microduck-walk

<!-- microduck-skill:preview -->

Preview

<video controls muted loop width="100%"> <source src="https://huggingface.co/neil-jo/microduck-walk/resolve/main/preview.mp4" type="video/mp4"> </video>

Sim checkpoint replay (play.sh). Not the ONNX runtime path, not a real robot, not an official Pollen policy.

<!-- /microduck-skill:preview -->

walk

Velocity-tracking flat walk from Mjlab-Velocity-Flat-MicroDuck, 4096 envs × 4000 iters on Hugging Face Jobs. Community example from the microduck-plugin loop. Not an official Pollen policy. Weights live here; they are not bundled in the plugin git.

A perpetual policy for the microduck (61-D observation, 14 actions, 50 Hz). Runs until told otherwise — a gait for the walk slot.

Agent plugin / skill

git clone https://github.com/acnlabs/microduck-plugin
# Point an Agent Plugins client at that folder (the one with plugin.json).

Skill-only clients:

npx skills add acnlabs/microduck-plugin@microduck-skill -g -y

Then the agent runs localhost control (do not type this yourself unless you are debugging):

control.sh start --repo neil-jo/microduck-walk --detach

ClawHub listing for microduck-skill is still pending review.

If you own a Microduck

sudo robotctl policy load walk neil-jo/microduck-walk

The observation normalizer is baked into policy.onnx; feed raw observations. manifest.json follows schema 2 of the microduck policy manifest (docs/policy-manifest.md in the daemon repo).

Training

• repo: pollen-robotics/microduck_rl
• branch: develop
• commit: 29e887ecf