Blind Precise Assembly: Proprioceptive Sim-to-Real Peg-in-Hole for Diverse Parts

Key points

• Every robot already carries high-resolution contact sensors: its own joints.1
• We study this question by training blind assembly policies that observe only a short history of joint positions, velocities, and their own actions, without visual, force, torque, or tactile sensing.1
• Trained with reinforcement learning entirely in simulation over randomized parts, grasps, and contacts, such policies learn peg insertion, gear meshing, and nut threading.1
• Training such policies needs thousands of different assemblies, each simulated with its clearance intact, so we also release an open-source generator that builds them procedurally from exact convex pieces for MuJoCo; it is more accurate and faster than convex decomposition or MuJoCo's signed-distance collision.1

Sources (1)

• [1]Blind Precise Assembly: Proprioceptive Sim-to-Real Peg-in-Hole for Diverse Parts
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 16, 05:58 PM

  Is joint sensing also enough for precise, contact-rich assembly, where the robot must find a millimeter-scale opening, mesh gear teeth, or catch a thread through light contact?

  Every robot already carries high-resolution contact sensors: its own joints.

Extractive summary: sentences quoted from the sources.