Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly?
Precise assembly is contact-rich and sparse-reward, making demonstrations costly and reinforcement learning from scratch difficult.
Verified eventStanford UniversityJun 24, 2026
Evidence notes
- Play2Perfect first trains a dexterous hand on diverse free-space object play, then fine-tunes the reusable grasping, reorientation, and pose-reaching prior on exact assembly tasks.
- The authors report 33-times better sample efficiency than training from scratch, plus zero-shot sim-to-real success of 60% on 0.5 mm-clearance insertion and over 50% on multi-part assembly and screwing.
Company context
Research university; ToddlerBot humanoid research uses ROBOTIS DYNAMIXEL actuators per public bundle and paper materials.