Reproduce “Towards Pose-Robust Few-Shot Vision-Language-Action 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

Wenhao Li, Zhongyi Han, Yuhao Wei, Liang Han, Yilong Yin, Liqiang Nie

WACO estimates camera and robot-base shifts from arm geometry and a static scene, then canonicalizes images and robot state for a few-shot policy. It fine-tunes the policy on correction errors. Across shifted trials, it achieves 87.2% success in simulation and 80.0% on four UR5e tasks, versus 74.0% and 43.3% for the standard recipe. It assumes setup calibration and RGB-D/joint input.

Summary by alphaXiv.

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