Coordinated Humanoid Manipulation with Choice Policies

Sliding and Pickup Hand-eye Coordination Eye Tracking Successful Insertion Fig. 1 : Coordinated Humanoid Manipulation. We present a teleoperation system and a policy learning framework for coordinated humanoid manipulation. Our modular interface enables efficient data collection, while the proposed Choice Policy learns to handle multimodal demonstrations with a single forward pass. These components allow the humanoid robot to perform complex, long-horizon tasks, such as dishwasher loading (top row) and whole-body loco-manipulation for whiteboard wiping (bottom row). These tasks require precise coordination between the hands, eyes, and locomotion systems. Videos and additional results are available on https://choice-policy.github.io.