Reproduce “ME-Brain-1.0: Memory, Cognition and Action for Evolving Embodied Intelligence” 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

Wei He, Hengtao Li, Chenfeng Wang, Zhongrui Yu, Xuhan Zhu, Maokui He, Zide Liu, Xiyue Zhang and 16 more

ME-Brain is a system designed to use robot experience in later actions without retraining model weights. Its memory stores records outside the context window, while cognition plans and checks outcomes; Focus-VLWA uses selected history and local predictions around physical interactions to guide action. On RoboMME, it achieves 47.88% mean success, versus 44.62% for a baseline; this comparison does not isolate component effects.

Summary by alphaXiv.

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