sreetz-nv/so101_vials_actuators_sysid

SO-101 vials-to-rack — SysID actuator fit

This is the SysID arm. Actuator parameters resolve from the SO-101 asset's physics variant, which authors the AnchorBench system-identification fit.

Paired counterpart: sreetz-nv/so101_vials_actuators_hand_tuned, identical in every respect except that it substitutes the workshop's pre-SysID hand-tuned per-joint gains. The two were collected back to back for a controlled actuator-fit A/B; use them together or the comparison is lost.

What it is

100 scripted-success episodes of the SO-101 arm picking a vial off a mat and placing it in a yellow tube rack, recorded in Isaac Lab with Newton/MJWarp physics and the OVRtx renderer. Task instruction: "Pick up the vial and place it in the rack."

episodes	100
frames	33,441
mean frames/episode	334
fps	30
robot type	so101_follower
cameras	wrist + external, 480x640 RGB
environment	Newton-So101-Auto-Vials-State-Machine-DR-v0 (domain randomized)
collection	scripted state machine; only episodes accepted by the success termination term were kept

Actuator parameters (the variable under test)

Resolved per joint from the asset, not set in Python:

joint	stiffness	damping	armature	static friction	viscous friction	effort limit	velocity limit
shoulder_pan	48.42	3.335	0.0676	0.347	1.591	3.35	30
shoulder_lift	47.78	2.563	0.0276	0.345	0.561	3.35	30
elbow_flex	14.72	0.343	0.0377	0.411	0.797	3.35	30
wrist_flex	43.61	4.139	0.0507	0.248	1.072	3.35	30
wrist_roll	54.87	2.968	0.0549	0.222	1.638	3.35	30
gripper	68.25	2.819	0.0776	0.083	0.959	3.35	30

The grasp force servo is paired to the 68.25 N.m/rad gripper drive with gain 0.066 rad/(N.s) and a 0.198 rad/s jaw rate cap.

Held constant against the paired dataset

Same USD asset (so101_new_calib_SysID) with the same Robot/Sensor/Physics variants, same collision geometry, same contact material, same wrist camera and intrinsics, same solver and nconmax=512, same task, same domain randomization. Only the drive parameters differ.

Caveats

• Sim only. Nothing here speaks to real-robot transfer.
• The paired datasets have different episode lengths (334 against 290 mean frames), so a fixed step budget gives the two arms different epoch counts.
• Scripted acceptance filters scene difficulty; the acceptance rate differs between the paired arms.