Reproduce “VLAA-GUI: Knowing When to Stop, Recover, and Search, A Modular Framework for GUI Automation” 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
Qijun Han, Haoqin Tu, Zijun Wang, Haoyue Dai, Yiyang Zhou, Nancy Lau, Alvaro A. Cardenas, Yuhui Xu and 6 more
VLAA-GUI checks whether requested interface outcomes are achieved, rather than relying on completed actions. Its Completeness-Aware Planner tracks visible end-state criteria; its Loop Breaker redirects repeated ineffective actions, and a searcher retrieves procedural guidance. With Claude Opus 4.6, it reaches 77.45% success on OSWorld’s 361-task set at 100 steps, above the 72.4% human reference, a benchmark comparison rather than a matched human study.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.