Reproduce “Subgoal-aware Memory Graph for GUI Agent” 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

Dezhao Luo, Daniel Beechey, Derek Yuen, Weilin Luo

The authors propose a subgoal-graph memory for mobile GUI agents, storing contiguous action sequences by intent, visible change, and outcome, then linking visually similar screen states for retrieval. On SPA-Bench’s English, single-app, level-3 split, using Qwen3-VL-8B-Instruct, it reports 45.30% per-attempt success versus 34.19% without memory, across 13 tasks; this is evidence only on that evaluation.

Summary by alphaXiv.

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