Technology1:32 video2 primary sources

An MIT robot fits parts together by feel, with no camera

A Massachusetts Institute of Technology robot fits parts by joint sense, with no camera.

Credit: Bronars et al., arXiv (2026), CC BY 4.0 · Credit: Bronars et al., arXiv (2026), CC BY 4.0 (cropped)

Narrated and written with AI, checked against primary sources. Our standards

The gist

  • It uses only its own joint positions and speeds, not a camera, force, torque, or touch.
  • On a real seven-joint arm, it inserted pegs in 73% of trials with no extra training.
  • In simulation, robots without vision handled unseen shapes better than matched camera robots, but more slowly.

What happened

Researchers at the Massachusetts Institute of Technology trained a robot to fit parts using only its sense of its own joints, with no camera. On a real arm with seven joints, it inserted pegs in 73% of trials, with no extra training on the real robot. Accuracy tests used a gap of 100 micrometers, and the arm updated 50 times a second.

Why it matters

If a robot can assemble precise parts without a camera, some factory work could get simpler, though this result is not yet reviewed by other scientists.

How we know

MIT researchers reported a 73% success rate on a real robot arm in an October 16, 2026 arXiv preprint that has not yet been peer-reviewed.

What we don’t know yet

The paper is a preprint that other scientists have not yet reviewed, so the findings are early.

Sources & credits2 sources · 6 credits

Sources

  1. Blind Precise Assembly: Proprioceptive Sim-to-Real Peg-in-Hole for Diverse Parts (Bronars, Park, Kedia, Agrawal, MIT; Oct. 16, 2026; CC BY 4.0; not peer reviewed)arXiv preprint · Oct 16, 2026 · arxiv.org
  2. Blind Precise Assembly (MIT)Project page · Oct 16, 2026 · younghyopark.me

Image, footage and music credits

  • Credit: Bronars et al., arXiv (2026), CC BY 4.0
  • Credit: Bronars et al., arXiv (2026), CC BY 4.0 (cropped)
  • Illustration: made for Wunder News by Astra (Codex image generation, AI)
  • Data: arXiv preprint
  • Audio credit: narration by ElevenLabs AI voice (Max)
  • Credit: music and sound effects AI-generated for Wunder News with ElevenLabs

Corrections

No corrections recorded. See the corrections log. Report an error

Published . By , our AI correspondent, drawn only from the sources above. Our standards

Up next