MicroDuck Cloud Policies (official ONNX used in duck_play demos)
ONNX policies from the MicroDuck cloud simulation runtime, used by the duck_play closed-loop demos (multi-round ball chase, L1 person follow, "bring ball to owner").
Each policy follows the shared contract [1, 61] -> [1, 14] (obs: 48 proprioception + 13 command slots).
| File | Use |
|---|---|
alpha_walking.onnx |
velocity-commanded walking/standing (default locomotion) |
alpha_stand.onnx |
standing / settle before kicks |
ball_kick_left.onnx |
left-foot kick |
ball_kick_right.onnx |
right-foot kick |
Origin & license
Exported from the open-source pollen-robotics/microduck runtime (microduck/policies), Apache-2.0. Hosted here for reproducible demos (see XenderYang/microduck-duck-play code).