Look Before You Leap: A GUI-Critic-R1 Model for Pre-Operative Error Diagnosis in GUI Automation

Fei Huang (Alibaba Group) · Xi Zhang (Nanyang Technological University) · Ming Yan (Alibaba Group) · Haiyang Xu (University of California, San Diego) · Yuyang Wanyan (Institute of automation, Chinese academy of science, Chinese Academy of Sciences) · Haowei Liu (Institute of Automation, Chinese Academy of Sciences) · Junyang Wang (Beijing Jiaotong University) · Jiabo Ye (East China Normal University) · Yutong Kou (Institute of Automation, Chinese Academy of Sciences) · Xiaoshan Yang (Institute of automation, Chinese academy of science, Chinese Academy of Sciences) · Weiming Dong (Institute of Automation, Chinese Academy of Sciences) · Changsheng Xu (Institute of automation, Chinese academy of science, Chinese Academy of Sciences)
critic accuracycritic modeldata collection pipelinedecision-making errorsdynamic evaluationgui automationgui-critic-testgui-critic-trainmultimodal large language modelsonline interactive environmentsoperational efficiencypre-operative critic mechanismreasoning-bootstrappingsuggestion rewardsuggestion-aware group relative policy optimization

In recent years, Multimodal Large Language Models (MLLMs) have been extensively utilized for multimodal reasoning tasks, including Graphical User Interface (GUI) automation. Unlike general offline multimodal tasks, GUI automation is executed in online interactive environments, necessitating step-by-step decision-making based on the real-time status of the environment. This task has a lower tolerance for decision-making errors at each step, as any mistakes may cumulatively disrupt the process and potentially lead to irreversible outcomes like deletions or payments. To address these issues, we introduce a pre-operative critic mechanism that provides effective feedback prior to the actual execution, by reasoning about the potential outcome and correctness of actions. Specifically, we propose a Suggestion-aware Group Relative Policy Optimization (S-GRPO) strategy to construct our pre-operative critic model GUI-Critic-R1, incorporating a novel suggestion reward to enhance the reliability of the model's feedback. Furthermore, we develop a reasoning-bootstrapping based data collection pipeline to create a GUI-Critic-Train and a GUI-Critic-Test, filling existing gaps in GUI critic data. Static experiments on the GUI-Critic-Test across both mobile and web domains reveal that our GUI-Critic-R1 offers significant advantages in critic accuracy compared to current MLLMs. Dynamic evaluation on GUI automation benchmark further highlights the effectiveness and superiority of our model, as evidenced by improved success rates and operational efficiency. The code is available at https://github.com/X-PLUG/MobileAgent/tree/main/GUI-Critic-R1.