Reproduce “Flip Anything with Agentic Robotics Self-Learning” 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
Shuangyu Xie, Kaiyuan Chen, Ken Goldberg
The authors develop a coding agent that improves a robot controller for flipping objects by revising inspectable graph-based code in physics simulation. Starting with bimanual handover, it adds strategies and recovery logic in response to failures. On 24 objects, success rises from 24.7% to 95.0% across 360 episodes; this is a best tuned simulation result, not evidence of real-world generalization.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.