Reproduce “World In-Context Learning for Agentic Robot Manipulation” 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
Heng Fang, Zehui Guo, Yuehan Zhang, Mingyang Du, Dingkang Liang, Xiang Bai
RoboScript lets a frozen model learn manipulation tasks by rehearsing from one demonstration in simulation and recording task steps, object roles, and relative geometry in text memory without updating its weights. On RoboTwin 2.0’s 50 dual-arm tasks, GPT-6.1-Sol achieves 83.4% mean success, versus 61.7% for 4D-WAM, after clean-only rehearsal; this result is limited to the benchmark and protocol.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.