play dead flat
"a duck that plays dead: it drops to the floor and lies there completely still, flat on the ground, and stays down"
A Microduck move trained by tfrere with
Microduck Academy. Family standup
(get up / land in a pose), tier None, kind episodic.
Judge (code, on the numbers): PASS (score 0.966) - {"height_ratio": 0.335, "height_m": 0.0385, "speed_mps": 0.0, "displacement_mps": 0.0, "pitch_deg": 1.6, "max_tilt_deg": 8.3, "yaw_rate_rps": 0.0, "head_yaw_ptp_rad": 0.005, "knee_left_rad": -0.022, "contact_fraction": [1.0, 1.0], "travel_m": 0.0, "heading_drift_rad": 0.001, "foot_lifts": [0, 0], "f
Eye (VLM, on the frames): not sure: spawned already down
Score shown in Discover: 75 % (the judge's score, capped at 75 % when the eye is not sure and at 50 % when it disagrees; see docs/TRAINING.md section 8).
| round | judge | judge score | eye (VLM) | score |
|---|---|---|---|---|
| 1 | pass | 0.976 | disagrees: standing bow instead of drop | 50 % |
| 2 | pass | 0.976 | disagrees: crouching instead of lying down | 50 % |
| 3 | pass | 0.966 | not sure: spawned already down | 75 % |
Files: policy.onnx (play it), model.pt (remix = fine-tune from it), rollouts/*.traj (trajectory.v1
recordings), checkpoints/r<round>-<iter>.traj (the training steps of every round, listed in
manifest.checkpoints), manifest.json (schema 2 + judge_score / vlm / score + academy block with
prompt, family, judge, lineage), video.mp4 + poster.jpg (the card clip).
Run it on a robot
sudo robotctl policy add play-dead-flat tfrere/microduck-move-play-dead-flat
robotctl robot do play-dead-flat
play dead flat
a duck that plays dead: it drops to the floor and lies there completely still, flat on the ground, and stays down