Tag
This paper introduces hybrid interaction for mobile GUI agents by augmenting with app-native deeplinks to reduce navigation steps and improve task success. The proposed GUI-Hopper agent learns when to use deeplinks for direct navigation and GUI actions for fallback, showing gains in both simulated and real-device evaluations.
StepReflect reformulates per-step GUI reflection for mobile agents as supervised structured prediction, achieving higher transition accuracy than GPT-5.2 on AndroidWorld while reducing API costs.
Proposes Gated Hindsight Distillation (GHD), a method that uses future screenshots as privileged information to recover correct reasoning during training of mobile GUI agents, improving task success on AndroidWorld and AndroidLab across two vision-language models.
Introduces SeerGuard, a consequence-aware safety framework for mobile GUI agents that uses a safety-augmented world model to assess risks before execution, improving safety-utility scores.
This paper introduces Task-State Representation (TSR), a training-free framework that decouples persistent task state from transient screen observations to improve long-horizon mobile GUI agents, achieving up to 12 absolute point increases in success rate on complex tasks.
MobileForge presents an annotation-free adaptation system for mobile GUI agents that uses real app interaction and hierarchical feedback-guided policy optimization to improve performance, achieving near state-of-the-art results on AndroidWorld with open data.
UI-KOBE proposes a framework that enhances lightweight mobile GUI agents by constructing and leveraging app-specific knowledge graphs to improve task planning and execution efficiency.
MobileExplorer is a new framework that accelerates on-device inference for mobile GUI agents by performing lightweight parallel exploration of UI elements during model inference, reducing reasoning steps and latency by 23% while maintaining or improving task success rates.
SimuWoB is a synthetic benchmark with 120 challenging tasks for mobile GUI agents, using high-fidelity virtual environments and automatic reward generation. Experiments reveal that current agents achieve only 27.92% average success rate, dropping to 17.82% on long-horizon tasks, indicating substantial weaknesses in complex scenarios.