Reproduce “Utility-WAM: Correcting Future-Induced Action Changes” 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
Feihong Zhang, Xi Lin, Yulong Shi, Zuxing Lu, Yanghong Mei, Zhaopeng Gu, Tingyu Yuan, Zihao Liang and 2 more
Utility-WAM corrects action shifts caused by generated future video in a world-action model, then gates video generation according to predicted benefit over a future-free path. On RoboTwin 2.0, selective routing reaches 93.3% closed-loop success, uses video for 24.5% of queries, and cuts latency 37.9% versus always-video inference. Correction reduces two severe task losses, but residual losses remain.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.