Reproduce “MobileGym: A Verifiable and Highly Parallel Simulation Platform for Mobile GUI Agent Research” 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

Dingbang Wu, Rui Hao, Haiyang Wang, Shuzhe Wu, Han Xiao, Zhenghong Li, Bojiang Zhou, Zheng Ju and 3 more

MobileGym provides a restorable phone simulation for training and evaluating GUI agents. It represents environment state as structured data researchers can configure, restore, and fork, while agents interact through screenshots and actions. On 59 real-device tasks, training raised success from 32.2% to 72.9%, retaining 95.1% of the simulation gain; the authors call this an existence proof, not comprehensive sim-to-real evidence.

Summary by alphaXiv.

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