+            with open(sc_dir / "metadata.json", "w") as f:
+                json.dump(metadata, f, indent=2)
+
+            if instruction_override:
+                print(f"  NOTE: instruction_override is set ({instruction_override!r}) but is not "
+                      "yet wired into the LIBERO env path; it will be ignored by this rollout.")
+
+            # resume?
+            if args.resume and (sc_dir / "rollout_summary.json").exists():
+                try:
+                    prev = json.loads((sc_dir / "rollout_summary.json").read_text())
+                except Exception:
+                    prev = {}
+                print(f"  --resume: existing rollout_summary.json found, skipping.")
+                rows.append({
+                    "scenario_id": sid, "label": label, "seed": seed,
+                    "rollout_started": prev.get("error") is None,
+                    "actions_produced": bool(prev.get("actions_path")),
+                    "video_saved": bool(prev.get("video_path")),
+                    "success": prev.get("success", "unknown"),
+                    "output_dir": str(sc_dir),
+                    "error_if_any": prev.get("error"),
+                    "resumed": True,
+                })
+                continue
+
+            if args.dry_run:
+                with open(sc_dir / "rollout_summary.json", "w") as f:
+                    json.dump({"status": "dry_run", "env_name": env_name, "seed": seed}, f, indent=2)
+                rows.append({
+                    "scenario_id": sid, "label": label, "seed": seed,
+                    "rollout_started": False, "actions_produced": False, "video_saved": False,
+                    "success": "unknown", "output_dir": str(sc_dir),
+                    "error_if_any": "dry-run (no rollout executed)",
+                })
+                continue
+
+            # build worker command
+            cmd = [
+                libero_python, str(WORKER),
+                "--env-name", env_name,
+                "--host", args.host, "--port", str(args.port),
+                "--max-episode-steps", str(mes),
+                "--n-action-steps", str(n_action_steps),
+                "--seed", str(seed),
+                "--out-dir", str(sc_dir),
+                "--obs-noise-std", str(obs_noise_std),
+                "--action-repeat", str(action_repeat),
+            ]
+            cmd += ["--save-video"] if save_video else ["--no-save-video"]
+            if args.render and save_video:
+                cmd += ["--save-frames"]
+
+            stdout_path = sc_dir / "stdout.log"
+            stderr_path = sc_dir / "stderr.log"
+            t0 = time.time()
+            try:
+                with open(stdout_path, "w") as so, open(stderr_path, "w") as se:
+                    proc = subprocess.run(
+                        cmd, stdout=so, stderr=se, cwd=str(REPO_ROOT),
+                        timeout=args.per_scenario_timeout,
+                    )
+                rc = proc.returncode
+            except subprocess.TimeoutExpired:
+                rc = -9
+                with open(stderr_path, "a") as se:
+                    se.write(f"\n[runner] scenario timed out after {args.per_scenario_timeout}s\n")
+            elapsed = time.time() - t0
+
+            # parse worker result
+            result = None
+            try:
+                for line in reversed(stdout_path.read_text().splitlines()):
+                    line = line.strip()
+                    if line.startswith(RESULT_PREFIX):
+                        result = json.loads(line[len(RESULT_PREFIX):].strip())
+                        break
+            except Exception:
+                pass
+            if result is None:
+                err_tail = ""
+                try:
+                    err_tail = "\n".join(stderr_path.read_text().splitlines()[-5:])
+                except Exception:
+                    pass
+                result = {
+                    "ok": False, "rollout_started": False, "actions_produced": False,
+                    "video_saved": False, "success": None,
+                    "error": (f"worker exited rc={rc} with no result line; "
+                              f"see {stderr_path}. tail:\n{err_tail}"),
+                }
+
+            status = "ok" if result.get("ok") else "FAILED"
+            print(f"  -> {status}  rc={rc}  {elapsed:.1f}s  "
+                  f"actions={'yes' if result.get('actions_produced') else 'no'}  "
+                  f"video={'yes' if result.get('video_saved') else 'no'}  "
+                  f"success={result.get('success')}")
+            if result.get("error"):
+                print(f"     error: {result['error']}")
+
+            rows.append({
+                "scenario_id": sid, "label": label, "seed": seed,
+                "rollout_started": bool(result.get("rollout_started")),
+                "actions_produced": bool(result.get("actions_produced")),
+                "video_saved": bool(result.get("video_saved")),
+                "success": result.get("success") if result.get("success") is not None else "unknown",
+                "output_dir": str(sc_dir),
+                "error_if_any": result.get("error"),
+                "rc": rc,
+                "elapsed_sec": round(elapsed, 1),
+                "n_action_calls": result.get("n_action_calls"),
+                "episode_length": result.get("episode_length"),
+                "video_path": result.get("video_path"),
+                "actions_path": result.get("actions_path"),
+            })
+    finally:
+        if server_proc is not None:
+            print("\nStopping GR00T server subprocess...")
+            server_proc.terminate()
+            try:
+                server_proc.wait(timeout=15)
+            except subprocess.TimeoutExpired:
+                server_proc.kill()
+            if server_log_f is not None:
+                server_log_f.close()
+
+    # ---- summary -------------------------------------------------------------
+    summary = {
+        "run_dir": str(run_dir),
+        "manifest": str(manifest_path),
+        "model_path": str(model_path),
+        "policy_server": {"host": args.host, "port": args.port},
+        "server_command": server_cmd_str,
+        "dry_run": args.dry_run,
+        "n_scenarios": len(scenarios),
+        "n_normal": n_normal,
+        "n_abnormal_probe": n_abnormal,
+        "timestamp": _dt.datetime.now().isoformat(timespec="seconds"),
+        "scenarios": rows,
+        "totals": {
+            "rollout_started": sum(1 for r in rows if r.get("rollout_started")),
+            "actions_produced": sum(1 for r in rows if r.get("actions_produced")),
+            "video_saved": sum(1 for r in rows if r.get("video_saved")),
+            "errors": sum(1 for r in rows if r.get("error_if_any") and "dry-run" not in str(r.get("error_if_any"))),
+        },
+    }
+    write_summary(run_dir, summary)
+    print_table(rows)
+    print(f"\nWrote: {run_dir / 'summary.json'}")
+    print(f"Wrote: {run_dir / 'summary.md'}")
+    print(f"\nReview the videos with:\n  python {Path(__file__).resolve().parent / 'review_smoke_tests.py'} "
+          f"--run-dir {run_dir}\n")
+
+    # exit code: 0 if dry-run or every scenario at least started; else 1
+    if args.dry_run:
+        return 0
+    failed = summary["totals"]["errors"]
+    if failed == 0:
+        return 0
+    print(f"{failed}/{len(scenarios)} scenario(s) reported an error -- exiting non-zero.", file=sys.stderr)
+    return 1
+
+
+def write_summary(run_dir: Path, summary: dict) -> None:
+    """Write summary.json and summary.md. Returns nothing; raises on JSON errors."""
+    with open(run_dir / "summary.json", "w") as f:
+        json.dump(summary, f, indent=2)
+    # markdown
+    lines = []
+    lines.append(f"# LIBERO smoke-test summary\n")
+    lines.append(f"- run dir: `{summary['run_dir']}`")
+    lines.append(f"- model: `{summary['model_path']}`")
+    lines.append(f"- server: `{summary['policy_server']['host']}:{summary['policy_server']['port']}`")
+    lines.append(f"- server command: `{summary['server_command']}`")
+    lines.append(f"- scenarios: {summary['n_scenarios']} ({summary['n_normal']} normal, "
+                 f"{summary['n_abnormal_probe']} abnormal_probe)")
+    if summary.get("dry_run"):
+        lines.append(f"- **DRY RUN** (no rollouts executed)")
+    t = summary["totals"]
+    lines.append(f"- totals: rollout_started={t['rollout_started']}, actions_produced={t['actions_produced']}, "
+                 f"video_saved={t['video_saved']}, errors={t['errors']}\n")
+    lines.append("| scenario_id | label | seed | rollout_started | actions_produced | video_saved | success | output_dir | error_if_any |")
+    lines.append("|---|---|---|---|---|---|---|---|---|")
+    for r in summary["scenarios"]:
+        err = (str(r.get("error_if_any")) or "").replace("\n", " ").replace("|", "\\|")
+        if len(err) > 160:
+            err = err[:157] + "..."
+        lines.append("| {id} | {label} | {seed} | {rs} | {ap} | {vs} | {succ} | {od} | {err} |".format(
+            id=r["scenario_id"], label=r["label"], seed=r["seed"],
+            rs="yes" if r.get("rollout_started") else "no",
+            ap="yes" if r.get("actions_produced") else "no",
+            vs="yes" if r.get("video_saved") else "no",
+            succ=r.get("success"), od=r["output_dir"], err=err or "",
+        ))
+    (run_dir / "summary.md").write_text("\n".join(lines) + "\n")
+
+
+def print_table(rows: list[dict]) -> None:
+    headers = ["scenario_id", "label", "seed", "rollout_started", "actions_produced",
+               "video_saved", "success", "output_dir", "error_if_any"]
+    def cell(r, h):
+        if h == "rollout_started":
+            return "yes" if r.get("rollout_started") else "no"
+        if h == "actions_produced":
+            return "yes" if r.get("actions_produced") else "no"
+        if h == "video_saved":
+            return "yes" if r.get("video_saved") else "no"
+        if h == "error_if_any":
+            e = str(r.get("error_if_any") or "")
+            e = e.replace("\n", " ")
+            return (e[:57] + "...") if len(e) > 60 else e
+        return str(r.get(h, ""))
+    table = [headers] + [[cell(r, h) for h in headers] for r in rows]
+    widths = [max(len(row[i]) for row in table) for i in range(len(headers))]
+    print("\n" + "=" * 8 + " SMOKE-TEST SUMMARY " + "=" * 8)
+    for ri, row in enumerate(table):
+        print("  " + " | ".join(c.ljust(widths[i]) for i, c in enumerate(row)))
+        if ri == 0:
+            print("  " + "-+-".join("-" * w for w in widths))
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

examples/LIBERO/smoke_tests/scenarios_10.yaml

@@ -0,0 +1,167 @@
+# LIBERO simulation-only smoke-test manifest (10 scenarios)
+#
+# These are *short* rollouts intended to verify the deploy/eval plumbing for
+# nvidia/GR00T-N1.7-LIBERO (libero_10 checkpoint). They are NOT a benchmark:
+# max_episode_steps is tiny on purpose, so success rates here are meaningless.
+#
+# Scenario fields
+#   id                : unique short identifier (used as the output sub-directory name)
+#   label             : "normal" | "abnormal_probe"
+#   env_name          : gymnasium id registered by gr00t.eval.sim.LIBERO.libero_env
+#                       (register_libero_envs registers libero_10 / spatial / object / goal / 90)
+#   instruction       : human-readable task instruction (informational; the *real*
+#                       instruction is supplied by the LIBERO env itself)
+#   instruction_override : optional string; if set, overrides the env's task_description
+#                          (used for instruction-mismatch probes). null = use env default.
+#   seed              : RNG seed forwarded to env.reset()/seed_everything for reproducibility
+#   max_episode_steps : max inner-env steps for this rollout (kept small for a smoke test)
+#   n_action_steps    : action chunk size executed per policy.get_action() call
+#   save_video        : whether to record an mp4 / frames for human review
+#   obs_noise_std     : optional float; std-dev of Gaussian noise added to image
+#                       observations *before* policy inference (0 = disabled)
+#   action_repeat     : optional int >= 1; if > 1, the previous action chunk is re-used
+#                       instead of querying the policy on (action_repeat-1)/action_repeat
+#                       of the steps (1 = disabled, i.e. normal behaviour)
+#   notes             : free-form description
+
+scenarios:
+
+  # -------------------- 8 normal scenarios --------------------
+  - id: normal_kitchen3_moka_pot_on_stove
+    label: normal
+    env_name: libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it
+    instruction: "turn on the stove and put the moka pot on it"
+    instruction_override: null
+    seed: 1000
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Baseline libero_10 long-horizon task; short rollout to check the full server<->client loop."
+
+  - id: normal_kitchen4_bowl_in_drawer
+    label: normal
+    env_name: libero_sim/KITCHEN_SCENE4_put_the_black_bowl_in_the_bottom_drawer_of_the_cabinet_and_close_it
+    instruction: "put the black bowl in the bottom drawer of the cabinet and close it"
+    instruction_override: null
+    seed: 1001
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Drawer manipulation task; verifies action decoding + sim stepping."
+
+  - id: normal_living1_soup_and_cheese_in_basket
+    label: normal
+    env_name: libero_sim/LIVING_ROOM_SCENE1_put_both_the_alphabet_soup_and_the_cream_cheese_box_in_the_basket
+    instruction: "put both the alphabet soup and the cream cheese box in the basket"
+    instruction_override: null
+    seed: 1002
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Multi-object pick-and-place; checks video recording wrapper output."
+
+  - id: normal_living2_soup_and_tomato_in_basket
+    label: normal
+    env_name: libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket
+    instruction: "put both the alphabet soup and the tomato sauce in the basket"
+    instruction_override: null
+    seed: 1003
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Second living-room scene; different seed for variety."
+
+  - id: normal_study1_book_in_caddy
+    label: normal
+    env_name: libero_sim/STUDY_SCENE1_pick_up_the_book_and_place_it_in_the_back_compartment_of_the_caddy
+    instruction: "pick up the book and place it in the back compartment of the caddy"
+    instruction_override: null
+    seed: 1004
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Study scene; thin/awkward object grasp."
+
+  - id: normal_kitchen8_both_moka_pots_on_stove
+    label: normal
+    env_name: libero_sim/KITCHEN_SCENE8_put_both_moka_pots_on_the_stove
+    instruction: "put both moka pots on the stove"
+    instruction_override: null
+    seed: 1005
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Two-object task; checks repeated subgoal handling within a short window."
+
+  - id: normal_kitchen6_mug_in_microwave
+    label: normal
+    env_name: libero_sim/KITCHEN_SCENE6_put_the_yellow_and_white_mug_in_the_microwave_and_close_it
+    instruction: "put the yellow and white mug in the microwave and close it"
+    instruction_override: null
+    seed: 1006
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Articulated-object (microwave door) task."
+
+  - id: normal_living5_two_mugs_on_plates
+    label: normal
+    env_name: libero_sim/LIVING_ROOM_SCENE5_put_the_white_mug_on_the_left_plate_and_put_the_yellow_and_white_mug_on_the_right_plate
+    instruction: "put the white mug on the left plate and put the yellow and white mug on the right plate"
+    instruction_override: null
+    seed: 1007
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: "Long-horizon dual placement; longest of the normal set but still capped short."
+
+  # -------------------- 2 abnormal_probe scenarios --------------------
+  - id: abnormal_probe_obs_noise
+    label: abnormal_probe
+    env_name: libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it
+    instruction: "turn on the stove and put the moka pot on it"
+    instruction_override: null
+    seed: 2000
+    max_episode_steps: 50
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 6.0
+    action_repeat: 1
+    notes: >-
+      Simulation-only perturbation: mild zero-mean Gaussian noise (sigma=6 in
+      0-255 pixel units) is added to the image observations *before* they are
+      sent to the policy. Smoke test only - this is a placeholder for future,
+      more principled distribution-shift probes; no complex anomaly logic.
+
+  - id: abnormal_probe_short_timeout
+    label: abnormal_probe
+    env_name: libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket
+    instruction: "put both the alphabet soup and the tomato sauce in the basket"
+    instruction_override: null
+    seed: 2001
+    max_episode_steps: 16
+    n_action_steps: 8
+    save_video: true
+    obs_noise_std: 0.0
+    action_repeat: 1
+    notes: >-
+      Simulation-only perturbation: deliberately shortened episode budget
+      (16 inner steps ~= 2 action chunks) so the long-horizon task cannot
+      possibly complete. Verifies that truncation / "unknown success" is
+      reported cleanly rather than crashing.

