Reproduce “Blind Precise Assembly: Proprioceptive Sim-to-Real Peg-in-Hole for Diverse Parts” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Antonia Bronars, Younghyo Park, Kushal Kedia, Pulkit Agrawal
The authors train a policy to insert pegs using histories of joint-encoder motion to detect contact and guide local search, without images, force and torque sensing, a part model, or hole pose. On a real robot, it seats pegs in 173 of 237 trials (73%) with no fine-tuning, including three held-out pegs. Trials start near the hole.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.