maple.adapters.custom.openpi.alohasim
Adapter: OpenPI ↔ AlohaSim
- OpenPI
input: 224×224 RGB top image + end effector state + language instruction output: 7-dim action [x y z rx ry rz gripper] for each robot (continuous)
- LIBERO
AlohaSim keys: agentview_image, robot0_eye_in_hand_image, robot0_proprio-state, … action: 14-dim, gripper 1.0 = open, -1.0 = close
Classes
OpenPIAlohaSimAdapter. |