examples/LIBERO/smoke_tests/visualize_smoke_run.py

@@ -0,0 +1,739 @@
+# SPDX-FileCopyrightText: Copyright (c) 2026 NVIDIA CORPORATION & AFFILIATES. All rights reserved.
+# SPDX-License-Identifier: Apache-2.0
+#
+# Licensed under the Apache License, Version 2.0 (the "License").
+"""Build an HTML visual-review report for a completed LIBERO smoke-test run.
+
+Reads the artifacts produced by ``run_10_smoke_tests.py`` (summary.json,
+per-scenario metadata.json / rollout_summary.json / actions.npy / video.mp4 /
+frames/), computes a few simple action statistics, renders one card per
+scenario (with a video preview, links to the raw files, and 4 small plots),
+and writes:
+
+    <run-dir>/visual_report.html
+    <run-dir>/visual_summary.csv
+    <run-dir>/visual_summary.json
+    <run-dir>/plots/<scenario_id>/{action_norm,action_mean_per_dof,gripper_over_time,action_delta_norm}.png
+
+It does NOT touch any existing rollout output, does NOT need a GPU, and does
+NOT rerun LIBERO. Missing files become a warning inside the relevant card
+instead of crashing.
+
+CLI::
+
+    python examples/LIBERO/smoke_tests/visualize_smoke_run.py \
+        --run-dir outputs/libero_smoke_tests/20260512_122756 --open
+
+Flags: --run-dir  --open  --no-plots  --no-video-embed
+"""
+
+from __future__ import annotations
+
+import argparse
+import base64
+import csv
+import datetime as _dt
+import html
+import json
+import os
+from pathlib import Path
+import sys
+import tempfile
+import traceback
+
+# matplotlib must be configured before import; use a headless backend + a
+# writable config dir (the default ~/.config/matplotlib may be read-only).
+os.environ.setdefault("MPLCONFIGDIR", tempfile.mkdtemp(prefix="mpl-visualize-"))
+import numpy as np
+
+_MPL_OK = True
+_MPL_ERR = None
+try:
+    import matplotlib
+
+    matplotlib.use("Agg")
+    import matplotlib.pyplot as plt
+except Exception as e:  # noqa: BLE001
+    _MPL_OK = False
+    _MPL_ERR = repr(e)
+
+# The LIBERO action dict has 7 keys; `_libero_rollout_worker._action_to_numpy`
+# concatenates them in *sorted (alphabetical)* key order, so the columns of
+# actions.npy for this run are exactly:
+LIBERO_DOF_NAMES = [
+    "action.gripper",  # col 0
+    "action.pitch",    # col 1
+    "action.roll",     # col 2
+    "action.x",        # col 3
+    "action.y",        # col 4
+    "action.yaw",      # col 5
+    "action.z",        # col 6
+]
+LIBERO_GRIPPER_COL = 0
+
+
+# --------------------------------------------------------------------------- #
+# small IO helpers
+# --------------------------------------------------------------------------- #
+def _load_json(path: Path):
+    try:
+        with open(path) as f:
+            return json.load(f), None
+    except FileNotFoundError:
+        return None, f"missing file: {path.name}"
+    except Exception as e:  # noqa: BLE001
+        return None, f"failed to read {path.name}: {e!r}"
+
+
+def _rel(path: Path, start: Path) -> str:
+    try:
+        return os.path.relpath(path, start)
+    except ValueError:
+        return str(path)
+
+
+def _fmt(v) -> str:
+    if v is None:
+        return "—"
+    if isinstance(v, float):
+        return f"{v:.4g}"
+    return str(v)
+
+
+def _b64_data_uri(path: Path, mime: str) -> str | None:
+    try:
+        data = path.read_bytes()
+    except OSError:
+        return None
+    return f"data:{mime};base64," + base64.b64encode(data).decode("ascii")
+
+
+# --------------------------------------------------------------------------- #
+# action statistics
+# --------------------------------------------------------------------------- #
+def _squeeze_actions(arr: np.ndarray) -> np.ndarray:
+    """Best-effort normalize an actions.npy array to shape (n_calls, horizon, n_dof).
+
+    The worker saves shape (n_calls, batch=1, horizon, n_dof). We also tolerate
+    (n_calls, horizon, n_dof) and (n_calls, n_dof).
+    """
+    a = np.asarray(arr)
+    if a.dtype == object:
+        # list of heterogeneous things saved with allow_pickle; try to stack
+        try:
+            a = np.stack([np.asarray(x) for x in a], axis=0)
+        except Exception:
+            raise ValueError(f"cannot interpret object-dtype actions array of shape {arr.shape}")
+    if a.ndim == 4 and a.shape[1] == 1:
+        a = a[:, 0, :, :]
+    if a.ndim == 2:  # (n_calls, n_dof) -> add a length-1 horizon
+        a = a[:, None, :]
+    if a.ndim != 3:
+        raise ValueError(f"unexpected actions array shape {arr.shape} (squeezed to {a.shape})")
+    return a.astype(np.float64)
+
+
+def _detect_gripper_col(acts3: np.ndarray) -> tuple[int, list[str]]:
+    """Return (gripper_col_index, dof_names)."""
+    n_dof = acts3.shape[-1]
+    if n_dof == 7:
+        return LIBERO_GRIPPER_COL, list(LIBERO_DOF_NAMES)
+    names = [f"dof_{i}" for i in range(n_dof)]
+    # generic heuristic: the DoF whose values most often saturate near +/-1,
+    # falling back to the last column ("7th-style" gripper convention).
+    flat = acts3.reshape(-1, n_dof)
+    sat = (np.abs(flat) > 0.5).mean(axis=0)
+    col = int(np.argmax(sat)) if sat.size and sat.max() > 0.5 else n_dof - 1
+    return col, names
+
+
+def compute_action_stats(acts3: np.ndarray) -> dict:
+    """acts3: (n_calls, horizon, n_dof) -> dict of arrays + scalars."""
+    n_calls, horizon, n_dof = acts3.shape
+    gripper_col, dof_names = _detect_gripper_col(acts3)
+
+    # per-policy-call norms
+    step0_norm = np.linalg.norm(acts3[:, 0, :], axis=-1)             # (n_calls,) norm of the first executed step
+    chunk_mean = acts3.mean(axis=1)                                   # (n_calls, n_dof) mean over the chunk
+    chunk_mean_norm = np.linalg.norm(chunk_mean, axis=-1)             # (n_calls,)
+    per_step_norm = np.linalg.norm(acts3.reshape(-1, n_dof), axis=-1)  # (n_calls*horizon,)
+
+    # mean (and std) action value per DoF over all calls x steps
+    flat = acts3.reshape(-1, n_dof)
+    mean_per_dof = flat.mean(axis=0)
+    std_per_dof = flat.std(axis=0)
+
+    # gripper command over time: value at chunk-step 0 per call
+    gripper_step0 = acts3[:, 0, gripper_col]                          # (n_calls,)
+    gripper_all = flat[:, gripper_col]
+
+    # action-delta norm between consecutive *chunks* (Frobenius over horizon x dof)
+    if n_calls >= 2:
+        delta = acts3[1:] - acts3[:-1]                                # (n_calls-1, horizon, n_dof)
+        delta_norm = np.linalg.norm(delta.reshape(n_calls - 1, -1), axis=-1)
+    else:
+        delta_norm = np.zeros((0,), dtype=np.float64)
+
+    return {
+        "n_calls": n_calls,
+        "horizon": horizon,
+        "n_dof": n_dof,
+        "dof_names": dof_names,
+        "gripper_col": gripper_col,
+        "step0_norm": step0_norm,
+        "chunk_mean_norm": chunk_mean_norm,
+        "per_step_norm": per_step_norm,
+        "mean_per_dof": mean_per_dof,
+        "std_per_dof": std_per_dof,
+        "gripper_step0": gripper_step0,
+        "gripper_all": gripper_all,
+        "delta_norm": delta_norm,
+        # scalars for the comparison table
+        "mean_action_norm": float(per_step_norm.mean()) if per_step_norm.size else None,
+        "max_action_norm": float(per_step_norm.max()) if per_step_norm.size else None,
+        "mean_delta_norm": float(delta_norm.mean()) if delta_norm.size else 0.0,
+        "gripper_min": float(gripper_all.min()) if gripper_all.size else None,
+        "gripper_max": float(gripper_all.max()) if gripper_all.size else None,
+    }
+
+
+# --------------------------------------------------------------------------- #
+# plotting
+# --------------------------------------------------------------------------- #
+def _save_fig(fig, path: Path) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    fig.tight_layout()
+    fig.savefig(path, dpi=110)
+    plt.close(fig)
+
+
+def make_plots(stats: dict, out_dir: Path, scenario_id: str) -> dict:
+    """Write the 4 PNGs; return {logical_name: Path}. Requires matplotlib."""
+    out_dir.mkdir(parents=True, exist_ok=True)
+    calls = np.arange(stats["n_calls"])
+    paths: dict[str, Path] = {}
+
+    # 1) action norm over policy calls
+    fig, ax = plt.subplots(figsize=(5.6, 3.2))
+    ax.plot(calls, stats["step0_norm"], "o-", label="‖action‖ at chunk step 0")
+    ax.plot(calls, stats["chunk_mean_norm"], "s--", label="‖mean action over 16-step chunk‖")
+    ax.set_xlabel("policy call index")
+    ax.set_ylabel("L2 norm")
+    ax.set_title(f"{scenario_id}\naction L2 norm per policy call")
+    ax.grid(alpha=0.3)
+    ax.legend(fontsize=8)
+    p = out_dir / "action_norm.png"
+    _save_fig(fig, p)
+    paths["action_norm"] = p
+
+    # 2) mean action value per DoF (bar + std error bars)
+    fig, ax = plt.subplots(figsize=(5.6, 3.2))
+    x = np.arange(stats["n_dof"])
+    ax.bar(x, stats["mean_per_dof"], yerr=stats["std_per_dof"], capsize=3, color="#4C78A8")
+    ax.axhline(0.0, color="k", lw=0.6)
+    ax.set_xticks(x)
+    ax.set_xticklabels(stats["dof_names"], rotation=35, ha="right", fontsize=8)
+    ax.set_ylabel("mean value (±std) over all calls×steps")
+    gname = stats["dof_names"][stats["gripper_col"]]
+    ax.set_title(f"{scenario_id}\nmean action per DoF  (gripper DoF = {gname}, col {stats['gripper_col']})")
+    ax.grid(alpha=0.3, axis="y")
+    p = out_dir / "action_mean_per_dof.png"
+    _save_fig(fig, p)
+    paths["action_mean_per_dof"] = p
+
+    # 3) gripper command over time
+    fig, ax = plt.subplots(figsize=(5.6, 3.2))
+    ax.step(calls, stats["gripper_step0"], where="post", marker="o", color="#E45756")
+    ax.set_ylim(min(-1.1, float(stats["gripper_step0"].min()) - 0.1) if stats["gripper_step0"].size else -1.1,
+                max(1.1, float(stats["gripper_step0"].max()) + 0.1) if stats["gripper_step0"].size else 1.1)
+    ax.set_xlabel("policy call index")
+    ax.set_ylabel(f"{gname} (chunk step 0)")
+    ax.set_title(f"{scenario_id}\ngripper command over time  (DoF '{gname}', col {stats['gripper_col']})")
+    ax.grid(alpha=0.3)
+    p = out_dir / "gripper_over_time.png"
+    _save_fig(fig, p)
+    paths["gripper_over_time"] = p
+
+    # 4) action delta norm between consecutive chunks
+    fig, ax = plt.subplots(figsize=(5.6, 3.2))
+    if stats["delta_norm"].size:
+        ax.plot(np.arange(1, stats["n_calls"]), stats["delta_norm"], "o-", color="#54A24B")
+    else:
+        ax.text(0.5, 0.5, "only one policy call\n(no consecutive delta)", ha="center", va="center",
+                transform=ax.transAxes, fontsize=10)
+    ax.set_xlabel("policy call index i  (delta between chunk i and i-1)")
+    ax.set_ylabel("‖Δ chunk‖_F  (Frobenius over 16×n_dof)")
+    ax.set_title(f"{scenario_id}\naction delta norm between consecutive chunks")
+    ax.grid(alpha=0.3)
+    p = out_dir / "action_delta_norm.png"
+    _save_fig(fig, p)
+    paths["action_delta_norm"] = p
+
+    return paths
+
+
+# --------------------------------------------------------------------------- #
+# per-scenario processing
+# --------------------------------------------------------------------------- #
+def process_scenario(scenario_id: str, run_dir: Path, summary_row: dict | None,
+                     make_plots_flag: bool, embed_video: bool) -> dict:
+    """Return a dict with everything needed to render the card + table row."""
+    sc_dir = run_dir / scenario_id
+    warnings: list[str] = []
+    rec: dict = {"scenario_id": scenario_id, "scenario_dir": sc_dir, "warnings": warnings}
+
+    if not sc_dir.is_dir():
+        warnings.append(f"scenario directory not found: {sc_dir}")
+        rec.update(label=summary_row.get("label") if summary_row else None, seed=None,
