YAML Metadata Warning:empty or missing yaml metadata in repo card
Check out the documentation for more information.
tags:
microduck
robotics
reinforcement-learning
onnx
library_name: microduck
pipeline_tag: robotics
roller-baseline-s1
Mjlab-RollerSprint2m-MicroDuck
A perpetual policy for the microduck (61-D observation, 14 actions, 50 Hz, on rollers). Runs until told otherwise ๏ฟฝ a gait, loaded into a policy slot.
Run it on a robot
sudo robotctl policy load <slot> mubeeapps-1/microduck-roller-baseline-s1
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
task_id:
Mjlab-RollerSprint2m-MicroDuckrepo:
https://github.com/pollen-robotics/microduck-challenges.gitbranch:
maincommit:
4b675ebf4checkpoint:
2999seed:
1base:
mjlab-microduck 0.1.0 @ 1efd853bestarted:
2026-10-29T10:08:32Z
Reproduce
Same code, same uv.lock, same command, same seed. Training it again yields a comparable policy, not the same weights: GPU reinforcement learning is not bit-reproducible across machines.
git clone https://github.com/pollen-robotics/microduck-challenges.git
cd microduck-challenges
git checkout 4b675ebf4
uv sync
uv run train Mjlab-RollerSprint2m-MicroDuck --env.scene.num-envs 4096 --agent.max-iterations 3000 --agent.seed 1 --agent.logger tensorboard --agent.run-name roller-baseline-4096-s1
Local evaluation
Unchanged shipped Roller Sprint recipe:4096 environments,3000 iterations,seed1. Selected checkpoint2999:local3.118seconds,5/5 finishes at1.20m/s. Checkpoint1500:3.120seconds,5/5. Local Pollen reference:3.128seconds,5/5. The10ms local reference difference is small and does not establish a robust improvement. Checkpoints0,1250,2500,2750 had0/5;1750 had1/5. All13checkpoints were evaluated. No physical hardware test or scientific novelty is claimed.