Reproduce “SEAR: Self-Evolving Agentic Robotics” 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
Xinyi Wang, Bangzheng Li, Hanyu Wang, Lin Zhang, Andrea Yaoyun Cui, Zhenggang Tang, Shyamal Buch, Dejia Xu
SEAR tests whether language-model agents can improve robot control by revising code and retaining skills and notes across tasks. It evaluates seven models on 330 simulated tasks and five physical tasks, using no-library and budget-matched restart controls. On 105 RoboLab tasks, GPT-5.6-sol scored 46.9% without a library, 42.4% with its own, and 60.6% with Claude Opus 5’s, showing library contents matter but borrowed libraries may hurt.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.