+                   instruction=None, success=summary_row.get("success") if summary_row else None,
+                   max_episode_steps=None, num_policy_calls=None, actions_shape=None,
+                   error=(summary_row or {}).get("error_if_any"), stats=None, plot_paths={},
+                   video_path=None, frames_dir=None, n_frames=0,
+                   files={}, video_data_uri=None)
+        return rec
+
+    meta, err = _load_json(sc_dir / "metadata.json")
+    if err:
+        warnings.append(err)
+        meta = {}
+    roll, err = _load_json(sc_dir / "rollout_summary.json")
+    if err:
+        warnings.append(err)
+        roll = {}
+
+    manifest_entry = (meta or {}).get("manifest_entry", {}) or {}
+    resolved = (meta or {}).get("resolved", {}) or {}
+    label = manifest_entry.get("label") or (meta or {}).get("label") or (summary_row or {}).get("label")
+    seed = manifest_entry.get("seed", resolved.get("seed", (summary_row or {}).get("seed")))
+    instruction = manifest_entry.get("instruction") or roll.get("instruction")
+    env_name = resolved.get("env_name") or manifest_entry.get("env_name") or roll.get("env_name")
+    max_episode_steps = (resolved.get("max_episode_steps")
+                         or manifest_entry.get("max_episode_steps")
+                         or (roll or {}).get("requested_max_episode_steps"))
+    success = roll.get("success", (summary_row or {}).get("success"))
+    error = (roll or {}).get("error") or (summary_row or {}).get("error_if_any")
+    num_policy_calls = (roll or {}).get("n_get_action_calls") or (summary_row or {}).get("n_action_calls")
+    obs_noise_std = resolved.get("obs_noise_std", manifest_entry.get("obs_noise_std"))
+    action_repeat = resolved.get("action_repeat", manifest_entry.get("action_repeat"))
+    notes = manifest_entry.get("notes")
+
+    # actions.npy
+    actions_shape = None
+    stats = None
+    plot_paths: dict[str, Path] = {}
+    actions_path = sc_dir / "actions.npy"
+    if actions_path.exists():
+        try:
+            arr = np.load(actions_path, allow_pickle=True)
+            actions_shape = tuple(int(x) for x in np.asarray(arr).shape) if np.asarray(arr).dtype != object else f"object[{len(arr)}]"
+            acts3 = _squeeze_actions(arr)
+            stats = compute_action_stats(acts3)
+            if num_policy_calls is None:
+                num_policy_calls = stats["n_calls"]
+            if make_plots_flag and _MPL_OK:
+                try:
+                    plot_paths = make_plots(stats, run_dir / "plots" / scenario_id, scenario_id)
+                except Exception as e:  # noqa: BLE001
+                    warnings.append(f"plot generation failed: {e!r}")
+            elif make_plots_flag and not _MPL_OK:
+                warnings.append(f"matplotlib unavailable, plots skipped: {_MPL_ERR}")
+        except Exception as e:  # noqa: BLE001
+            warnings.append(f"failed to load/parse actions.npy: {e!r}")
+    else:
+        warnings.append("missing file: actions.npy")
+
+    # video
+    video_path = sc_dir / "video.mp4"
+    video_present = video_path.exists()
+    if not video_present:
+        warnings.append("missing file: video.mp4")
+    video_data_uri = None
+    if video_present and embed_video:
+        video_data_uri = _b64_data_uri(video_path, "video/mp4")
+        if video_data_uri is None:
+            warnings.append("failed to base64-embed video.mp4")
+
+    # frames/
+    frames_dir = sc_dir / "frames"
+    n_frames = 0
+    first_frame_uri = None
+    if frames_dir.is_dir():
+        frame_files = sorted(frames_dir.glob("*.png"))
+        n_frames = len(frame_files)
+        if not video_present and frame_files:
+            first_frame_uri = _b64_data_uri(frame_files[0], "image/png")
+    # (frames/ is optional; only warn if neither video nor frames exist)
+    if not video_present and n_frames == 0:
+        warnings.append("no video.mp4 and no frames/ — nothing to preview")
+
+    # raw-file links (present-or-not)
+    files = {}
+    for name in ("metadata.json", "rollout_summary.json", "stdout.log", "stderr.log"):
+        p = sc_dir / name
+        files[name] = {"rel": _rel(p, run_dir), "exists": p.exists()}
+        if not p.exists():
+            warnings.append(f"missing file: {name}")
+
+    rec.update(
+        label=label, seed=seed, instruction=instruction, env_name=env_name,
+        success=success, max_episode_steps=max_episode_steps, num_policy_calls=num_policy_calls,
+        actions_shape=actions_shape, error=error, obs_noise_std=obs_noise_std,
+        action_repeat=action_repeat, notes=notes, stats=stats, plot_paths=plot_paths,
+        video_path=video_path if video_present else None, video_present=video_present,
+        video_data_uri=video_data_uri, frames_dir=frames_dir if frames_dir.is_dir() else None,
+        n_frames=n_frames, first_frame_uri=first_frame_uri, files=files,
+        output_dir=str((summary_row or {}).get("output_dir") or _rel(sc_dir, run_dir.parent.parent)),
+        elapsed_sec=(roll or {}).get("elapsed_sec", (summary_row or {}).get("elapsed_sec")),
+        video_saved=video_present,
+    )
+    return rec
+
+
+# --------------------------------------------------------------------------- #
+# HTML rendering
+# --------------------------------------------------------------------------- #
+_CSS = """
+body{font-family:-apple-system,BlinkMacSystemFont,"Segoe UI",Roboto,Helvetica,Arial,sans-serif;
+     margin:0;padding:24px;background:#f4f5f7;color:#1a1a1a;}
+h1{margin:0 0 4px 0;} .sub{color:#666;margin-bottom:20px;font-size:14px;}
+.meta-grid{display:grid;grid-template-columns:repeat(auto-fit,minmax(220px,1fr));gap:6px 18px;
+           background:#fff;border:1px solid #e0e0e0;border-radius:8px;padding:14px 18px;margin-bottom:22px;font-size:13px;}
+.meta-grid b{color:#444;}
+table.cmp{border-collapse:collapse;width:100%;background:#fff;font-size:12.5px;margin-bottom:28px;
+          box-shadow:0 1px 3px rgba(0,0,0,.08);border-radius:8px;overflow:hidden;}
+table.cmp th,table.cmp td{border-bottom:1px solid #eee;padding:7px 10px;text-align:left;white-space:nowrap;}
+table.cmp th{background:#2d3748;color:#fff;position:sticky;top:0;}
+table.cmp tr:hover{background:#f7f9fc;}
+.badge{display:inline-block;padding:1px 8px;border-radius:10px;font-size:11px;font-weight:600;}
+.b-normal{background:#e3f0ff;color:#1a5fb4;} .b-abn{background:#ffe9d6;color:#b35a00;}
+.b-ok{background:#e6f6ea;color:#1a7f37;} .b-fail{background:#fde8e8;color:#b42318;} .b-unk{background:#eee;color:#555;}
+.card{background:#fff;border:1px solid #e0e0e0;border-radius:10px;padding:18px;margin-bottom:22px;
+      box-shadow:0 1px 3px rgba(0,0,0,.06);}
+.card h2{margin:0 0 2px 0;font-size:18px;} .card .cmeta{color:#555;font-size:13px;margin-bottom:10px;}
+.card .cols{display:grid;grid-template-columns:minmax(280px,360px) 1fr;gap:18px;}
+.kv{font-size:13px;line-height:1.7;} .kv b{color:#444;display:inline-block;min-width:130px;}
+.plots{display:grid;grid-template-columns:repeat(auto-fit,minmax(300px,1fr));gap:10px;}
+.plots img{width:100%;border:1px solid #eee;border-radius:6px;background:#fff;}
+video{width:100%;max-width:340px;border-radius:6px;background:#000;}
+.warn{background:#fff8e1;border:1px solid #ffe082;border-radius:6px;padding:8px 12px;margin:8px 0;font-size:12.5px;color:#7a5c00;}
+.err{background:#fdecea;border:1px solid #f5c6c0;border-radius:6px;padding:8px 12px;margin:8px 0;font-size:12.5px;color:#7a1f17;font-family:monospace;white-space:pre-wrap;}
+.files a{margin-right:12px;font-size:12.5px;} .files .missing{color:#aaa;text-decoration:line-through;}
+.notes{font-size:12.5px;color:#555;font-style:italic;margin-top:8px;}
+code{background:#f0f0f0;padding:1px 4px;border-radius:3px;font-size:12px;}
+"""
+
+
+def _badge_label(label: str | None) -> str:
+    if label == "normal":
+        return '<span class="badge b-normal">normal</span>'
+    if label == "abnormal_probe":
+        return '<span class="badge b-abn">abnormal_probe</span>'
+    return f'<span class="badge b-unk">{html.escape(str(label))}</span>'
+
+
+def _badge_success(s) -> str:
+    if s is True:
+        return '<span class="badge b-ok">success: true</span>'
+    if s is False:
+        return '<span class="badge b-fail">success: false</span>'
+    return '<span class="badge b-unk">success: unknown</span>'
+
+
+def render_html(run_dir: Path, summary: dict | None, records: list[dict], table_rows: list[dict],
+                args) -> str:
+    parts: list[str] = []
+    parts.append("<!DOCTYPE html><html><head><meta charset='utf-8'>")
+    parts.append(f"<title>LIBERO smoke-test visual report — {html.escape(run_dir.name)}</title>")
+    parts.append(f"<style>{_CSS}</style></head><body>")
+    parts.append("<h1>LIBERO smoke-test visual review</h1>")
+    parts.append(f"<div class='sub'>run dir: <code>{html.escape(str(run_dir))}</code> &nbsp;|&nbsp; "
+                 f"generated {html.escape(_dt.datetime.now().isoformat(timespec='seconds'))} &nbsp;|&nbsp; "
+                 f"scenarios: {len(records)}</div>")
+
+    # run-level metadata
+    if summary:
+        t = summary.get("totals", {})
+        parts.append("<div class='meta-grid'>")
+        for k, v in [
+            ("model_path", summary.get("model_path")),
+            ("policy_server", f"{summary.get('policy_server',{}).get('host')}:{summary.get('policy_server',{}).get('port')}"),
+            ("manifest", summary.get("manifest")),
+            ("server_command", summary.get("server_command")),
+            ("n_scenarios", summary.get("n_scenarios")),
+            ("n_normal / n_abnormal_probe", f"{summary.get('n_normal')} / {summary.get('n_abnormal_probe')}"),
+            ("rollout_started / actions_produced / video_saved / errors",
+             f"{t.get('rollout_started')} / {t.get('actions_produced')} / {t.get('video_saved')} / {t.get('errors')}"),
+            ("run timestamp", summary.get("timestamp")),
+            ("dry_run", summary.get("dry_run")),
+        ]:
+            parts.append(f"<div><b>{html.escape(str(k))}:</b> {html.escape(_fmt(v))}</div>")
+        parts.append("</div>")
+    if not _MPL_OK:
+        parts.append(f"<div class='warn'>matplotlib not available ({html.escape(str(_MPL_ERR))}) — plots were skipped.</div>")
+    if args.no_plots:
+        parts.append("<div class='warn'>--no-plots was set — no PNG plots were generated this run.</div>")
+
+    # comparison table
+    parts.append("<h2>Comparison across all scenarios</h2>")
+    cols = ["scenario_id", "label", "seed", "success", "num_policy_calls", "video_saved",
+            "mean_action_norm", "max_action_norm", "mean_delta_norm", "gripper_min", "gripper_max", "output_dir"]
+    parts.append("<table class='cmp'><thead><tr>" + "".join(f"<th>{html.escape(c)}</th>" for c in cols) + "</tr></thead><tbody>")
+    for row in table_rows:
+        parts.append("<tr>")
+        for c in cols:
+            v = row.get(c)
+            if c == "label":
+                cell = _badge_label(v)
+            elif c == "success":
+                cell = _badge_success(v)
+            elif c == "scenario_id":
+                cell = f"<a href='#{html.escape(str(v))}'>{html.escape(str(v))}</a>"
+            elif c in ("mean_action_norm", "max_action_norm", "mean_delta_norm", "gripper_min", "gripper_max"):
+                cell = "—" if v is None else f"{float(v):.4g}"
+            else:
+                cell = html.escape(_fmt(v))
+            parts.append(f"<td>{cell}</td>")
+        parts.append("</tr>")
+    parts.append("</tbody></table>")
+
+    # one card per scenario
+    for rec in records:
+        sid = rec["scenario_id"]
