roller-es1
Roller sprint, race-only. Base: our PPO policy stefo0/microduck-roller-warm1 (Pollen reference resumed with a skate-push reward). Fine-tuned by evolution strategies (OpenAI-ES, antithetic pairs, rank shaping, Adam; all actor weights, normaliser frozen; 300 generations x 64 candidates) against the Arena roller-sprint simulator, searched on start seeds 101-400 and validated on held-out seeds 1001-1030 (2.00 s); the board seeds 1-5 were never searched on. Simulation only; tuned to nominal Arena physics, not for a real robot.
A perpetual policy for the microduck (61-D observation, 14 actions, 50 Hz, on rollers). Runs until told otherwise — a gait, loaded into a policy slot.
Run it on a robot
sudo robotctl policy load <slot> stefo0/microduck-roller-es1
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:
main - commit:
423289dc8
- Downloads last month
- 18