Reproduce “Atomic-WAM: Execution-Grounded Atomic-Motion Reasoning for 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

Shijun Shi, Zhe Liu, Zhi Hou, Yinan Chen, Tao Zhang, Kai Hu, Yong Liu, Jiangning Zhang

Atomic-WAM addresses broad robot instructions by having a vision-language reasoner describe the next motion from the instruction and scene, then using a world action model to predict future video and an action chunk. It trains the reasoner with feedback on whether a frozen executor completes motions. On RoboLab-120, success is 31.6% under vague instructions versus 20.6% for Cosmos3-Nano-Policy, in real-to-sim evaluation.

Summary by alphaXiv.

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