+        parts.append(f"<div class='card' id='{html.escape(sid)}'>")
+        parts.append(f"<h2>{html.escape(sid)}</h2>")
+        parts.append("<div class='cmeta'>" + _badge_label(rec.get("label")) + " " + _badge_success(rec.get("success"))
+                     + f" &nbsp; seed={html.escape(_fmt(rec.get('seed')))}"
+                     + (f" &nbsp; env=<code>{html.escape(str(rec.get('env_name')))}</code>" if rec.get("env_name") else "")
+                     + "</div>")
+        if rec.get("error"):
+            parts.append(f"<div class='err'>error: {html.escape(str(rec['error']))}</div>")
+        for w in rec.get("warnings", []):
+            parts.append(f"<div class='warn'>⚠ {html.escape(str(w))}</div>")
+
+        parts.append("<div class='cols'>")
+
+        # left column: key/values + video + file links
+        parts.append("<div>")
+        st = rec.get("stats") or {}
+        kv = [
+            ("scenario_id", sid),
+            ("label", rec.get("label")),
+            ("seed", rec.get("seed")),
+            ("instruction", rec.get("instruction")),
+            ("success", rec.get("success")),
+            ("max_episode_steps", rec.get("max_episode_steps")),
+            ("num policy calls", rec.get("num_policy_calls")),
+            ("actions.npy shape", rec.get("actions_shape")),
+            ("action horizon × DoF", f"{st.get('horizon')} × {st.get('n_dof')}" if st else "—"),
+            ("gripper DoF (col)", (f"{st['dof_names'][st['gripper_col']]} (col {st['gripper_col']})" if st else "—")),
+            ("mean / max ‖action‖", f"{_fmt(st.get('mean_action_norm'))} / {_fmt(st.get('max_action_norm'))}" if st else "—"),
+            ("mean ‖Δ chunk‖", _fmt(st.get("mean_delta_norm")) if st else "—"),
+            ("gripper min / max", f"{_fmt(st.get('gripper_min'))} / {_fmt(st.get('gripper_max'))}" if st else "—"),
+            ("obs_noise_std", rec.get("obs_noise_std")),
+            ("action_repeat", rec.get("action_repeat")),
+            ("elapsed_sec", rec.get("elapsed_sec")),
+            ("frames", (f"{rec.get('n_frames')} PNGs" if rec.get("n_frames") else "none")),
+        ]
+        parts.append("<div class='kv'>")
+        for k, v in kv:
+            parts.append(f"<div><b>{html.escape(str(k))}:</b> {html.escape(_fmt(v))}</div>")
+        parts.append("</div>")
+
+        # video preview
+        parts.append("<div style='margin-top:10px'>")
+        if rec.get("video_data_uri"):
+            parts.append(f"<video controls preload='metadata' src='{rec['video_data_uri']}'></video>")
+        elif rec.get("video_present"):
+            vrel = _rel(rec["video_path"], run_dir)
+            parts.append(f"<video controls preload='metadata' src='{html.escape(vrel)}'></video>"
+                         f"<div style='font-size:12px'><a href='{html.escape(vrel)}'>{html.escape(vrel)}</a></div>")
+        elif rec.get("first_frame_uri"):
+            parts.append(f"<img src='{rec['first_frame_uri']}' style='max-width:340px;border-radius:6px'/>"
+                         f"<div class='warn'>no video.mp4 — showing first frame only</div>")
+        else:
+            parts.append("<div class='warn'>no video / frames available to preview</div>")
+        parts.append("</div>")
+
+        # file links
+        parts.append("<div class='files' style='margin-top:10px'>")
+        for name, info in (rec.get("files") or {}).items():
+            if info["exists"]:
+                parts.append(f"<a href='{html.escape(info['rel'])}'>{html.escape(name)}</a>")
+            else:
+                parts.append(f"<span class='missing'>{html.escape(name)}</span>")
+        if rec.get("frames_dir"):
+            parts.append(f"<a href='{html.escape(_rel(rec['frames_dir'], run_dir))}/'>frames/ ({rec.get('n_frames')})</a>")
+        parts.append("</div>")
+        if rec.get("notes"):
+            parts.append(f"<div class='notes'>note: {html.escape(str(rec['notes']))}</div>")
+        parts.append("</div>")  # end left column
+
+        # right column: plots
+        parts.append("<div>")
+        if rec.get("plot_paths"):
+            parts.append("<div class='plots'>")
+            for logical in ("action_norm", "action_mean_per_dof", "gripper_over_time", "action_delta_norm"):
+                p = rec["plot_paths"].get(logical)
+                if p and Path(p).exists():
+                    parts.append(f"<a href='{html.escape(_rel(Path(p), run_dir))}'>"
+                                 f"<img src='{html.escape(_rel(Path(p), run_dir))}' alt='{logical}'/></a>")
+                else:
+                    parts.append(f"<div class='warn'>plot '{logical}.png' not available</div>")
+            parts.append("</div>")
+        elif rec.get("stats") is None:
+            parts.append("<div class='warn'>no actions.npy — no action statistics / plots</div>")
+        else:
+            parts.append("<div class='warn'>plots not generated (use without --no-plots, and ensure matplotlib is installed)</div>")
+        parts.append("</div>")  # end right column
+
+        parts.append("</div>")  # end cols
+        parts.append("</div>")  # end card
+
+    parts.append("</body></html>")
+    return "\n".join(parts)
+
+
+# --------------------------------------------------------------------------- #
+# main
+# --------------------------------------------------------------------------- #
+def _discover_scenarios(run_dir: Path, summary: dict | None) -> tuple[list[str], dict[str, dict]]:
+    rows_by_id: dict[str, dict] = {}
+    order: list[str] = []
+    if summary and isinstance(summary.get("scenarios"), list):
+        for r in summary["scenarios"]:
+            sid = r.get("scenario_id")
+            if sid:
+                rows_by_id[sid] = r
+                order.append(sid)
+    # add any scenario subdirs not already listed
+    skip = {"plots"}
+    for p in sorted(run_dir.iterdir()):
+        if p.is_dir() and p.name not in skip and p.name not in order:
+            order.append(p.name)
+    return order, rows_by_id
+
+
+def write_table_outputs(run_dir: Path, table_rows: list[dict]) -> tuple[Path, Path]:
+    cols = ["scenario_id", "label", "seed", "success", "num_policy_calls", "video_saved",
+            "mean_action_norm", "max_action_norm", "mean_delta_norm", "gripper_min", "gripper_max", "output_dir"]
+    csv_path = run_dir / "visual_summary.csv"
+    with open(csv_path, "w", newline="") as f:
+        w = csv.DictWriter(f, fieldnames=cols)
+        w.writeheader()
+        for row in table_rows:
+            w.writerow({c: ("" if row.get(c) is None else row.get(c)) for c in cols})
+    json_path = run_dir / "visual_summary.json"
+    with open(json_path, "w") as f:
+        json.dump({"run_dir": str(run_dir), "n_scenarios": len(table_rows),
+                   "generated": _dt.datetime.now().isoformat(timespec="seconds"),
+                   "columns": cols, "rows": table_rows}, f, indent=2)
+    return csv_path, json_path
+
+
+def main() -> int:
+    ap = argparse.ArgumentParser(description=__doc__, formatter_class=argparse.RawDescriptionHelpFormatter)
+    ap.add_argument("--run-dir", required=True, help="Path to the completed run dir (contains summary.json + scenario subdirs).")
+    ap.add_argument("--open", action="store_true", help="Open the generated HTML report in a browser.")
+    ap.add_argument("--no-plots", action="store_true", help="Do not (re)generate the PNG plots.")
+    ap.add_argument("--no-video-embed", action="store_true", help="Link videos by relative path instead of base64-embedding them in the HTML.")
+    args = ap.parse_args()
+
+    run_dir = Path(args.run_dir).resolve()
+    if not run_dir.is_dir():
+        print(f"ERROR: --run-dir does not exist or is not a directory: {run_dir}", file=sys.stderr)
+        return 2
+
+    summary, serr = _load_json(run_dir / "summary.json")
+    if serr:
+        print(f"WARNING: {serr} — proceeding by scanning scenario subdirectories.", file=sys.stderr)
+    # summary.md is read only to confirm it exists / surface it; we don't parse it.
+    summary_md_present = (run_dir / "summary.md").exists()
+    if not summary_md_present:
+        print("WARNING: summary.md not found (continuing).", file=sys.stderr)
+
+    order, rows_by_id = _discover_scenarios(run_dir, summary)
+    if not order:
+        print(f"ERROR: no scenarios found under {run_dir}", file=sys.stderr)
+        return 2
+    print(f"Found {len(order)} scenario(s) in {run_dir}")
+
+    make_plots_flag = not args.no_plots
+    embed_video = not args.no_video_embed
+
+    records: list[dict] = []
+    table_rows: list[dict] = []
+    all_warnings: list[str] = []
+    n_videos_embedded = 0
+    n_videos_present = 0
+    for sid in order:
+        try:
+            rec = process_scenario(sid, run_dir, rows_by_id.get(sid), make_plots_flag, embed_video)
+        except Exception as e:  # noqa: BLE001
+            tb = traceback.format_exc()
+            print(f"WARNING: scenario {sid} raised {e!r}; recording as failed card.", file=sys.stderr)
+            rec = {"scenario_id": sid, "warnings": [f"internal error: {e!r}", tb], "stats": None,
+                   "plot_paths": {}, "files": {}, "label": None, "seed": None, "success": None,
+                   "video_present": False, "video_data_uri": None, "n_frames": 0}
+        records.append(rec)
+        all_warnings += [f"[{sid}] {w}" for w in rec.get("warnings", [])]
+        if rec.get("video_present"):
+            n_videos_present += 1
+        if rec.get("video_data_uri"):
+            n_videos_embedded += 1
+        st = rec.get("stats") or {}
+        table_rows.append({
+            "scenario_id": sid,
+            "label": rec.get("label"),
+            "seed": rec.get("seed"),
+            "success": rec.get("success"),
+            "num_policy_calls": rec.get("num_policy_calls"),
+            "video_saved": bool(rec.get("video_present")),
+            "mean_action_norm": st.get("mean_action_norm"),
+            "max_action_norm": st.get("max_action_norm"),
+            "mean_delta_norm": st.get("mean_delta_norm"),
+            "gripper_min": st.get("gripper_min"),
+            "gripper_max": st.get("gripper_max"),
+            "output_dir": rec.get("output_dir") or str(run_dir / sid),
+        })
+
+    csv_path, json_path = write_table_outputs(run_dir, table_rows)
+    html_str = render_html(run_dir, summary, records, table_rows, args)
+    html_path = run_dir / "visual_report.html"
+    html_path.write_text(html_str, encoding="utf-8")
+
+    # console summary
+    print(f"\nWrote: {html_path}")
+    print(f"Wrote: {csv_path}")
+    print(f"Wrote: {json_path}")
+    if make_plots_flag and _MPL_OK:
+        print(f"Plots: {run_dir / 'plots'}/<scenario_id>/*.png")
+    print(f"Videos: {n_videos_present}/{len(order)} present; "
+          f"{'all ' if n_videos_embedded == len(order) and embed_video else ''}"
+          f"{n_videos_embedded}/{len(order)} embedded in HTML"
+          + ("" if embed_video else "  (--no-video-embed: linked by path instead)"))
+    if all_warnings:
+        print(f"\n{len(all_warnings)} warning(s):")
+        for w in all_warnings:
+            print(f"  - {w}")
+    else:
+        print("\nNo missing files or warnings.")
+
+    if args.open:
+        import webbrowser
+
+        url = html_path.as_uri()
+        opened = False
+        try:
+            opened = webbrowser.open(url)
+        except Exception:
+            opened = False
+        if opened:
+            print(f"\nOpened {url} in a browser.")
+        else:
+            print(f"\nCould not auto-open a browser. Open this file manually:\n  {html_path}")
+
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

