Reproduce “RoboAtom: Large-scale Atomic-Action Supervision for Robot Learning” 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
Xueyuan Wei, Xuanxu Wang, Xudong Li, Yuping He, Wanshui Gan, Jiaqi Peng, Wenqi Guo, Jianjun Zhou and 5 more
RoboAtom builds a shared dataset of time-aligned robot action segments, each with a skill label, description, and frame boundaries. It reviews seed examples, then uses state and action signals to propagate boundaries across executions with the same ordered skills. The dataset contains 1,446,974 trajectories and 6,876,282 skill-labeled segments. On this shared-sequence setup, PhysExtend scores 71.0 Skill [email protected], versus Cortex’s 48.5.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.