Research map / Vision-language and multimodal / Robot action and navigation
VisCritic: Visual State Comparison as Process Reward for GUI Agents
Jiachen Qian
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 Change Region Attention (ablation) | VisCritic · AndroidWorld ablation (ShowUI base) | success rate: 23.4 -> 24.8 | Improved |
| add multi-task head over success-only prediction (ablation) | VisCritic · AndroidWorld ablation (ShowUI base) | success rate: 24.1 -> 24.8 | Improved |
| add contrastive pre-training (ablation) | VisCritic · AndroidWorld ablation (ShowUI base) | success rate: 22.0 -> 24.8 | Improved |
| add post-execution verification (ablation) | VisCritic · AndroidWorld ablation (ShowUI base) | success rate: 21.4 -> 24.8 | Improved |
| use learned VDE instead of pixel-level differences (ablation) | VisCritic · AndroidWorld ablation (ShowUI base) | success rate: 21.2 -> 24.8 | Improved |
| visual state comparison instead of textual critic (ablation) | Critic head · AndroidWorld + Mind2Web verification | F1: 66.2 -> 84.6 | Improved |
| visual features instead of pixel difference (ablation) | Critic head · AndroidWorld + Mind2Web verification | F1: 70.5 -> 84.6 | Improved |
| add VisCritic to ShowUI | ShowUI · WebArena | task SR: 17.6 -> 23.2 | Improved |
| add VisCritic to Qwen2.5-VL agent vs text critic on OSWorld | Qwen2.5-VL Agent · OSWorld | task SR: 17.4 -> 17.1 | Hurt |
| add VisCritic to SeeClick | SeeClick · Mind2Web | step SR: 33.4 -> 37.6 | Improved |
| swap InternViT-6B for SigLIP-400M backbone (ablation) | VisCritic · AndroidWorld | F1: 80.3% F1 and 23.1% task SR | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ShowUI — ShowUI: One Vision-Language-Action Model for GUI Visual Agent (CVPR 2025) · better on 4 of 4 shared results
- CogAgent — CogAgent: A Visual Language Model for GUI Agents (CVPR 2024) · 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
- ShowUI: One Vision-Language-Action Model for GUI Visual Agent (CVPR 2025 · similarity 0.89)
- GUI-Eyes: Tool-Augmented Perception for Visual Grounding in GUI Agents (AAAI 2026 · similarity 0.89)
- HATS: Hardness-Aware Trajectory Synthesis for GUI Agents (CVPR 2026 · similarity 0.88)
- GUI-Shift: Enhancing VLM-Based GUI Agents through Self-supervised Reinforcement Learning (ICLR 2026 · similarity 0.87)
- Reflection with Action-Induced Visual Differences for Desktop GUI Agents (NeurIPS 2026 · similarity 0.87)
- CodeV: Code with Images for Faithful Visual Reasoning via Tool-Aware Policy Optimization (CVPR 2026 · similarity 0.86)
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.