outputs/_setup_logs/_dryrun.log

@@ -0,0 +1,57 @@
+Loaded 10 scenarios from /home/ubuntu/VLA_SAE/examples/LIBERO/smoke_tests/scenarios_10.yaml
+  labels: 8 normal, 2 abnormal_probe
+
+Expected GR00T server command (run this in a separate terminal):
+  uv run python gr00t/eval/run_gr00t_server.py --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 --embodiment-tag LIBERO_PANDA --use-sim-policy-wrapper
+
+NOTE: The GR00T-N1.7 backbone 'nvidia/Cosmos-Reason2-2B' is a GATED HuggingFace repo.
+To start the server you must:
+  1. Request access at https://huggingface.co/nvidia/Cosmos-Reason2-2B (one click, usually instant).
+  2. Authenticate, e.g.  export HF_TOKEN=hf_xxx   (or: uv run hf auth login)
+  3. Re-run the server / smoke tests.
+
+Model checkpoint OK: checkpoints/GR00T-N1.7-LIBERO/libero_10
+Run directory: outputs/libero_smoke_tests/20260512_120217
+(dry-run) skipping server reachability requirement.
+LIBERO rollout interpreter: /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/python
+
+[1/10] normal_kitchen3_moka_pot_on_stove  (normal)  env=libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it  seed=1000  max_episode_steps=50
+
+[2/10] normal_kitchen4_bowl_in_drawer  (normal)  env=libero_sim/KITCHEN_SCENE4_put_the_black_bowl_in_the_bottom_drawer_of_the_cabinet_and_close_it  seed=1001  max_episode_steps=50
+
+[3/10] normal_living1_soup_and_cheese_in_basket  (normal)  env=libero_sim/LIVING_ROOM_SCENE1_put_both_the_alphabet_soup_and_the_cream_cheese_box_in_the_basket  seed=1002  max_episode_steps=50
+
+[4/10] normal_living2_soup_and_tomato_in_basket  (normal)  env=libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket  seed=1003  max_episode_steps=50
+
+[5/10] normal_study1_book_in_caddy  (normal)  env=libero_sim/STUDY_SCENE1_pick_up_the_book_and_place_it_in_the_back_compartment_of_the_caddy  seed=1004  max_episode_steps=50
+
+[6/10] normal_kitchen8_both_moka_pots_on_stove  (normal)  env=libero_sim/KITCHEN_SCENE8_put_both_moka_pots_on_the_stove  seed=1005  max_episode_steps=50
+
+[7/10] normal_kitchen6_mug_in_microwave  (normal)  env=libero_sim/KITCHEN_SCENE6_put_the_yellow_and_white_mug_in_the_microwave_and_close_it  seed=1006  max_episode_steps=50
+
+[8/10] normal_living5_two_mugs_on_plates  (normal)  env=libero_sim/LIVING_ROOM_SCENE5_put_the_white_mug_on_the_left_plate_and_put_the_yellow_and_white_mug_on_the_right_plate  seed=1007  max_episode_steps=50
+
+[9/10] abnormal_probe_obs_noise  (abnormal_probe)  env=libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it  seed=2000  max_episode_steps=50
+
+[10/10] abnormal_probe_short_timeout  (abnormal_probe)  env=libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket  seed=2001  max_episode_steps=16
+
+======== SMOKE-TEST SUMMARY ========
+  scenario_id                              | label          | seed | rollout_started | actions_produced | video_saved | success | output_dir                                                                          | error_if_any                 
+  -----------------------------------------+----------------+------+-----------------+------------------+-------------+---------+-------------------------------------------------------------------------------------+------------------------------
+  normal_kitchen3_moka_pot_on_stove        | normal         | 1000 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_kitchen3_moka_pot_on_stove        | dry-run (no rollout executed)
+  normal_kitchen4_bowl_in_drawer           | normal         | 1001 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_kitchen4_bowl_in_drawer           | dry-run (no rollout executed)
+  normal_living1_soup_and_cheese_in_basket | normal         | 1002 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_living1_soup_and_cheese_in_basket | dry-run (no rollout executed)
+  normal_living2_soup_and_tomato_in_basket | normal         | 1003 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_living2_soup_and_tomato_in_basket | dry-run (no rollout executed)
+  normal_study1_book_in_caddy              | normal         | 1004 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_study1_book_in_caddy              | dry-run (no rollout executed)
+  normal_kitchen8_both_moka_pots_on_stove  | normal         | 1005 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_kitchen8_both_moka_pots_on_stove  | dry-run (no rollout executed)
+  normal_kitchen6_mug_in_microwave         | normal         | 1006 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_kitchen6_mug_in_microwave         | dry-run (no rollout executed)
+  normal_living5_two_mugs_on_plates        | normal         | 1007 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/normal_living5_two_mugs_on_plates        | dry-run (no rollout executed)
+  abnormal_probe_obs_noise                 | abnormal_probe | 2000 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/abnormal_probe_obs_noise                 | dry-run (no rollout executed)
+  abnormal_probe_short_timeout             | abnormal_probe | 2001 | no              | no               | no          | unknown | outputs/libero_smoke_tests/20260512_120217/abnormal_probe_short_timeout             | dry-run (no rollout executed)
+
+Wrote: outputs/libero_smoke_tests/20260512_120217/summary.json
+Wrote: outputs/libero_smoke_tests/20260512_120217/summary.md
+
+Review the videos with:
+  python /home/ubuntu/VLA_SAE/examples/LIBERO/smoke_tests/review_smoke_tests.py --run-dir outputs/libero_smoke_tests/20260512_120217
+

outputs/_setup_logs/_hf_download.log

@@ -0,0 +1,4 @@
+/home/ubuntu/.local/lib/python3.10/site-packages/huggingface_hub/cli/download.py:146: UserWarning: Ignoring `--include` since filenames have being explicitly set.
+  warnings.warn("Ignoring `--include` since filenames have being explicitly set.")
+Warning: You are sending unauthenticated requests to the HF Hub. Please set a HF_TOKEN to enable higher rate limits and faster downloads.
+path=/home/ubuntu/VLA_SAE/checkpoints/GR00T-N1.7-LIBERO

