Research map / Reinforcement learning / RL from feedback
InfiGUI-G1: Advancing GUI Grounding with Adaptive Exploration Policy Optimization
Yuhang Liu, Zeyu Liu, Shuanghe Zhu, Pengxiang Li, Congkai Xie, Jiasheng Wang, Xueyu Hu, Xiaotian Han, Jianbo Yuan, Xinyao Wang, Shengyu Zhang, Hongxia Yang and 1 more
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 multi-answer generation to RLVR (ablation) | InfiGUI-G1-7B · ScreenSpot-Pro | Top-1 accuracy: 47.6 -> 51.9 | Improved |
| use Adaptive Exploration Reward instead of naive reward (ablation) | InfiGUI-G1-7B · ScreenSpot-Pro | Top-1 accuracy: 41.4 -> 51.9 | Improved |
| add rank factor k to AER (ablation) | InfiGUI-G1-7B · ScreenSpot-Pro | Top-1 accuracy: 44.0 -> 51.9 | Improved |
| add collinear penalty (ablation) | InfiGUI-G1-7B · ScreenSpot-Pro | Top-1 accuracy: 37.0 -> 51.9 | Improved |
| add collinear penalty (ablation) | InfiGUI-G1-3B · ScreenSpot-Pro | Top-1 accuracy: 35.3 -> 45.2 | Improved |
| add AER to 3B model (ablation) | InfiGUI-G1-3B · ScreenSpot-Pro | Top-1 accuracy: 38.4 -> 45.2 | Improved |
| add AEPO (multi-answer + AER) vs naive RLVR | Qwen2.5-VL-7B · ScreenSpot-Pro | Top-1 accuracy: 47.6 -> 51.9 | Improved |
| add AEPO vs naive RLVR | Qwen2.5-VL-3B · UI-Vision | Overall accuracy: 19.4 -> 22.0 | Improved |
| add AEPO vs naive RLVR | Qwen2.5-VL-7B · ScreenSpot-Pro hard subset | accuracy: 10.8 -> 17.4 | Improved |
| add AEPO vs naive RLVR | Qwen2.5-VL-3B · ScreenSpot-Pro hard subset | accuracy: 25.5 -> 31.4 | Improved |
| single-pass multi-answer vs naive RLVR given multiple attempts | InfiGUI-G1-7B · ScreenSpot-Pro | success rate: 52.1 (pass@4) -> 58.0 | Improved |
| AEPO on easy ScreenSpot-Pro samples | InfiGUI-G1 3B and 7B · ScreenSpot-Pro easy subset | accuracy: 100 -> 100 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- GUI-G2-7B — GUI-G²: Gaussian Reward Modeling for GUI Grounding (AAAI 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- GuirlVG: Incentivize GUI Visual Grounding via Empirical Exploration on Reinforcement Learning (ICLR 2026 · similarity 0.91)
- Test-Time Reinforcement Learning for GUI Grounding via Region Consistency (AAAI 2026 · similarity 0.91)
- GUI-G1: Understanding R1-Zero-Like Training for Visual Grounding in GUI Agents (NeurIPS 2025 · similarity 0.90)
- Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding (NeurIPS 2026 · similarity 0.90)
- SE-GUI: Enhancing Visual Grounding for GUI Agents via Self-Evolutionary Reinforcement Learning (NeurIPS 2025 · similarity 0.89)
- UI-Ins: Enhancing GUI Grounding with Multi-Perspective Instruction as Reasoning (ICLR 2026 · similarity 0.89)
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.
