Play2Perfect: What Matters in Dexterous Play Pretraining for Precise Assembly?

2026-06-24

AI Summary

This work addresses the low sample efficiency in high-precision assembly tasks with multi-fingered dexterous hands, which stems from dense contacts and sparse rewards. The authors propose a two-stage approach: first, task-agnostic “play” pre-training to acquire general manipulation priors—such as grasping and in-hand reorientation—and then fine-tuning these skills for precise assembly. This strategy substantially improves both sample efficiency and generalization. The study presents the first systematic investigation into how object diversity, goal specification, trajectory variability, and target precision affect transfer performance. Leveraging reinforcement learning, the method conducts task-free pre-training across diverse objects, focusing on high-precision interactions during contact-intensive phases. Compared to training from scratch, it achieves a 33-fold improvement in sample efficiency; in zero-shot sim-to-real transfer, it attains a 60% success rate for tight-fitting insertions with a 0.5 mm clearance and over 50% success in long-horizon, multi-part assembly tasks.

Abstract

Multi-fingered robots promise the speed and dexterity of human hands, yet challenging problems such as precise assembly have remained out of reach. These tasks are contact-rich, making data collection for imitation learning difficult, and sparse-reward, making direct exploration with reinforcement learning (RL) intractable. Consequently, prior work has made progress by structuring the problem with specialized grippers, tool attachments, and environment fixtures. In this work, we argue that before a robot can perfect precise assembly, it must first learn to play. We further ask the question: what factors in the process of learning to play matter for precise assembly? We propose Play2Perfect, an RL framework for task-agnostic pretraining through play on diverse objects and goals, which is then perfected on precise assembly. The goal of play is to acquire reusable manipulation priors, such as grasping, in-hand reorientation and pose reaching. Finetuning then adapts this general prior to assembly, focusing exploration on the final contact-rich, high-precision interactions needed for success. We systematically study key design choices in play pretraining, including object diversity, training objective, trajectory diversity, and goal precision. We show that our prior is 33x more sample-efficient than RL training from scratch, even when provided with dense, multi-stage rewards. We demonstrate zero-shot sim-to-real transfer, achieving 60% success on tight insertions with only 0.5 mm contact clearance, and over 50% success on long-horizon multi-part assembly and screwing.

Problem

• dexterous manipulation
• precise assembly
• contact-rich tasks
• sparse-reward
• multi-fingered robots

Innovation

• play-based pretraining
• dexterous manipulation
• sample-efficient RL
• sim-to-real transfer
• precise assembly