outputs/_setup_logs/_libero_setup.log

@@ -0,0 +1,439 @@
++++ dirname gr00t/eval/sim/LIBERO/setup_libero.sh
+++ cd gr00t/eval/sim/LIBERO
+++ pwd
++ SCRIPT_DIR=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO
++ LIBERO_REPO=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../../external_dependencies/LIBERO
++ PROJECT_REPO=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../..
++ LIBERO_UV_ENV=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv
++ git submodule update --init /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../../external_dependencies/LIBERO
++ rm -rf /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv
++ mkdir -p /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv
++ uv venv /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv --python 3.10
+Using CPython 3.10.12 interpreter at: /usr/bin/python3.10
+Creating virtual environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Activate with: source gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/activate
++ source /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/activate
+++ '[' -z '' ']'
+++ '[' -n x ']'
+++ SCRIPT_PATH=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/activate
+++ '[' /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/activate = gr00t/eval/sim/LIBERO/setup_libero.sh ']'
+++ deactivate nondestructive
+++ unset -f pydoc
+++ '[' -z '' ']'
+++ '[' -z '' ']'
+++ hash -r
+++ '[' -z '' ']'
+++ unset VIRTUAL_ENV
+++ unset VIRTUAL_ENV_PROMPT
+++ '[' '!' nondestructive = nondestructive ']'
+++ VIRTUAL_ENV=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv
+++ '[' linux-gnu = cygwin ']'
+++ '[' linux-gnu = msys ']'
+++ export VIRTUAL_ENV
+++ '[' -z '' ']'
+++ unset SCRIPT_PATH
+++ _OLD_VIRTUAL_PATH=/home/ubuntu/.local/bin:/home/ubuntu/.vscode-server/cli/servers/Stable-034f571df509819cc10b0c8129f66ef77a542f0e/server/bin/remote-cli:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
+++ PATH=/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin:/home/ubuntu/.local/bin:/home/ubuntu/.vscode-server/cli/servers/Stable-034f571df509819cc10b0c8129f66ef77a542f0e/server/bin/remote-cli:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games:/snap/bin
+++ export PATH
+++ '[' x '!=' x ']'
++++ basename /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv
+++ VIRTUAL_ENV_PROMPT=.venv
+++ export VIRTUAL_ENV_PROMPT
+++ '[' -z '' ']'
+++ '[' -z '' ']'
+++ _OLD_VIRTUAL_PS1=
+++ PS1='(.venv) '
+++ export PS1
+++ alias pydoc
+++ true
+++ hash -r
++ uv pip install --requirements /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../../external_dependencies/LIBERO/requirements.txt
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 120 packages in 34ms
+   Building gym==0.25.2
+   Building bddl==1.0.1
+   Building promise==2.3
+   Building future==0.18.2
+   Building egl-probe==1.0.2
+   Building robomimic==0.2.0
+   Building pathtools==0.1.2
+   Building easydict==1.9
+Downloading wandb (1.7MiB)
+Downloading pyopengl (3.0MiB)
+Downloading transformers (4.4MiB)
+Downloading nvidia-cuda-runtime (2.1MiB)
+Downloading cuda-bindings (6.0MiB)
+Downloading tokenizers (6.3MiB)
+Downloading nvidia-cuda-nvrtc (86.0MiB)
+Downloading nvidia-cuda-cupti (10.2MiB)
+Downloading h5py (4.8MiB)
+Downloading imageio-ffmpeg (28.1MiB)
+Downloading triton (179.4MiB)
+Downloading nvidia-cusparse (139.2MiB)
+Downloading numba (3.6MiB)
+Downloading torch (506.0MiB)
+Downloading nvidia-cublas (403.5MiB)
+Downloading numpy (16.0MiB)
+Downloading torchvision (7.2MiB)
+Downloading llvmlite (53.7MiB)
+Downloading nvidia-cufft (204.2MiB)
+Downloading scipy (36.8MiB)
+Downloading setuptools (1.0MiB)
+Downloading nvidia-cusolver (191.6MiB)
+Downloading opencv-python (58.1MiB)
+Downloading mujoco (6.9MiB)
+Downloading tensorboard-data-server (6.3MiB)
+Downloading nvidia-cudnn-cu13 (349.1MiB)
+Downloading nvidia-curand (56.8MiB)
+Downloading nvidia-cusparselt-cu13 (162.0MiB)
+Downloading nvidia-nvjitlink (38.8MiB)
+Downloading nvidia-nccl-cu13 (187.4MiB)
+Downloading robosuite (184.5MiB)
+Downloading matplotlib (11.3MiB)
+Downloading grpcio (6.5MiB)
+Downloading tensorboard (5.3MiB)
+Downloading nvidia-nvshmem-cu13 (57.6MiB)
+      Built easydict==1.9
+      Built pathtools==0.1.2
+      Built promise==2.3
+      Built robomimic==0.2.0
+      Built bddl==1.0.1
+      Built gym==0.25.2
+      Built future==0.18.2
+ Downloaded nvidia-cuda-runtime
+ Downloaded setuptools
+ Downloaded wandb
+      Built egl-probe==1.0.2
+ Downloaded h5py
+ Downloaded numba
+ Downloaded transformers
+ Downloaded tensorboard
+ Downloaded cuda-bindings
+ Downloaded tensorboard-data-server
+ Downloaded tokenizers
+ Downloaded grpcio
+ Downloaded mujoco
+ Downloaded torchvision
+ Downloaded nvidia-cuda-cupti
+ Downloaded matplotlib
+ Downloaded numpy
+ Downloaded imageio-ffmpeg
+ Downloaded nvidia-nvjitlink
+ Downloaded scipy
+ Downloaded pyopengl
+ Downloaded llvmlite
+ Downloaded nvidia-curand
+ Downloaded nvidia-nvshmem-cu13
+ Downloaded opencv-python
+ Downloaded nvidia-cuda-nvrtc
+ Downloaded nvidia-cusparse
+ Downloaded nvidia-cusparselt-cu13
+ Downloaded triton
+ Downloaded nvidia-nccl-cu13
+ Downloaded nvidia-cusolver
+ Downloaded nvidia-cufft
+ Downloaded nvidia-cudnn-cu13
+ Downloaded nvidia-cublas
+ Downloaded robosuite
+ Downloaded torch
+Prepared 43 packages in 17.97s
+Installed 120 packages in 192ms
+ + absl-py==2.4.0
+ + antlr4-python3-runtime==4.9.3
+ + attrs==26.1.0
+ + bddl==1.0.1
+ + certifi==2026.4.22
+ + charset-normalizer==3.4.7
+ + click==8.3.3
+ + cloudpickle==2.1.0
+ + cuda-bindings==13.2.0
+ + cuda-pathfinder==1.5.4
+ + cuda-toolkit==13.0.2
+ + cycler==0.12.1
+ + docker-pycreds==0.4.0
+ + easydict==1.9
+ + egl-probe==1.0.2
+ + einops==0.4.1
+ + etils==1.13.0
+ + exceptiongroup==1.3.1
+ + fastjsonschema==2.21.2
+ + filelock==3.29.0
+ + fonttools==4.62.1
+ + fsspec==2026.4.0
+ + future==0.18.2
+ + gitdb==4.0.12
+ + gitpython==3.1.50
+ + glfw==2.10.0
+ + grpcio==1.80.0
+ + gym==0.25.2
+ + gym-notices==0.1.0
+ + h5py==3.16.0
+ + hf-xet==1.5.0
+ + huggingface-hub==0.36.2
+ + hydra-core==1.2.0
+ + idna==3.14
+ + imageio==2.37.3
+ + imageio-ffmpeg==0.6.0
+ + importlib-resources==7.1.0
+ + iniconfig==2.3.0
+ + jinja2==3.1.6
+ + jsonschema==4.26.0
+ + jsonschema-specifications==2025.9.1
+ + jupyter-core==5.9.1
+ + jupytext==1.19.2
+ + kiwisolver==1.5.0
+ + llvmlite==0.47.0
+ + markdown==3.10.2
+ + markdown-it-py==4.2.0
+ + markupsafe==3.0.3
+ + matplotlib==3.5.3
+ + mdit-py-plugins==0.6.0
+ + mdurl==0.1.2
+ + mpmath==1.3.0
+ + mujoco==3.8.1
+ + nbformat==5.10.4
+ + networkx==3.4.2
+ + numba==0.65.1
+ + numpy==1.22.4
+ + nvidia-cublas==13.1.0.3
+ + nvidia-cuda-cupti==13.0.85
+ + nvidia-cuda-nvrtc==13.0.88
+ + nvidia-cuda-runtime==13.0.96
+ + nvidia-cudnn-cu13==9.19.0.56
+ + nvidia-cufft==12.0.0.61
+ + nvidia-cufile==1.15.1.6
+ + nvidia-curand==10.4.0.35
+ + nvidia-cusolver==12.0.4.66
+ + nvidia-cusparse==12.6.3.3
+ + nvidia-cusparselt-cu13==0.8.0
+ + nvidia-nccl-cu13==2.28.9
+ + nvidia-nvjitlink==13.0.88
+ + nvidia-nvshmem-cu13==3.4.5
+ + nvidia-nvtx==13.0.85
+ + omegaconf==2.3.0
+ + opencv-python==4.6.0.66
+ + packaging==26.2
+ + pathtools==0.1.2
+ + pillow==12.2.0
+ + platformdirs==4.9.6
+ + pluggy==1.6.0
+ + promise==2.3
+ + protobuf==3.20.3
+ + psutil==7.2.2
+ + pygments==2.20.0
+ + pyopengl==3.1.10
+ + pyparsing==3.3.2
+ + pytest==9.0.3
+ + python-dateutil==2.9.0.post0
+ + pyyaml==6.0.3
+ + referencing==0.37.0
+ + regex==2026.5.9
+ + requests==2.34.0
+ + robomimic==0.2.0
+ + robosuite==1.4.0
+ + rpds-py==0.30.0
+ + scipy==1.13.1
+ + sentry-sdk==2.59.0
+ + setproctitle==1.3.7
+ + setuptools==81.0.0
+ + shortuuid==1.0.13
+ + six==1.17.0
+ + smmap==5.0.3
+ + sympy==1.14.0
+ + tensorboard==2.20.0
+ + tensorboard-data-server==0.7.2
+ + tensorboardx==2.6.5
+ + termcolor==3.3.0
+ + thop==0.1.1.post2209072238
+ + tokenizers==0.12.1
+ + tomli==2.4.1
+ + torch==2.11.0
+ + torchvision==0.26.0
+ + tqdm==4.67.3
+ + traitlets==5.15.0
+ + transformers==4.21.1
+ + triton==3.6.0
+ + typing-extensions==4.15.0
+ + urllib3==2.7.0
+ + wandb==0.13.1
+ + werkzeug==3.1.8
+ + zipp==3.23.1
++ uv pip install -e /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../../external_dependencies/LIBERO --config-settings editable_mode=compat
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 1 package in 440ms
+   Building libero @ file:///home/ubuntu/VLA_SAE/external_dependencies/LIBERO
+      Built libero @ file:///home/ubuntu/VLA_SAE/external_dependencies/LIBERO
+Prepared 1 package in 211ms
+Installed 1 package in 1ms
+ + libero==0.1.0 (from file:///home/ubuntu/VLA_SAE/external_dependencies/LIBERO)
++ uv pip install --editable /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../.. --no-deps
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 1 package in 1ms
+   Building gr00t @ file:///home/ubuntu/VLA_SAE
+      Built gr00t @ file:///home/ubuntu/VLA_SAE
+Prepared 1 package in 569ms
+Installed 1 package in 0.62ms
+ + gr00t==0.1.0 (from file:///home/ubuntu/VLA_SAE)
++ uv pip install torch==2.5.1 torchvision==0.20.1 pydantic av tianshou==0.5.1 tyro pandas dm_tree einops==0.8.1 albumentations==1.4.18 zmq
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 72 packages in 543ms
+   Building zmq==0.0.0
+Downloading pandas (12.2MiB)
+Downloading nvidia-cuda-cupti-cu12 (13.2MiB)
+Downloading opencv-python-headless (57.6MiB)
+Downloading sympy (5.9MiB)
+Downloading av (35.9MiB)
+Downloading numpy (16.0MiB)
+Downloading torchvision (6.9MiB)
+Downloading nvidia-cufft-cu12 (201.7MiB)
+Downloading nvidia-cublas-cu12 (346.6MiB)
+Downloading nvidia-cusparse-cu12 (197.8MiB)
+Downloading nvidia-cuda-nvrtc-cu12 (23.5MiB)
+Downloading nvidia-cusolver-cu12 (122.0MiB)
+Downloading nvidia-nvjitlink-cu12 (20.1MiB)
+Downloading nvidia-nccl-cu12 (179.9MiB)
+Downloading nvidia-curand-cu12 (53.7MiB)
+Downloading nvidia-cudnn-cu12 (634.0MiB)
+Downloading torch (864.5MiB)
+Downloading triton (199.8MiB)
+      Built zmq==0.0.0
+ Downloaded torchvision
+ Downloaded nvidia-cuda-cupti-cu12
+ Downloaded numpy
+ Downloaded nvidia-nvjitlink-cu12
+ Downloaded nvidia-cuda-nvrtc-cu12
+ Downloaded sympy
+ Downloaded av
+ Downloaded nvidia-curand-cu12
+ Downloaded opencv-python-headless
+ Downloaded nvidia-cusolver-cu12
+ Downloaded pandas
+ Downloaded nvidia-nccl-cu12
+ Downloaded triton
+ Downloaded nvidia-cufft-cu12
+ Downloaded nvidia-cusparse-cu12
+ Downloaded nvidia-cublas-cu12
+ Downloaded nvidia-cudnn-cu12
+ Downloaded torch
+Prepared 26 packages in 25.57s
+Uninstalled 6 packages in 141ms
+Installed 44 packages in 266ms
+ + albucore==0.0.17
+ + albumentations==1.4.18
+ + annotated-types==0.7.0
+ + av==17.0.1
+ + dm-tree==0.1.10
+ + docstring-parser==0.18.0
+ - einops==0.4.1
+ + einops==0.8.1
+ + eval-type-backport==0.3.1
+ + farama-notifications==0.0.6
+ + gymnasium==1.3.0
+ + lazy-loader==0.5
+ - numpy==1.22.4
+ + numpy==2.2.6
+ + nvidia-cublas-cu12==12.4.5.8
+ + nvidia-cuda-cupti-cu12==12.4.127
+ + nvidia-cuda-nvrtc-cu12==12.4.127
+ + nvidia-cuda-runtime-cu12==12.4.127
+ + nvidia-cudnn-cu12==9.1.0.70
+ + nvidia-cufft-cu12==11.2.1.3
+ + nvidia-curand-cu12==10.3.5.147
+ + nvidia-cusolver-cu12==11.6.1.9
+ + nvidia-cusparse-cu12==12.3.1.170
+ + nvidia-nccl-cu12==2.21.5
+ + nvidia-nvjitlink-cu12==12.4.127
+ + nvidia-nvtx-cu12==12.4.127
+ + opencv-python-headless==4.13.0.92
+ + pandas==2.3.3
+ + pettingzoo==1.26.1
+ + pydantic==2.13.4
+ + pydantic-core==2.46.4
+ + pytz==2026.2
+ + pyzmq==27.1.0
+ + scikit-image==0.25.2
+ - sympy==1.14.0
+ + sympy==1.13.1
+ + tianshou==0.5.1
+ + tifffile==2025.5.10
+ - torch==2.11.0
+ + torch==2.5.1
+ - torchvision==0.26.0
+ + torchvision==0.20.1
+ - triton==3.6.0
+ + triton==3.1.0
+ + typeguard==4.5.1
+ + typing-inspection==0.4.2
+ + tyro==1.0.13
+ + tzdata==2026.2
+ + wrapt==2.1.2
+ + zmq==0.0.0
++ uv pip install transformers==4.57.3 msgpack==1.1.0 msgpack-numpy==0.4.8 gymnasium==0.29.1
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 23 packages in 109ms
+Prepared 1 package in 95ms
+Uninstalled 3 packages in 19ms
+Installed 6 packages in 40ms
+ - gymnasium==1.3.0
+ + gymnasium==0.29.1
+ + msgpack==1.1.0
+ + msgpack-numpy==0.4.8
+ + safetensors==0.7.0
+ - tokenizers==0.12.1
+ + tokenizers==0.22.2
+ - transformers==4.21.1
+ + transformers==4.57.3
++ uv pip install numpy==1.26.4
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 1 package in 6ms
+Uninstalled 1 package in 11ms
+Installed 1 package in 16ms
+ - numpy==2.2.6
+ + numpy==1.26.4
++ uv pip install --editable /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/../../../.. --no-deps
+Using Python 3.10.12 environment at: gr00t/eval/sim/LIBERO/libero_uv/.venv
+Resolved 1 package in 1ms
+   Building gr00t @ file:///home/ubuntu/VLA_SAE
+      Built gr00t @ file:///home/ubuntu/VLA_SAE
+Prepared 1 package in 682ms
+Uninstalled 1 package in 0.33ms
+Installed 1 package in 0.55ms
+ ~ gr00t==0.1.0 (from file:///home/ubuntu/VLA_SAE)
++ rm -rf /home/ubuntu/.libero
++ printf 'n\n'
++ python -c 'from gr00t.eval.sim.LIBERO.libero_env import register_libero_envs'
+Do you want to specify a custom path for the dataset folder? (Y/N): [robosuite WARNING] No private macro file found! (__init__.py:7)
+[robosuite WARNING] It is recommended to use a private macro file (__init__.py:8)
+[robosuite WARNING] To setup, run: python /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/robosuite/scripts/setup_macros.py (__init__.py:9)
+Matplotlib created a temporary config/cache directory at /tmp/matplotlib-igojt7pq because the default path (/home/ubuntu/.config/matplotlib) is not a writable directory; it is highly recommended to set the MPLCONFIGDIR environment variable to a writable directory, in particular to speed up the import of Matplotlib and to better support multiprocessing.
+Gym has been unmaintained since 2022 and does not support NumPy 2.0 amongst other critical functionality.
+Please upgrade to Gymnasium, the maintained drop-in replacement of Gym, or contact the authors of your software and request that they upgrade.
+See the migration guide at https://gymnasium.farama.org/introduction/migration_guide/ for additional information.
+Initializing the default config file...
+The following information is stored in the config file: /home/ubuntu/.libero/config.yaml
+benchmark_root: /home/ubuntu/VLA_SAE/external_dependencies/LIBERO/libero/libero
+bddl_files: /home/ubuntu/VLA_SAE/external_dependencies/LIBERO/libero/libero/./bddl_files
+init_states: /home/ubuntu/VLA_SAE/external_dependencies/LIBERO/libero/libero/./init_files
+datasets: /home/ubuntu/VLA_SAE/external_dependencies/LIBERO/libero/libero/../datasets
+assets: /home/ubuntu/VLA_SAE/external_dependencies/LIBERO/libero/libero/./assets
++ python -
+[robosuite WARNING] No private macro file found! (__init__.py:7)
+[robosuite WARNING] It is recommended to use a private macro file (__init__.py:8)
+[robosuite WARNING] To setup, run: python /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/robosuite/scripts/setup_macros.py (__init__.py:9)
+Matplotlib created a temporary config/cache directory at /tmp/matplotlib-fwatww__ because the default path (/home/ubuntu/.config/matplotlib) is not a writable directory; it is highly recommended to set the MPLCONFIGDIR environment variable to a writable directory, in particular to speed up the import of Matplotlib and to better support multiprocessing.
+Gym has been unmaintained since 2022 and does not support NumPy 2.0 amongst other critical functionality.
+Please upgrade to Gymnasium, the maintained drop-in replacement of Gym, or contact the authors of your software and request that they upgrade.
+See the migration guide at https://gymnasium.farama.org/introduction/migration_guide/ for additional information.
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:159: UserWarning: WARN: The obs returned by the `reset()` method is not within the observation space.
+  logger.warn(f"{pre} is not within the observation space.")
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:135: UserWarning: WARN: The obs returned by the `reset()` method was expecting numpy array dtype to be float32, actual type: float64
+  logger.warn(
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:131: UserWarning: WARN: The obs returned by the `reset()` method was expecting a numpy array, actual type: <class 'list'>
+  logger.warn(
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/spaces/box.py:240: UserWarning: WARN: Casting input x to numpy array.
+  gym.logger.warn("Casting input x to numpy array.")
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+Env OK: <class 'gymnasium.wrappers.order_enforcing.OrderEnforcing'>

