GenHOI: Zero-Shot Contact-Aware Humanoid-Object Interaction via Video Imitation

Researchers have developed GenHOI, a novel framework enabling humanoid robots to perform diverse object-interaction tasks in a zero-shot manner by directly imitating generated videos, without task-specific training or physical demonstration data. The framework reconstructs the robot-object scene in simulation, generates task-oriented interaction videos, and extracts contact-relevant constraints that convert visual cues into physically grounded optimization priors. Validated across tasks including box grasping, bimanual chair carrying, table lifting, and cylindrical-object enveloping.