Research map / Reinforcement learning / RL from feedback
FDC-Ground: Improving GRPO for GUI Grounding via Exponential Rewards and Fact-Aligned Pruning
Xiangjian Zeng, Wenjing Li, Qingqiang Wu, Liang Zhang
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 |
|---|---|---|---|
| replace sparse binary reward with EDDR (ablation) | GRPO · ScreenSpot | average accuracy (%): 74.4 -> 85.9 | Improved |
| replace square decayed distance reward with EDDR (ablation) | GRPO · ScreenSpot | average accuracy (%): 85.5 -> 85.9 | Improved |
| add top-m selection by absolute advantage (ablation) | EDDR + GRPO · ScreenSpot | average accuracy (%): 85.9 -> 86.7 | Improved |
| add fact-aligned filtering to top-m pruning (ablation) | EDDR + Top-m · ScreenSpot | average accuracy (%): 86.7 -> 88.1 | Improved |
| apply FDC-Pruning (g=4, m=2) versus plain GRPO (ablation) | GRPO · ScreenSpot Web subset (Qwen2.5-VL-3B) | average accuracy (%): 66.1 -> 72.0 | Improved |
| apply FDC-Pruning (g=4, m=2) versus plain GRPO (ablation) | GRPO · ScreenSpot Web subset (Qwen2.5-VL-3B) | training time (s): 11541 -> 9789 | Improved |
| increase group size from 4 to 16 (ablation) | FDC-Pruning · ScreenSpot Web subset (Qwen2.5-VL-3B) | average accuracy (%): 72.0 -> 76.6 | Improved |
| increase group size from 16 to 64 (ablation) | FDC-Pruning · ScreenSpot Web subset (Qwen2.5-VL-3B) | average accuracy (%): 76.6 -> 74.8 | Hurt |
| increase group size from 16 to 64 (ablation) | FDC-Pruning · ScreenSpot Web subset (Qwen2.5-VL-3B) | training time (s): 19675 -> 57702 | Hurt |
| train with FDC-Ground RL | Qwen2.5-VL-3B · ScreenSpot-v2 | average accuracy (%): 46.9 -> 85.5 | Improved |
| add entropy regularization | FDC-Ground · GUI grounding training | reward plateau | No clear effect |
| add exploration-oriented rewards | FDC-Ground · GUI grounding training | accuracy | Hurt |
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.
- SE-GUI-7B — SE-GUI: Enhancing Visual Grounding for GUI Agents via Self-Evolutionary Reinforcement Learning (NeurIPS 2025) · better on 2 of 2 shared results
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
- Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding (NeurIPS 2026 · similarity 0.90)
- GUI-G1: Understanding R1-Zero-Like Training for Visual Grounding in GUI Agents (NeurIPS 2025 · similarity 0.90)
- Test-Time Reinforcement Learning for GUI Grounding via Region Consistency (AAAI 2026 · similarity 0.89)
- GuirlVG: Incentivize GUI Visual Grounding via Empirical Exploration on Reinforcement Learning (ICLR 2026 · similarity 0.89)
- Learning GUI Grounding with Spatial Reasoning from Visual Feedback (ICML 2026 · similarity 0.89)
- BAMI: Training-Free Bias Mitigation in GUI Grounding (CVPR 2026 · similarity 0.88)
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.
