Reproduce “HexaShell: A Universal Harness for Robot Coding Agents in Simulation and the Real World” 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
Hongcheng Gao, Jianshu Zeng, Alex Zhou, Shijia Ge, Xuan Shangguan, Yazhe Wang, Yexing Wan, Zelin Zheng and 8 more
HexaShell provides robot-coding agents reusable perception, motion-checking, and error-reporting tools, with task-specific knowledge in reviewed natural-language skills. On RoboDojo, its GPT-6-Astra setup without skills achieved a 69.9% macro success rate, versus 22.5% for another agent using the same model. The comparison uses different agents and protocols, so it does not isolate which system component accounts for the gap.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.