Computer-use agents improve themselves with real-world data loops
CUA-Master: Scaling Computer-use Agents through Open-World Recursive Improvements
Artificial Intelligence
Summary
Training computer-use agents is hard because they need lots of diverse and reliable interaction data. The authors present CUA-Master, a system that keeps learning by collecting real-world computer tasks from videos and then uses those tasks to teach the agents better ways to interact with software. Their method uses two feedback loops: one keeps updating the system with new data and fixes its weaknesses, while the other turns example tasks into smaller building blocks for creating even more complex tasks. This way, the agents get better over time by learning from genuine human workflows instead of only their own experiences.
What this means in practice
- •For software tool developers: Create smarter user-assisting agents that learn continuously from real software usage videos and data.
- •For automation engineers: Build reliable automation systems by synthesizing complex tasks from smaller reusable primitives mined from real user behaviors.
Authors
Xinlei Yu, Gen Li, Ji Zhao, Zhangquan Chen, Congcong Wang, Shuai Dong, Zelin Peng, Yuqi Xu, Hongrui Wu, Kaiwen Tuo, Cheng Yang, Kaituo Feng, Yifan Wu, Jiayi Zhang, Dongming Wu, Wenqi Shao, Jiaqi Wang, Xiangyu Yue
Abstract
Despite recent advances in CUAs, their development remains constrained by the limited diversity, controllability, and reliability of interaction data. Moreover, conventional optimization pipelines lack iterative feedback, while existing recursive improvement approaches often rely on self-generated experiences, limiting ability expansion and yielding diminishing returns. To address these challenges, we introduce CUA-Master, a nested double-loop recursive improvement framework. At the system level, an outer world-data-model flywheel continuously incorporates open-world corpus and leverages evolving model weaknesses to guide subsequent data construction and optimization iterations. At the data level, an inner flywheel couples open-world seeding and autocatalytic propagation. The former mines authentic workflows as seed tasks from instructional videos, and employs visual-semantic dual-path trajectory reconstruction; The latter transforms these seeds into reusable primitives for controllable synthesis of diverse, complex, and verifiable tasks. To enable efficient and reliable trajectory collection, we further introduce a feedback-guided snapshot-rollback retry mechanism in various real environments. Extensive experiments demonstrate that CUA-Master-27B achieves competitive performance across GUI-only and GUI-CLI hybrid benchmarks, scoring 80.2 on OSWorld Verified, 36.9 on OSWorld 2.0, 43.9 on WeaveBench, and 25.0 on Agents' Last Exam. Our framework also substantially improves the quality and efficiency of data synthesis, and exhibits continuous gains across recursive iterations.