Reproduce “Demystifying When and Why VLAs Fail in Contact-Rich Tasks and How to Fix Them” 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

Carlota Parés-Morlans, Nils Kuhn, Y. Isabel Liu, Alberta Longhini, Jeannette Bohg

FACT improves contact-rich robot manipulation by reweighting training toward low-noise action predictions and conditioning action generation on current force and recent force history. On five tasks, it reaches an unweighted average success rate of 66%, compared with 40.5% for ForceVLA. This is a descriptive comparison, evaluated on one Franka robot with a wrist-mounted force/torque sensor.

Summary by alphaXiv.

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