CUA-Master Implements Nested Double-Loop Recursive Improvement for Computer-Use Agents

CUA-Master utilizes a nested double-loop framework to scale computer-use agents. An outer flywheel integrates open-world data to address model weaknesses, while an inner flywheel uses autocatalytic propagation to optimize data construction and interaction reliability.