outputs/_setup_logs/_server.log

@@ -0,0 +1,85 @@
+/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
+  check_for_updates()
+flash_attn is not installed. Falling back to sdpa attention. Install flash-attn for better performance: pip install flash-attn
+Starting GR00T inference server...
+  Embodiment tag: EmbodimentTag.LIBERO_PANDA
+  Model path: checkpoints/GR00T-N1.7-LIBERO/libero_10
+  Device: cuda
+  Host: 127.0.0.1
+  Port: 5555
+Traceback (most recent call last):
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", line 403, in hf_raise_for_status
+    response.raise_for_status()
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/requests/models.py", line 1167, in raise_for_status
+    raise HTTPError(http_error_msg, response=self)
+requests.exceptions.HTTPError: 401 Client Error: Unauthorized for url: https://huggingface.co/nvidia/Cosmos-Reason2-2B/resolve/main/config.json
+
+The above exception was the direct cause of the following exception:
+
+Traceback (most recent call last):
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/utils/hub.py", line 479, in cached_files
+    hf_hub_download(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
+    return fn(*args, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 1014, in hf_hub_download
+    return _hf_hub_download_to_cache_dir(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 1121, in _hf_hub_download_to_cache_dir
+    _raise_on_head_call_error(head_call_error, force_download, local_files_only)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 1662, in _raise_on_head_call_error
+    raise head_call_error
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 1550, in _get_metadata_or_catch_error
+    metadata = get_hf_file_metadata(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/utils/_validators.py", line 114, in _inner_fn
+    return fn(*args, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 1467, in get_hf_file_metadata
+    r = _request_wrapper(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 283, in _request_wrapper
+    response = _request_wrapper(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/file_download.py", line 307, in _request_wrapper
+    hf_raise_for_status(response)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/huggingface_hub/utils/_http.py", line 420, in hf_raise_for_status
+    raise _format(GatedRepoError, message, response) from e
+huggingface_hub.errors.GatedRepoError: 401 Client Error. (Request ID: Root=1-6a0314fd-4cda431e47f56d0801840f67;ad3e8e30-c9ce-490a-ba30-5986386d4d7c)
+
+Cannot access gated repo for url https://huggingface.co/nvidia/Cosmos-Reason2-2B/resolve/main/config.json.
+Access to model nvidia/Cosmos-Reason2-2B is restricted. You must have access to it and be authenticated to access it. Please log in.
+
+The above exception was the direct cause of the following exception:
+
+Traceback (most recent call last):
+  File "/home/ubuntu/VLA_SAE/gr00t/eval/run_gr00t_server.py", line 167, in <module>
+    main(config)
+  File "/home/ubuntu/VLA_SAE/gr00t/eval/run_gr00t_server.py", line 86, in main
+    policy = Gr00tPolicy(
+  File "/home/ubuntu/VLA_SAE/gr00t/policy/gr00t_policy.py", line 100, in __init__
+    model = AutoModel.from_pretrained(model_dir)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/models/auto/auto_factory.py", line 604, in from_pretrained
+    return model_class.from_pretrained(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 277, in _wrapper
+    return func(*args, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 4971, in from_pretrained
+    model = cls(config, *model_args, **model_kwargs)
+  File "/home/ubuntu/VLA_SAE/gr00t/model/gr00t_n1d7/gr00t_n1d7.py", line 516, in __init__
+    self.backbone = backbone_cls(
+  File "/home/ubuntu/VLA_SAE/gr00t/model/modules/qwen3_backbone.py", line 80, in __init__
+    self.model = Qwen3VLForConditionalGeneration.from_pretrained(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 277, in _wrapper
+    return func(*args, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/modeling_utils.py", line 4843, in from_pretrained
+    config, model_kwargs = cls.config_class.from_pretrained(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/configuration_utils.py", line 622, in from_pretrained
+    config_dict, kwargs = cls.get_config_dict(pretrained_model_name_or_path, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/configuration_utils.py", line 662, in get_config_dict
+    config_dict, kwargs = cls._get_config_dict(pretrained_model_name_or_path, **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/configuration_utils.py", line 721, in _get_config_dict
+    resolved_config_file = cached_file(
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/utils/hub.py", line 322, in cached_file
+    file = cached_files(path_or_repo_id=path_or_repo_id, filenames=[filename], **kwargs)
+  File "/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/transformers/utils/hub.py", line 543, in cached_files
+    raise OSError(
+OSError: You are trying to access a gated repo.
+Make sure to have access to it at https://huggingface.co/nvidia/Cosmos-Reason2-2B.
+401 Client Error. (Request ID: Root=1-6a0314fd-4cda431e47f56d0801840f67;ad3e8e30-c9ce-490a-ba30-5986386d4d7c)
+
+Cannot access gated repo for url https://huggingface.co/nvidia/Cosmos-Reason2-2B/resolve/main/config.json.
+Access to model nvidia/Cosmos-Reason2-2B is restricted. You must have access to it and be authenticated to access it. Please log in.

outputs/_setup_logs/_uv_sync.log

@@ -0,0 +1,6 @@
+Using CPython 3.10.12 interpreter at: /usr/bin/python3
+Creating virtual environment at: .venv
+error: Failed to generate package metadata for `flash-attn==2.7.4.post1 @ path+scripts/deployment/dgpu/wheels/flash_attn-2.7.4.post1-cp310-cp310-linux_aarch64.whl`
+  Caused by: Failed to extract archive: flash_attn-2.7.4.post1-cp310-cp310-linux_aarch64.whl
+  Caused by: Invalid zip file structure
+  Caused by: Encountered an unexpected header (actual: 0x73726576, expected: 0x4034b50).

outputs/_setup_logs/_venv_install.log

@@ -0,0 +1,218 @@
++ uv pip install --python .venv/bin/python -e . --no-deps
+Resolved 1 package in 1ms
+   Building gr00t @ file:///home/ubuntu/VLA_SAE
+      Built gr00t @ file:///home/ubuntu/VLA_SAE
+Prepared 1 package in 848ms
+Installed 1 package in 0.80ms
+ + gr00t==0.1.0 (from file:///home/ubuntu/VLA_SAE)
++ uv pip install --python .venv/bin/python torch==2.7.1 torchvision==0.22.1 transformers==4.57.3 numpy==1.26.4 albumentations==1.4.18 av==16.1.0 diffusers==0.35.1 dm-tree lmdb==1.7.5 msgpack==1.1.0 msgpack-numpy==0.4.8 pandas==2.2.3 peft==0.17.1 termcolor==3.2.0 tyro==0.9.17 click==8.1.8 datasets==3.6.0 cryptography einops==0.8.1 gitpython==3.1.46 jsonlines==4.0.0 gymnasium==1.2.2 matplotlib==3.10.1 omegaconf==2.3.0 scipy==1.15.3 torchcodec==0.4.0 wandb==0.23.0 pyzmq==27.0.1 'huggingface-hub[cli]' 'opencv-python-headless>=4.5,<4.13' safetensors accelerate sentencepiece protobuf pyyaml tqdm
+Resolved 127 packages in 1.16s
+Downloading aiohttp (1.6MiB)
+Downloading nvidia-cufile-cu12 (1.1MiB)
+Downloading sympy (6.0MiB)
+Downloading hf-xet (4.3MiB)
+Downloading nvidia-curand-cu12 (53.7MiB)
+Downloading scipy (35.9MiB)
+Downloading pyarrow (46.6MiB)
+Downloading scikit-image (14.1MiB)
+Downloading cryptography (4.5MiB)
+Downloading nvidia-nvjitlink-cu12 (18.8MiB)
+Downloading numpy (17.4MiB)
+Downloading diffusers (3.9MiB)
+Downloading tokenizers (3.1MiB)
+Downloading kiwisolver (1.6MiB)
+Downloading opencv-python-headless (47.7MiB)
+Downloading transformers (11.4MiB)
+Downloading nvidia-cusolver-cu12 (150.9MiB)
+Downloading nvidia-cusparse-cu12 (206.5MiB)
+Downloading pandas (12.5MiB)
+Downloading sentencepiece (1.3MiB)
+Downloading nvidia-cuda-cupti-cu12 (8.5MiB)
+Downloading pillow (6.7MiB)
+Downloading nvidia-cufft-cu12 (190.9MiB)
+Downloading networkx (1.6MiB)
+Downloading nvidia-cuda-nvrtc-cu12 (22.6MiB)
+Downloading pydantic-core (2.0MiB)
+Downloading fonttools (4.7MiB)
+Downloading av (38.4MiB)
+Downloading nvidia-cudnn-cu12 (544.5MiB)
+Downloading nvidia-nccl-cu12 (192.0MiB)
+Downloading nvidia-cublas-cu12 (374.9MiB)
+Downloading nvidia-cusparselt-cu12 (149.5MiB)
+Downloading pygments (1.2MiB)
+Downloading torchcodec (1.3MiB)
+Downloading wandb (19.3MiB)
+Downloading matplotlib (8.2MiB)
+Downloading torchvision (7.1MiB)
+Downloading triton (148.4MiB)
+Downloading torch (783.1MiB)
+ Downloaded nvidia-cufile-cu12
+ Downloaded sentencepiece
+ Downloaded kiwisolver
+ Downloaded aiohttp
+ Downloaded torchcodec
+ Downloaded pygments
+ Downloaded pydantic-core
+   Building antlr4-python3-runtime==4.9.3
+      Built antlr4-python3-runtime==4.9.3
+ Downloaded networkx
+ Downloaded tokenizers
+ Downloaded diffusers
+ Downloaded hf-xet
+ Downloaded fonttools
+ Downloaded cryptography
+ Downloaded pillow
+ Downloaded torchvision
+ Downloaded nvidia-cuda-cupti-cu12
+ Downloaded matplotlib
+ Downloaded sympy
+ Downloaded pandas
+ Downloaded scikit-image
+ Downloaded numpy
+ Downloaded nvidia-nvjitlink-cu12
+ Downloaded transformers
+ Downloaded wandb
+ Downloaded nvidia-cuda-nvrtc-cu12
+ Downloaded scipy
+ Downloaded av
+ Downloaded opencv-python-headless
+ Downloaded nvidia-curand-cu12
+ Downloaded pyarrow
+ Downloaded nvidia-cusolver-cu12
+ Downloaded nvidia-cusparselt-cu12
+ Downloaded nvidia-cufft-cu12
+ Downloaded nvidia-cusparse-cu12
+ Downloaded nvidia-nccl-cu12
+ Downloaded triton
+ Downloaded nvidia-cublas-cu12
+ Downloaded nvidia-cudnn-cu12
+ Downloaded torch
+Prepared 125 packages in 43.34s
+Installed 127 packages in 197ms
+ + absl-py==2.4.0
+ + accelerate==1.13.0
+ + aiohappyeyeballs==2.6.1
+ + aiohttp==3.13.5
+ + aiosignal==1.4.0
+ + albucore==0.0.17
+ + albumentations==1.4.18
+ + annotated-types==0.7.0
+ + antlr4-python3-runtime==4.9.3
+ + async-timeout==5.0.1
+ + attrs==26.1.0
+ + av==16.1.0
+ + certifi==2026.4.22
+ + cffi==2.0.0
+ + charset-normalizer==3.4.7
+ + click==8.1.8
+ + cloudpickle==3.1.2
+ + contourpy==1.3.2
+ + cryptography==48.0.0
+ + cycler==0.12.1
+ + datasets==3.6.0
+ + diffusers==0.35.1
+ + dill==0.3.8
+ + dm-tree==0.1.10
+ + docstring-parser==0.18.0
+ + einops==0.8.1
+ + eval-type-backport==0.3.1
+ + farama-notifications==0.0.6
+ + filelock==3.29.0
+ + fonttools==4.62.1
+ + frozenlist==1.8.0
+ + fsspec==2025.3.0
+ + gitdb==4.0.12
+ + gitpython==3.1.46
+ + gymnasium==1.2.2
+ + hf-xet==1.5.0
+ + huggingface-hub==0.36.2
+ + idna==3.14
+ + imageio==2.37.3
+ + importlib-metadata==9.0.0
+ + inquirerpy==0.3.4
+ + jinja2==3.1.6
+ + jsonlines==4.0.0
+ + kiwisolver==1.5.0
+ + lazy-loader==0.5
+ + lmdb==1.7.5
+ + markdown-it-py==4.2.0
+ + markupsafe==3.0.3
+ + matplotlib==3.10.1
+ + mdurl==0.1.2
+ + mpmath==1.3.0
+ + msgpack==1.1.0
+ + msgpack-numpy==0.4.8
+ + multidict==6.7.1
+ + multiprocess==0.70.16
+ + networkx==3.4.2
+ + numpy==1.26.4
+ + nvidia-cublas-cu12==12.6.4.1
+ + nvidia-cuda-cupti-cu12==12.6.80
+ + nvidia-cuda-nvrtc-cu12==12.6.77
+ + nvidia-cuda-runtime-cu12==12.6.77
+ + nvidia-cudnn-cu12==9.5.1.17
+ + nvidia-cufft-cu12==11.3.0.4
+ + nvidia-cufile-cu12==1.11.1.6
+ + nvidia-curand-cu12==10.3.7.77
+ + nvidia-cusolver-cu12==11.7.1.2
+ + nvidia-cusparse-cu12==12.5.4.2
+ + nvidia-cusparselt-cu12==0.6.3
+ + nvidia-nccl-cu12==2.26.2
+ + nvidia-nvjitlink-cu12==12.6.85
+ + nvidia-nvtx-cu12==12.6.77
+ + omegaconf==2.3.0
+ + opencv-python-headless==4.11.0.86
+ + packaging==26.2
+ + pandas==2.2.3
+ + peft==0.17.1
+ + pfzy==0.3.4
+ + pillow==12.2.0
+ + platformdirs==4.9.6
+ + prompt-toolkit==3.0.52
+ + propcache==0.5.2
+ + protobuf==6.33.6
+ + psutil==7.2.2
+ + pyarrow==24.0.0
+ + pycparser==3.0
+ + pydantic==2.13.4
+ + pydantic-core==2.46.4
+ + pygments==2.20.0
+ + pyparsing==3.3.2
+ + python-dateutil==2.9.0.post0
+ + pytz==2026.2
+ + pyyaml==6.0.3
+ + pyzmq==27.0.1
+ + regex==2026.5.9
+ + requests==2.34.0
+ + rich==15.0.0
+ + safetensors==0.7.0
+ + scikit-image==0.25.2
+ + scipy==1.15.3
+ + sentencepiece==0.2.1
+ + sentry-sdk==2.59.0
+ + setuptools==82.0.1
+ + shtab==1.8.0
+ + six==1.17.0
+ + smmap==5.0.3
+ + sympy==1.14.0
+ + termcolor==3.2.0
+ + tifffile==2025.5.10
+ + tokenizers==0.22.2
+ + torch==2.7.1
+ + torchcodec==0.4.0
+ + torchvision==0.22.1
+ + tqdm==4.67.3
+ + transformers==4.57.3
+ + triton==3.3.1
+ + typeguard==4.5.1
+ + typing-extensions==4.15.0
+ + typing-inspection==0.4.2
+ + tyro==0.9.17
+ + tzdata==2026.2
+ + urllib3==2.7.0
+ + wandb==0.23.0
+ + wcwidth==0.7.0
+ + wrapt==2.1.2
+ + xxhash==3.7.0
+ + yarl==1.23.0
+ + zipp==3.23.1

