Reproduce “ForceHarness: Turning Force into Action and Physics for Frozen Frontier Models” 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

Ke Sun, Kevin Yuchen Ma, Yike Zhang, Yanzhe Chen, Mike Zheng Shou

ForceHarness lets a frozen API model use force feedback for contact-rich robot tasks. Fast control primitives handle contact beneath the model, while the harness translates force traces into physical measurements for later decisions. On a real robot, GPT-6 Astra succeeded in 54 of 60 episodes across six tasks, versus 2 of 60 without force; these results cover the tested tasks.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.