Reproduce “Co^2Gen: Collaboratively Generating Task Plans for Decentralized Multi-Robot and Human Collaboration” 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
Yi Loo, Siddhant Shrivastava, Minh Long Pham, Malika Meghjani
Co²Gen generates condition-checking shared plans that sequence familiar robot skills, then uses successful episodes to train separate vision-language-action (VLA) policies. A proposal-and-critique process refines executable plans, while a human verifies episode success. On two physical dual-arm tasks, its generated-data-plus-single-arm-demonstration model matched bimanual-human-data performance on storage within error and exceeded it on stacking; results are limited to these tasks.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.