MIT's blind assembly robot hits 73% peg-insertion success with zero-shot sim-to-real
chris_j_paxton · x · 2026-10-24
MIT Improbable AI Lab's "Blind Precise Assembly" paper shows robots can perform precise contact-rich assembly—peg insertion, gear meshing, nut threading—using only joint proprioception (positions, velocities, action history), with no cameras, tactile/force/torque sensors, or knowledge of part geometry and poses.
Key points:
- Policies trained entirely in simulation with heavy randomization of geometries, poses, and contacts
- Peg-insertion policy transfers zero-shot to real hardware, achieving 73% success on diverse tight-tolerance 3D-printed parts
- Gear meshing and nut threading achieve near-perfect success in sim; all three tasks run in real time
- Takeaway: rich sensory feedback isn't necessary for contact-rich assembly when policies learn to interpret their own motion—motor encoders are already high-resolution contact sensors
Author Antonia Bronars is on the job market for robotics research scientist roles starting 2027.
More from Embodied
- Robot foundation model Dream-1 serves 1.5M inference calls in 4 days across 60 countries — gharik · 2026-10-24
- NVIDIA team recruiting 2027 interns for humanoids, motion generation, and RL research — xbpeng4 · 2026-10-24
- "RL doesn't work until it does": researchers showcase RL for high-reliability robot manipulation — chris_j_paxton · 2026-10-24
- mujoco_warp_pbr: batched ray-traced PBR renderer for MuJoCo in NVIDIA Warp — mishig25 · 2026-10-24
- Claude Code taught a Roborock to escape from under the stairs, 10/10 success — fillyourcoffee · 2026-10-24
- NVIDIA expands Isaac robotics platform at GTC Berlin; enact_intel stealths with $30M pre-seed — lukas_m_ziegler · 2026-10-24