UI-MOPD: Multiplatform policy distillation for GUI agents

The advancement of multimodal models has enabled GUI agents to evolve from single-platform tasks to multiplatform interactions. However, building agents that operate coherently across multiple environments remains a major technical challenge: high-quality executable trajectory data is scarce, and each platform presents distinct interaction conventions that can lead to behavioral mixing or loss of previously acquired capabilities. To address this, the UI-MOPD proposal introduces a novel approach based on policy distillation with multiple teachers, dynamically selecting a platform-specific teacher to transfer behavioral biases to a shared policy through environment-conditioned distillation. This continual learning method allows adaptation to new platforms without sacrificing performance on known ones, achieving success rates of 38.2% on OSWorld and 12.0% on MobileWorld. For companies seeking to implement intelligent agents capable of orchestrating tasks across different applications, having a technology partner that offers artificial intelligence and AI agent services is key. At Q2BSTUDIO we develop custom applications and custom software that integrate artificial intelligence, cybersecurity, AWS and Azure cloud services, as well as business intelligence solutions such as Power BI. Our experience in enterprise AI allows us to design autonomous agents that learn and adapt to multiple environments, facilitating the automation of complex processes. Multiplatform policy distillation is just one example of how cutting-edge research can translate into real value for organizations.