Search papers, labs, and topics across Lattice.
This paper introduces UI-MOPD, a novel approach that utilizes multi-teacher on-policy distillation for continual learning in GUI agents, addressing the challenges of cross-platform interaction. By constructing the Uni-GUI dataset and dynamically selecting platform-specific teachers, the method effectively transfers behavioral priors to a shared policy, mitigating issues like catastrophic forgetting and capability degradation. Experiments reveal that UI-MOPD achieves task success rates of 38.2% and 12.0% on OSWorld and MobileWorld, respectively, highlighting its effectiveness in balancing retention of existing capabilities while adapting to new platforms.
UI-MOPD achieves a remarkable balance between retaining existing capabilities and adapting to new platforms, with task success rates that challenge conventional approaches in GUI agent learning.
Recent advances in multimodal foundation models and agent systems have driven GUI agents from single-platform task execution toward cross-platform interaction. However, building multi-platform GUI agents remains challenging. On one hand, high-quality and executable cross-platform interaction trajectories are still scarce, and existing data often suffer from limited platform coverage. On the other hand, different platforms exhibit distinct interaction conventions, making joint or continual training prone to behavioral pattern mixing, platform-specific capability degradation, and catastrophic forgetting. To address these challenges, we construct Uni-GUI, a high-quality cross-platform GUI interaction dataset, and propose UI-MOPD, the first method that incorporates multi-teacher on-policy distillation into continual learning for GUI agents. UI-MOPD dynamically selects a platform-specific teacher according to the current environment and transfers platform-specific behavioral priors to a shared policy through platform-conditioned distillation, enabling adaptation to new platforms while preserving capabilities on existing ones. Experiments on OSWorld and MobileWorld show that UI-MOPD achieves task success rates of 38.2% and 12.0%, respectively, demonstrating its effectiveness in balancing cross-platform capability retention and new-platform adaptation. Project page: https://elispectre.github.io/UI-MOPD/.
Tsinghua AI