Reproduce “HexaShell: A Thin Harness for Exploring the Limits of VLM Agents as Robot Coding Agents” 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, Alex Zhou, Jay Zhu, Zelin Zheng, Shijia Ge, Xiping Li, Di Wu, Yazhe Wang and 8 more
HexaShell lets a coding agent control a robot and improve reusable tools after failures. It uses RGB camera views to estimate object positions through triangulation, while a separate optimizer diagnoses recorded episodes and tests code repairs. On RoboDojo’s hidden seed, it reached 46.9% success and 53.2 progress, leading the leaderboard; however, success was only 32.0% on Precision.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.