Reproduce “CauSyncVLA: Causal Latent World Modeling for Asynchronous Robot Control” 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

Xu Pan, Chubin Zhang, Xingrui Yu, Runmin Lan, Zihao Wang, Wenkai Guo, Bofang Jia, Bo An and 1 more

CauSyncVLA lets a robot replan while visual processing runs slowly. A learned predictor updates cached vision-language memory from observed state changes and executed commands; a shifted remainder of the previous action plan is corrected with one step of a frozen expert. On LIBERO-10, it succeeds on 469 of 500 starts versus 456 for synchronous control; 5.04× measures model replanning capacity, not end-to-end speedup.

Summary by alphaXiv.

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