Reproduce “OmniBrain-0: A Unified Agentic Model for 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
Xiaomi Embodied Intelligence Team, Yuechen Luo, Chong He, Fang Li, Shaoqing Xu, Jingkai Wang, Menglun Su, Jiaxin Liu and 24 more
OmniBrain-0 coordinates specialist robot tools across manipulation and navigation, using instructions, observations, and execution feedback to choose tools and revise decisions. The authors train its decision model through curated interactions and multi-teacher on-policy distillation. On LIBERO-Pro, it achieves 65.5% success in 1.6 minutes, below few-shot GPT-5.6-Sol’s 78.3% in 13.8 minutes. Unlike OmniBrain-0’s main-table evaluation, teacher results use an Experience Library.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.