limbo lean hold
"a duck that stands still on the spot, feet planted, and leans its whole body way back like it is going under a limbo bar, knees pushed forward, beak up, and holds that lean without walking"
A Microduck move trained by tfrere with
Microduck Academy. Family velocity
(walking styles (gaits)), tier 2, kind perpetual.
Judge (code, on the numbers): PASS (score 0.929) - {"height_ratio": 0.849, "height_m": 0.0976, "speed_mps": 0.0, "displacement_mps": 0.0, "pitch_deg": 5.2, "max_tilt_deg": 6.4, "yaw_rate_rps": 0.003, "head_yaw_ptp_rad": 0.025, "knee_left_rad": 0.698, "contact_fraction": [1.0, 1.0], "travel_m": 0.002, "heading_drift_rad": 0.019, "foot_lifts": [0, 0],
Eye (VLM, on the frames): not sure: no body lean - on the frames of checkpoint 3200, the one that ships
Commands (the battery, telemetry only): answers 2 of 2 commands. (command area 100 %; see docs/TRAINING.md, the command battery)
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 | 0.929 | not sure: no body lean | 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 limbo-lean-hold tfrere/microduck-move-limbo-lean-hold
robotctl robot do limbo-lean-hold
limbo lean hold
Hold a deep backward-leaning stationary pose with knees pushed forward, trunk angled back, and head pitched up, without stepping or traveling.