sprint-k2
Sprint 2 m: lane progress reward, one-sided speed tracking, free head at speed, mirror loss + lane drift; local 1.412 s 5/5
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 <slot> heyskylark/microduck-sprint-k2
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:
git@github.com:heyskylark/microduck-challenges.git - branch:
main - commit:
dfd67e0da - checkpoint:
1999 - seed:
2 - base:
mjlab-microduck 0.1.0 @ 8d0dbc94c - started:
2026-10-31T03:27:55Z
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 git@github.com:heyskylark/microduck-challenges.git
cd microduck-challenges
git checkout dfd67e0da
uv sync
uv run train Mjlab-Sprint2m-MicroDuck --env.scene.num-envs 4096 --agent.max-iterations 2000 --agent.seed 2 --agent.logger tensorboard --agent.run-name K2-mirror-drift-s2
- Downloads last month
- 11