HERO: Learning Humanoid End-Effector Control for Visual Whole-Body Open-Vocabulary Object Grasping
Researchers have developed HERO, a modular system for humanoid robots to grasp objects in diverse real-world settings. This system combines large vision models for scene understanding with a novel residual-aware end-effector tracking policy for precise control. HERO achieved a 2.44cm end-effector tracking error, significantly outperforming previous methods, and demonstrated reliable grasping of everyday items in environments like offices and coffee shops. AI
IMPACT Enables more versatile and precise object interaction for humanoid robots in unstructured environments.