Download 2026-09-24-walk-then-sit-take2/run.json from witcheer/microduck-hermes-runs: direct link, hf CLI and curl.
- Browser
- Download file 1.98 kB
-
https://huggingface.co/datasets/witcheer/microduck-hermes-runs/resolve/main/2026-09-24-walk-then-sit-take2/run.json
- Command line
-
hf download hf://datasets/witcheer/microduck-hermes-runs/2026-09-24-walk-then-sit-take2/run.json
-
curl -L -o run.json https://huggingface.co/datasets/witcheer/microduck-hermes-runs/resolve/main/2026-09-24-walk-then-sit-take2/run.json
1.98 kB
| { | |
| "sentence": "make the duck walk forward for five seconds then sit", | |
| "agent": "Hermes Agent 0.18.0 CLI on the same Linux box as the simulator: hermes chat -q \"<sentence>\" -s microduck-sim -t terminal --source tool --max-turns 20", | |
| "model": "Qwen3.8-27B Q6_K GGUF, llama.cpp llama-server via llama-swap (the alias in transcript.json meta reads Qwen3.6-27B-NVFP4; the file served was Qwen3.8-27B-Q6_K.gguf), one RTX 5090, local, no cloud model", | |
| "skill": "tools/microduck-sim-SKILL.md", | |
| "client": "tools/duckctl.py (our client; not Pollen's own duckctl app)", | |
| "simulator": "Pollen microduck duck-sim at 6507d2e: real robotd 0.12.0 (API 28) + MuJoCo body server, 50 Hz control loop, walking policy = Pollen's stock walk", | |
| "label": "simulation only. no real robot involved.", | |
| "start_posture": "sit", | |
| "model_state": "warm", | |
| "client_version": "before the posture fix (health prints robot.mode = walk, which is the locomotion hardware mode, not posture)", | |
| "tool_calls": [ | |
| "health", | |
| "walk --vx 0.3 --seconds 5", | |
| "stand", | |
| "walk --vx 0.3 --seconds 5", | |
| "sit", | |
| "status" | |
| ], | |
| "session_seconds_first_to_last_message": 32.8, | |
| "sim_truth_walk_m": 0.577, | |
| "robot_odometry_walk_m": 0.545, | |
| "outcome": "pass after self-correction", | |
| "note": "the model read mode walk as standing and sent a walk to a seated duck; the walk output showed odometry delta 0.0 and policy sit, so it stood the duck up and walked again. this run is why health now prints posture", | |
| "files": { | |
| "transcript.json": "full session: user sentence, model reasoning, tool calls, tool outputs (system prompt removed)", | |
| "transcript.md": "same, readable", | |
| "state.jsonl": "simulator ground truth at 30 Hz from the body server: wall time, sim time, trunk xyz, trunk quaternion, 14 joint positions (robotd wire order)", | |
| "monitor.log": "robotd's own monitor at 2 Hz (requested vs applied velocity, policy, deadman)", | |
| "replay.mp4": "offscreen MuJoCo replay of state.jsonl, fixed camera, real time" | |
| } | |
| } |