Ground pick
An official Microduck move: Pollen Robotics' ground pick,
retrained from scratch by Microduck Academy with the upstream recipe
as-is (task Mjlab-GroundPick-Flat-MicroDuck of pollen-robotics/microduck_rl at 2b25a48,
5999 iterations, 4096 envs, no reward edit), so that it carries the
base/model.pt checkpoint the Academy needs to remix it. The policy Pollen ships is
alpha_ground_pick.onnx.
Pollen Robotics' ground pick: the beak goes down to the floor and comes back up, driven by a phase command over 2.8 s.
Fidelity check (this move vs Pollen's policy, same scene, same command, take phase)
| number | this move | Pollen's alpha_ground_pick.onnx |
|---|---|---|
| final_upright | True | True |
| max_tilt | 0.502 | 0.638 |
| base_z_mean | 0.0914 | 0.0979 |
| base_z_min | 0.0805 | 0.0839 |
| mouth_z_min | 0.0212 | 0.0204 |
| t_mouth_min_s | 0.26 | 1.08 |
| mouth_z_start | 0.188 | 0.1883 |
| mouth_z_end | 0.2162 | 0.2166 |
| base_z_end | 0.1166 | 0.1164 |
Recorded with space/payload/render/rollout.py on the upstream robot MJCF (spikes/base_moves/fidelity.py).
Run it on a robot
robotctl policy load ground_pick tfrere/microduck-move-base-ground-pick
This move takes a phase command (command.encoding: phase, period 4.0 s, stops at phase 0.7; the take lasts 2.8 s). robotctl policy add refuses phase commands: policy load puts it in the ground_pick slot, the runtime drives the phase itself.
Files
policy.onnx (play it, normalizer baked in), base/model.pt + base/agent.yaml + base/env.yaml (the rsl_rl
checkpoint and configs: remix = fine-tune from it), rollouts/0.traj (trajectory.v1 showcase take, with the
commands), video.mp4 + poster.jpg (the card clip), manifest.json
(schema 2 + academy block: origin, source task and policy, training facts, fidelity numbers), fidelity.json,
train.json.
Credits and license
The task, the reward design, the robot model and the training recipe are Pollen Robotics' work (pollen-robotics/microduck_rl, Apache-2.0). This repo republishes a retrained policy of that recipe under the same Apache-2.0 license, with the checkpoint, so the Academy's users can start from it.