Anonymous-neurips26/Mobile-RobustBench

Mobile-RobustBench是一个用于评估移动GUI代理在视觉扰动下鲁棒性的基准数据集。它包含9个应用、143个任务和712条轨迹，平均每条轨迹有5.48个步骤。数据集支持10种扰动类型，如元素缺失、元素遮挡、布局偏移等，覆盖感知和接地阶段。数据以JSON、PNG和XML格式存储，包括干净的轨迹和受扰动的轨迹，用于测试代理在真实移动环境中的性能。Mobile-RobustBench is a benchmark for evaluating the robustness of mobile GUI agents under visual perturbations. It includes 9 apps, 143 tasks, and 712 traces, with an average of 5.48 steps per trace. The dataset supports 10 perturbation types, such as element absence, element obscured, layout shift, etc., covering perception and grounding stages. Data is stored in JSON, PNG, and XML formats, including clean traces and perturbed traces, to test agent performance in real mobile environments.