Reproduce “Master Rules from Chaos: Learning to Reason, Plan, and Interact from Chaos for Tangram Assembly” 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
Chao Zhao, Chunli Jiang, Lifan Luo, Guanlan Zhang, Hongyu Yu, Michael Yu Wang, Qifeng Chen
MRChaos trains a robot in simulation to assemble unfamiliar seven-piece tangrams from silhouette prompts. The policy uses both the silhouette and a depth-camera view, learning through pixel-overlap feedback in a two-stage curriculum. On previously unseen human-created H-Fiendish tasks, it reaches 73.6% final coverage, versus 62.3% for behavior cloning. Real-world H-Hard deployment reaches 62.4% final coverage, within fixed-order planar assembly.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.