Reproduce “WAM-TTT: Steering World-Action Models by Watching Human Play at Test Time” 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

Yusen Feng, Bingchen Han, Jiangran Lyu, Kai Liu, Yixin Zheng, Yuxuan Wan, Weiheng Liu, Sun Han and 8 more

WAM-TTT uses human videos to steer a pretrained robot policy through a temporary fast-weight memory, rather than converting them into robot actions. Paired human–robot data teach the memory during meta-training; at deployment, only video-side fast weights adapt. Across nine real-robot tasks in a combined household-scene shift, mean progress is 46.2%, versus 7.1% for WAM-ICL; tasks remain within the training distribution.

Summary by alphaXiv.

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