sprint-baseline-s1
Microduck 2 m Sprint baseline trained locally with the unchanged official recipe seed 1; checkpoint 2750.
A perpetual policy for the microduck (61-D observation, 14 actions, 50 Hz). Runs until told otherwise — a gait, loaded into a policy slot.
Run it on a robot
sudo robotctl policy load walk mubeeapps-1/microduck-sprint-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-Sprint2m-MicroDuck - repo:
https://github.com/pollen-robotics/microduck-challenges.git - branch:
main - commit:
4b675ebf4 - checkpoint:
2750 - seed:
1 - base:
mjlab-microduck 0.1.0 @ 1efd853be - started:
2026-10-28T10:43:31Z
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-Sprint2m-MicroDuck --env.scene.num-envs 4096 --agent.max-iterations 3000 --agent.seed 1 --agent.logger tensorboard --agent.run-name baseline-s1
Validation
Trained locally using the unchanged official Sprint recipe, seed 1. Selected checkpoint 2750 completed 5/5 local Arena trials with a mean score of 5.123 seconds; an independent local retest reproduced that result. At the selected 0.60 m/s command, additional start seeds 6 to 25 all finished (mean 5.11775 seconds). These are simulation results on one Windows PC, not official server scores or physical robot validation.
This policy is a baseline reproduction, not a claim of a novel learning method.
- Downloads last month
- 17