outputs/_setup_logs/server_live.log

@@ -0,0 +1,20 @@
+Starting GR00T inference server...
+  Embodiment tag: EmbodimentTag.LIBERO_PANDA
+  Model path: checkpoints/GR00T-N1.7-LIBERO/libero_10
+  Device: cuda
+  Host: 127.0.0.1
+  Port: 5555
+/home/ubuntu/VLA_SAE/.venv/lib/python3.10/site-packages/albumentations/__init__.py:13: UserWarning: A new version of Albumentations is available: 2.0.8 (you have 1.4.18). Upgrade using: pip install -U albumentations. To disable automatic update checks, set the environment variable NO_ALBUMENTATIONS_UPDATE to 1.
+  check_for_updates()
+flash_attn is not installed. Falling back to sdpa attention. Install flash-attn for better performance: pip install flash-attn
+/home/ubuntu/VLA_SAE/gr00t/model/modules/dit.py:255: FutureWarning: Accessing config attribute `compute_dtype` directly via 'AlternateVLDiT' object attribute is deprecated. Please access 'compute_dtype' over 'AlternateVLDiT's config object instead, e.g. 'unet.config.compute_dtype'.
+  embedding_dim=self.inner_dim, compute_dtype=self.compute_dtype
+/home/ubuntu/VLA_SAE/gr00t/model/modules/dit.py:286: FutureWarning: Accessing config attribute `output_dim` directly via 'AlternateVLDiT' object attribute is deprecated. Please access 'output_dim' over 'AlternateVLDiT's config object instead, e.g. 'unet.config.output_dim'.
+  self.proj_out_2 = nn.Linear(self.inner_dim, self.output_dim)
+Total number of DiT parameters:  1091722240
+Total number of SelfAttentionTransformer parameters:  201433088
+
+
+✓ Server ready — listening on 127.0.0.1:5555
+
+Server is ready and listening on tcp://127.0.0.1:5555

outputs/_setup_logs/smoke_run.log

@@ -0,0 +1,67 @@
+Loaded 10 scenarios from examples/LIBERO/smoke_tests/scenarios_10.yaml
+  labels: 8 normal, 2 abnormal_probe
+
+Expected GR00T server command (run this in a separate terminal):
+  uv run python gr00t/eval/run_gr00t_server.py --model-path checkpoints/GR00T-N1.7-LIBERO/libero_10 --embodiment-tag LIBERO_PANDA --use-sim-policy-wrapper
+
+NOTE: The GR00T-N1.7 backbone 'nvidia/Cosmos-Reason2-2B' is a GATED HuggingFace repo.
+To start the server you must:
+  1. Request access at https://huggingface.co/nvidia/Cosmos-Reason2-2B (one click, usually instant).
+  2. Authenticate, e.g.  export HF_TOKEN=hf_xxx   (or: uv run hf auth login)
+  3. Re-run the server / smoke tests.
+
+Model checkpoint OK: checkpoints/GR00T-N1.7-LIBERO/libero_10
+Run directory: outputs/libero_smoke_tests/20260512_122756
+GR00T server reachable at 127.0.0.1:5555.
+LIBERO rollout interpreter: /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/python
+
+[1/10] normal_kitchen3_moka_pot_on_stove  (normal)  env=libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it  seed=1000  max_episode_steps=50
+  -> ok  rc=0  17.3s  actions=yes  video=yes  success=False
+
+[2/10] normal_kitchen4_bowl_in_drawer  (normal)  env=libero_sim/KITCHEN_SCENE4_put_the_black_bowl_in_the_bottom_drawer_of_the_cabinet_and_close_it  seed=1001  max_episode_steps=50
+  -> ok  rc=0  12.1s  actions=yes  video=yes  success=False
+
+[3/10] normal_living1_soup_and_cheese_in_basket  (normal)  env=libero_sim/LIVING_ROOM_SCENE1_put_both_the_alphabet_soup_and_the_cream_cheese_box_in_the_basket  seed=1002  max_episode_steps=50
+  -> ok  rc=0  13.3s  actions=yes  video=yes  success=False
+
+[4/10] normal_living2_soup_and_tomato_in_basket  (normal)  env=libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket  seed=1003  max_episode_steps=50
+  -> ok  rc=0  14.2s  actions=yes  video=yes  success=False
+
+[5/10] normal_study1_book_in_caddy  (normal)  env=libero_sim/STUDY_SCENE1_pick_up_the_book_and_place_it_in_the_back_compartment_of_the_caddy  seed=1004  max_episode_steps=50
+  -> ok  rc=0  12.0s  actions=yes  video=yes  success=False
+
+[6/10] normal_kitchen8_both_moka_pots_on_stove  (normal)  env=libero_sim/KITCHEN_SCENE8_put_both_moka_pots_on_the_stove  seed=1005  max_episode_steps=50
+  -> ok  rc=0  10.8s  actions=yes  video=yes  success=False
+
+[7/10] normal_kitchen6_mug_in_microwave  (normal)  env=libero_sim/KITCHEN_SCENE6_put_the_yellow_and_white_mug_in_the_microwave_and_close_it  seed=1006  max_episode_steps=50
+  -> ok  rc=0  12.1s  actions=yes  video=yes  success=False
+
+[8/10] normal_living5_two_mugs_on_plates  (normal)  env=libero_sim/LIVING_ROOM_SCENE5_put_the_white_mug_on_the_left_plate_and_put_the_yellow_and_white_mug_on_the_right_plate  seed=1007  max_episode_steps=50
+  -> ok  rc=0  15.1s  actions=yes  video=yes  success=False
+
+[9/10] abnormal_probe_obs_noise  (abnormal_probe)  env=libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it  seed=2000  max_episode_steps=50
+  -> ok  rc=0  10.8s  actions=yes  video=yes  success=False
+
+[10/10] abnormal_probe_short_timeout  (abnormal_probe)  env=libero_sim/LIVING_ROOM_SCENE2_put_both_the_alphabet_soup_and_the_tomato_sauce_in_the_basket  seed=2001  max_episode_steps=16
+  -> ok  rc=0  13.1s  actions=yes  video=yes  success=False
+
+======== SMOKE-TEST SUMMARY ========
+  scenario_id                              | label          | seed | rollout_started | actions_produced | video_saved | success | output_dir                                                                          | error_if_any
+  -----------------------------------------+----------------+------+-----------------+------------------+-------------+---------+-------------------------------------------------------------------------------------+-------------
+  normal_kitchen3_moka_pot_on_stove        | normal         | 1000 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_kitchen3_moka_pot_on_stove        |             
+  normal_kitchen4_bowl_in_drawer           | normal         | 1001 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_kitchen4_bowl_in_drawer           |             
+  normal_living1_soup_and_cheese_in_basket | normal         | 1002 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_living1_soup_and_cheese_in_basket |             
+  normal_living2_soup_and_tomato_in_basket | normal         | 1003 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_living2_soup_and_tomato_in_basket |             
+  normal_study1_book_in_caddy              | normal         | 1004 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_study1_book_in_caddy              |             
+  normal_kitchen8_both_moka_pots_on_stove  | normal         | 1005 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_kitchen8_both_moka_pots_on_stove  |             
+  normal_kitchen6_mug_in_microwave         | normal         | 1006 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_kitchen6_mug_in_microwave         |             
+  normal_living5_two_mugs_on_plates        | normal         | 1007 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/normal_living5_two_mugs_on_plates        |             
+  abnormal_probe_obs_noise                 | abnormal_probe | 2000 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise                 |             
+  abnormal_probe_short_timeout             | abnormal_probe | 2001 | yes             | yes              | yes         | False   | outputs/libero_smoke_tests/20260512_122756/abnormal_probe_short_timeout             |             
+
+Wrote: outputs/libero_smoke_tests/20260512_122756/summary.json
+Wrote: outputs/libero_smoke_tests/20260512_122756/summary.md
+
+Review the videos with:
+  python /home/ubuntu/VLA_SAE/examples/LIBERO/smoke_tests/review_smoke_tests.py --run-dir outputs/libero_smoke_tests/20260512_122756
+

outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/actions.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b7b5e56b63aa6efeeb95fae090d9988add5621839fd3eb16aa95c6b22fad60fe
+size 3264

outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/metadata.json

@@ -0,0 +1,36 @@
+{
+  "scenario_id": "abnormal_probe_obs_noise",
+  "label": "abnormal_probe",
+  "manifest_entry": {
+    "id": "abnormal_probe_obs_noise",
+    "label": "abnormal_probe",
+    "env_name": "libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it",
+    "instruction": "turn on the stove and put the moka pot on it",
+    "instruction_override": null,
+    "seed": 2000,
+    "max_episode_steps": 50,
+    "n_action_steps": 8,
+    "save_video": true,
+    "obs_noise_std": 6.0,
+    "action_repeat": 1,
+    "notes": "Simulation-only perturbation: mild zero-mean Gaussian noise (sigma=6 in 0-255 pixel units) is added to the image observations *before* they are sent to the policy. Smoke test only - this is a placeholder for future, more principled distribution-shift probes; no complex anomaly logic."
+  },
+  "resolved": {
+    "env_name": "libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it",
+    "seed": 2000,
+    "max_episode_steps": 50,
+    "n_action_steps": 8,
+    "save_video": true,
+    "save_frames": true,
+    "obs_noise_std": 6.0,
+    "action_repeat": 1,
+    "instruction_override": null
+  },
+  "model_path": "checkpoints/GR00T-N1.7-LIBERO/libero_10",
+  "policy_server": {
+    "host": "127.0.0.1",
+    "port": 5555
+  },
+  "libero_python": "/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/bin/python",
+  "timestamp": "2026-05-12T12:29:46"
+}

outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/rollout_summary.json

@@ -0,0 +1,22 @@
+{
+  "env_name": "libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it",
+  "seed": 2000,
+  "requested_max_episode_steps": 50,
+  "n_action_steps": 8,
+  "n_get_action_calls": 7,
+  "episode_successes": [
+    false
+  ],
+  "success": false,
+  "episode_length": 7,
+  "episode_reward": 0.0,
+  "elapsed_sec": 6.133596658706665,
+  "obs_noise_std": 6.0,
+  "action_repeat": 1,
+  "video_path": "outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/video.mp4",
+  "actions_path": "outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/actions.npy",
+  "policy_server": {
+    "host": "127.0.0.1",
+    "port": 5555
+  }
+}

outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/stderr.log

@@ -0,0 +1,22 @@
+[robosuite WARNING] No private macro file found! (__init__.py:7)
+[robosuite WARNING] It is recommended to use a private macro file (__init__.py:8)
+[robosuite WARNING] To setup, run: python /home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/robosuite/scripts/setup_macros.py (__init__.py:9)
+Matplotlib created a temporary config/cache directory at /tmp/matplotlib-syt5xusc because the default path (/home/ubuntu/.config/matplotlib) is not a writable directory; it is highly recommended to set the MPLCONFIGDIR environment variable to a writable directory, in particular to speed up the import of Matplotlib and to better support multiprocessing.
+Gym has been unmaintained since 2022 and does not support NumPy 2.0 amongst other critical functionality.
+Please upgrade to Gymnasium, the maintained drop-in replacement of Gym, or contact the authors of your software and request that they upgrade.
+See the migration guide at https://gymnasium.farama.org/introduction/migration_guide/ for additional information.
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:159: UserWarning: WARN: The obs returned by the `reset()` method is not within the observation space.
+  logger.warn(f"{pre} is not within the observation space.")
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:135: UserWarning: WARN: The obs returned by the `reset()` method was expecting numpy array dtype to be float32, actual type: float64
+  logger.warn(
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:131: UserWarning: WARN: The obs returned by the `reset()` method was expecting a numpy array, actual type: <class 'list'>
+  logger.warn(
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/spaces/box.py:240: UserWarning: WARN: Casting input x to numpy array.
+  gym.logger.warn("Casting input x to numpy array.")
+
+  logger.warn(f"{pre} is not within the observation space.")
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:135: UserWarning: WARN: The obs returned by the `step()` method was expecting numpy array dtype to be float32, actual type: float64
+  logger.warn(
+/home/ubuntu/VLA_SAE/gr00t/eval/sim/LIBERO/libero_uv/.venv/lib/python3.10/site-packages/gymnasium/utils/passive_env_checker.py:131: UserWarning: WARN: The obs returned by the `step()` method was expecting a numpy array, actual type: <class 'list'>
+  logger.warn(
+

outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/stdout.log

@@ -0,0 +1,7 @@
+Running collecting 1 episodes for libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it with 1 vec envs
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+[info] using task orders [0, 1, 2, 3, 4, 5, 6, 7, 8, 9]
+Collecting 1 episodes took 6.133342027664185 seconds
+SMOKE_RESULT_JSON: {"ok": true, "rollout_started": true, "actions_produced": true, "video_saved": true, "success": false, "error": null, "env_name": "libero_sim/KITCHEN_SCENE3_turn_on_the_stove_and_put_the_moka_pot_on_it", "seed": 2000, "n_action_calls": 7, "episode_length": 7, "episode_reward": 0.0, "video_path": "outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/video.mp4", "actions_path": "outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/actions.npy", "video_source_name": "1c2407cd-5353-4895-957f-f8b3986573d2_s0.mp4", "frames_dir": "outputs/libero_smoke_tests/20260512_122756/abnormal_probe_obs_noise/frames", "n_frames": 25}