SEAgent: Self-Evolving Computer Use Agent with Autonomous Learning from Experience
Zeyi Sun, Ziyu Liu, Yuhang Zang, Yuhang Cao, Xiaoyi Dong, Tong Wu, Dahua Lin, Jiaqi Wang
2025-08-07
Summary
This paper talks about SEAgent, a new system that helps computer-use agents learn to use new software on their own by gaining experience and going through a series of tasks designed to teach them step-by-step.
What's the problem?
The problem is that current computer-use agents usually need a lot of human help or programming to learn new software, which makes it hard for them to adapt quickly to different programs or tasks without starting from scratch.
What's the solution?
The solution was to build SEAgent, a self-evolving framework where agents learn from their own past experiences and improve over time by completing a series of learning tasks. This method lets the agents master new software more independently and efficiently than before.
Why it matters?
This matters because it allows AI agents to become more flexible and capable, learning new programs without constant human guidance. It makes technology more useful and accessible by enabling smarter agents that can quickly adapt to many different computer tasks through their own learning.
Abstract
SEAgent, an agentic self-evolving framework, enables computer-use agents to autonomously master novel software through experiential learning and a curriculum of tasks, achieving superior performance compared to existing methods.