Robot learns precise insertion skills from simulation to real world
InsertAnything: Generalizable Contact-Rich Precision Insertion from Simulation to Reality
Robotics
Summary
Putting small parts together like puzzle pieces is tricky for robots because even tiny mistakes can cause the pieces to jam or collide. The authors created a way for a robot to practice these precise insertions only in computer simulations. Using simple force sensors and position targets, the robot learns to adjust and fix alignment errors on the fly. The trained robot can then successfully insert parts in real life without extra teaching or adjustments, even for new shapes it hasn’t seen before.
What this means in practice
- •For industrial robot programmers: Deploy trained simulation-only policies to real-world robotic assembly tasks for fast and reliable precision insertions without extra tuning.
- •For automation system integrators: Use generalizable insertion skills to handle multiple part geometries and clearances in manufacturing lines with less manual configuration.
Authors
Zhenghua Ma, Xinpan Meng, Zeyu Liu, Muyuan Ma, Hengdi Zhang, Houcheng Li, Long Cheng
Abstract
Contact-rich precision insertion is a key manipulation skill in robotic assembly. Tight clearances make insertion more sensitive to alignment errors and prone to collisions and jamming, while variations in geometry and clearance across parts further complicate policy reuse. We present a reinforcement learning framework that trains insertion policies entirely in simulation for direct deployment without real-world demonstrations or policy fine-tuning. By combining target poses with compact three-dimensional fingertip force feedback, the policy learns to search for alignment and correct its motion despite errors in the estimated hole position. A decoupled gated reward coordinates alignment and insertion. Force-signal smoothing and state-independent standard deviations stabilize the learning process. The resulting policies perform real-world insertion across multiple hole geometries with a minimum nominal clearance of 0.02 mm and improve success while reducing peak contact forces under hole-position errors. Cross-clearance and cross-geometry evaluations further confirm policy generalization. The system achieved the first perfect score of 20/20 on ManipulationNet's peg-in-hole benchmark under its Human-in-the-Loop protocol, with fully autonomous insertion motions. A single policy trained only on a simulated hexagonal insertion task achieved an overall success rate of 95.0% across eight unseen real-world insertion tasks. These results show that learning entirely in simulation can yield precision insertion skills that can be deployed directly and reused across real-world tasks. The project website (https://mzhsoul.github.io/InsertAnything/) provides open-source simulation and real-robot experiment scripts, assets, and trained checkpoints.