Reproduce “Mind2Act World: A Diagnostic Benchmark for Embodied Mind, Physical Action, and Mind-Action Composition in Robotic Manipulation” 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
Yuzhi Huang, Jie Wu, Fanding Huang, Junchao Lin, Weichen Zhang, Ziyi Xiong, Yang Li, Zhengwei Xie and 35 more
Mind2Act World tests whether robots use memory, execution, and feedback across related simulated tabletop tasks. It separates memory-focused Mind tasks, action-focused Act tasks, and workflows where outcomes guide later decisions. GPT-6 Astra succeeds on 52.7% of Mind episodes, 39.0% of Act episodes, and 12.0% of Mind2Act episodes. These task-set rates do not isolate composition as the cause, and physical-robot transfer remains unmeasured.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.