UI-MOPD: Multi-Platform On-Policy Distillation for Continual GUI Agent Learning

2026-07-05

AI Summary

This work addresses key challenges in multi-platform GUI agents, including scarcity of high-quality interaction data, limited platform coverage, and issues in continual learning such as behavioral interference, platform-specific capability degradation, and catastrophic forgetting. To tackle these problems, the authors introduce Uni-GUI, a cross-platform GUI interaction dataset, and propose UI-MOPD—a novel method that, for the first time, integrates multi-teacher online policy distillation into a continual learning framework for GUI agents. UI-MOPD dynamically selects platform-specific teachers and employs a platform-conditional distillation mechanism to transfer behavioral priors into a shared policy, effectively balancing adaptation to new platforms with retention of capabilities on previously learned ones. Evaluated on OSWorld and MobileWorld, the approach achieves task success rates of 38.2% and 12.0%, respectively, significantly mitigating interference and forgetting in cross-platform learning.

Abstract

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/.

Problem

• cross-platform GUI agents
• continual learning
• catastrophic forgetting
• interaction trajectories
• platform-specific capabilities

Innovation

• multi-platform GUI agents
• on-policy distillation
• continual learning
• cross-platform interaction
• platform-conditioned distillation