Reproduce “CoRE: Learning Collaboration-Role Experts for Decentralized Collaborative Manipulation with One Policy” 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

Yanan Zhou, Zhaoyan Qian, Zihao Li, Mingyuan Ba, Ranpeng Qiu, Weiming Zhi

CoRE trains one shared robot policy to coordinate manipulation from each robot’s local observations, without robot identities, role labels, or inter-robot messages. Pooled demonstrations teach adaptable expert paths, while action–expert alignment trains a router to select among them at each action position. In RoboFactory, CoRE reaches 84.2% average full-task success versus 58.4% for CHORUS, with Take Photo at 29/100.

Summary by alphaXiv.

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