Reproduce “GaP: A Graph-as-Policy Multi-Agent Self-Learning Harness For Variational Automation Tasks” 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

Kaiyuan Chen, Shuangyu Xie, Letian Fu, Justin Yu, William Pacini, Sandeep Bajamahal, Hudson Kim, Jaimyn Drake and 16 more

Graph-as-Policy (GaP) assembles reusable robot skills into an executable graph, rehearses and revises it in simulation, and deploys the selected program for repeated tasks with bounded object and pose variation in a known workcell. On the popcorn task, refinement raised simulation success from 33% to 94%, and physical trials succeeded 18 of 20 times; this does not establish industry-level reliability.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.