Reproduce “PICA-WAM: Visual Perturbation Invariance via Cross-Appearance Self-Alignment in World 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

Yude Zou, Yifei Yao, Guanbo Jiao, Jianming Ma, Guanjie Zheng, Xing Gao

PICA-WAM trains a robot action representation to align across nominal and appearance-perturbed views of the same task state, while retaining video prediction. It uses an EMA teacher to guide the perturbed-view model. In real-robot tests, it achieved 67.5% success across 160 trials, versus 33.1% for OpenWAM-α and 55.6% for π0.5; results apply to the tested setup.

Summary by alphaXiv.

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