Reproduce “Coding Agents for Generalized Task and Motion Planning Problems” 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
Matteo Merler, Bowen Li, Josh Roy, Yichao Liang, Qianwei Wang, Yixuan Huang, Tom Silver
The authors use coding agents to synthesize reusable programs for generalized task-and-motion planning by probing simulators, writing tests, and revising code before freezing each program for evaluation. Across 16 environments with planner baselines, GPT-6 Astra averaged 95% held-out success versus 47% for planners. The evidence is limited to simulated, fully observed tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.