playdead
Finger gun... bang! Sits, flops onto its back, legs up, head to one side.
An episodic policy for the microduck (61-D observation, 14 actions, 50 Hz). Runs 5.0 s and ends lying on its back.
Run it on a robot
sudo robotctl policy add playdead drfonz-indigo/microduck-playdead
robotctl robot do playdead
Things to know
- It finishes on its back. When the 5 s are up, the next policy has to be one that can get up. In simulation,
velstandgot the duck up in 32 of 32 headless attempts, but it flips the duck onto its front and springs up quickly (up to 12.5 rad/s), so keep a hand ready the first time. - The beak is not part of the policy. The jaw has no servo among the 14 actions. To open the beak (at about 3.5 s, for comic effect), the runtime must drive it.
- Not yet tried on a real robot. Evaluated in simulation only: in 256-duck evaluations (still stand, randomised robot, and mid-walk) it never ended on a side or face, always on its back with its head turned (about 80°), and the trunk followed the timeline to within 2.7°.
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:
drfonz/microduck_rl(design notes:docs/PLAY_DEAD.md) - branch:
play-dead - commit:
09724f402