Research map / Vision-language and multimodal / Robot action and navigation
AnchorGUI: Asymmetric Memory for Dual-Scale Learning in GUI Navigation
Shengjie Jin, Zelong Sun, Hengbo Xu, Yanbiao Ma, Zhiwu Lu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add CSA dual-grained memory (ablation) | ReAct · AndroidWorld SR@1 | success rate: 43.7 -> 55.3 | Improved |
| add asymmetric window retention (ablation) | CSA memory · AndroidWorld SR@1 | success rate: 55.3 -> 57.3 | Improved |
| add asymmetric distillation instead of full visual distillation (ablation) | AnchorGUI · AndroidWorld SR@3 | success rate: 66.1 -> 69.2 | Improved |
| retain screenshots only for mismatched steps vs pure text retention (ablation) | AnchorGUI · AndroidWorld SR@3 | success rate: 62.1 -> 69.2 | Improved |
| retain screenshots only for mismatches vs full visual retention (ablation) | AnchorGUI · AndroidWorld SR@3 | success rate: 64.9 -> 69.2 | Improved |
| sliding window k=2 vs k=1 (ablation) | AnchorGUI · AndroidWorld SR@1 | success rate: 54.0 -> 57.3 | Improved |
| expand sliding window beyond k=3 (ablation) | AnchorGUI · AndroidWorld SR@1 | success rate: 57.3 -> 49.6 (k=8) | Hurt |
| use asymmetric distillation vs Reflexion | AnchorGUI · AndroidWorld SR@3 | success rate: 64.4 -> 69.2 | Improved |
| use AnchorGUI instead of ReAct | Qwen3-VL-8B · AndroidWorld SR@1 | success rate: 43.7 -> 57.3 | Improved |
| use AnchorGUI instead of ReAct | Qwen3-VL-8B · GUI-Odyssey | exact match: 58.9 -> 60.6 | No clear effect |
| use AnchorGUI vs fine-tuned OdysseyAgent | Qwen3-VL-8B · GUI-Odyssey | Type Match: 90.8 (OdysseyAgent) vs 80.6 | Hurt |
| use AnchorGUI instead of ReAct | Qwen3-VL-8B · AITZ | Type Match: 71.4 -> 75.2 | Improved |
Closest papers by meaning
- GUI Exploration Lab: Enhancing Screen Navigation in Agents via Multi-Turn Reinforcement Learning (NeurIPS 2025 · similarity 0.88)
- ShowUI: One Vision-Language-Action Model for GUI Visual Agent (CVPR 2025 · similarity 0.88)
- Scaling, Benchmarking, and Reasoning of Vision-Language Agents for Mobile GUI Navigation (ICML 2026 · similarity 0.88)
- GUI-Rise: Structured Reasoning and History Summarization for GUI Navigation (NeurIPS 2025 · similarity 0.87)
- FantasyVLN: Unified Multimodal Chain-of-Thought Reasoning for Vision-and-Language Navigation (CVPR 2026 · similarity 0.87)
- Aux-Think: Exploring Reasoning Strategies for Data-Efficient Vision-Language Navigation (NeurIPS 2025 · similarity 0.87)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
