Label-Free Mobile Agent Adapts to New Apps
jiqizhixin · x · 2026-07-17
This post introduces MobileForge, a new system for mobile GUI agents designed to help agents adapt to any mobile app without human annotation or demonstrations.
Methodology
- MobileGym: Automatically generates tasks and evaluates the agent during real app interactions.
- HiFPO: Converts both good and bad steps into learning signals, essentially providing the agent with step-by-step feedback to facilitate gradual improvement.
Results
- The adapted Qwen3-VL-8B achieved 67.2% Pass@3 on AndroidWorld, closely trailing the specifically trained GUI-Owl-1.5-8B at 69.0%.
- The authors' custom-trained ForgeOwl-8B further increased this to 77.6%.
- It achieved 41.0% on unseen MobileWorld tasks, making it the strongest reported open-data mobile agent to date.
Additional Info
The post also includes links to the paper, project page, GitHub, and institutional reports.
More from Research
- Chinese AI labs are now treating distillation obfuscation as the top research topic — pmddomingos · 2026-07-22
- Structural ensembles beat single predictions in TCR:pMHC generalization study — quaidmorris · 2026-07-22
- RSS launches under OMSF to push structural biology data modeling at scale — MoAlQuraishi · 2026-07-22
- enFoldX tops 8 neoantigen scans and an unseen-peptide benchmark — quaidmorris · 2026-07-22
- enFoldX reaches AUC 0.82 on human VDJdb and transfers to mouse at 0.76 — quaidmorris · 2026-07-22
- enFoldX gains accuracy as AF3 ensemble disagreement rises for non-binders — quaidmorris · 2026-07-22