Taeksoo Kim

The paper presents AutoDex, a system that automatically collects real-world data for robust dexterous grasping, achieving a 4.8x throughput improvement over teleoperation and higher success rate than simulation-only validation.

This paper proposes robot-factored world models for action-conditioned video prediction in robotics, which factor out action realization and robot rendering to avoid learning the realization process and future-state leakage.