Training High-Level Schedulers with Execution-Feedback Reinforcement Learning for Long-Horizon GUI Automation

Researchers developed CES, a multi-agent framework using reinforcement learning to improve GUI automation for long-horizon tasks. The system uses a Coordinator for planning, State Tracker for context management, and can integrate with any low-level Executor model to significantly enhance performance on complex automated tasks.

• →CES framework addresses GUI agent challenges in long-horizon tasks through specialized high-level scheduling models.
• →The system separates responsibilities with a Coordinator for strategic planning and State Tracker for context compression.
• →The framework is designed as a plug-and-play module that can enhance various existing Executor models.
• →Experiments demonstrate significant improvements in planning and state management capabilities for automated GUI tasks.
• →The approach uses execution-feedback reinforcement learning specifically for training high-level schedulers rather than unified policy models.