Reproduce “Robota-Embed: Unified VLA Embeddings for Robot In-Context Learning and Evolution” 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

Yunze Liu, Jing Wang, Shenrui Wu, Wenkun He, Guanqi Zhan, Xianzheng Ma, Junyu Chen, Wenhang Ge and 11 more

Robota-Embed retrieves recorded future robot experiences by matching task instructions, interaction images, and native actions between a current state and a later video continuation. A shared vision-language-action model maps both into vectors for cosine ranking. On a four-source benchmark, it achieves 47.33% exact-clip R@1, versus 25.24% for Gemini Embedding 2; episode matches are more common than exact continuations.

Summary by alphaXiv.

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