Ponder & Press: Advancing Visual GUI Agent towards General Computer Control
Yiqin Wang, Haoji Zhang, Jingqi Tian, Yansong Tang
- 🏛 Institutions
- Shenzhen International Graduate School, Tsinghua
- 📅 Date
- December 2, 2024
- 📑 Publisher
- Findings of ACL 2025
- 💻 Env
- Desktop Mobile Web
- 🔑 Keywords
TLDR
Ponder & Press is a pure-vision divide-and-conquer GUI-control framework that separates high-level instruction interpretation from element localization. It pairs a general-purpose MLLM interpreter with a GUI-specific locator, improves ScreenSpot grounding by 22.5%, and reports strong performance across web, desktop, and mobile GUI benchmarks.
Related papers
- UI-TARS: Pioneering Automated GUI Interaction with Native AgentsJanuary 21, 2025 · arXiv
- OS-ATLAS: A Foundation Action Model for Generalist GUI AgentsOctober 30, 2024 · ICLR 2025 (Spotlight)
- TinyClick: Single-Turn Agent for Empowering GUI AutomationOctober 9, 2024 · INTERSPEECH 2025
- SeeClick: Harnessing GUI Grounding for Advanced Visual GUI AgentsJanuary 17, 2024 · ACL 2024
- Aria-UI: Visual Grounding for GUI InstructionsDecember 20, 2024 · Findings of ACL 2025
- Iris: Breaking GUI Complexity with Adaptive Focus and Self-RefiningDecember 13, 2024 · arXiv