STEVE: AStep Verification Pipeline for Computer-use Agent Training
Fanbin Lu, Zhisheng Zhong, Ziqin Wei, Shu Liu, Chi-Wing Fu, Jiaya Jia
2025-03-20
Summary
This paper is about training AI to use computers like humans by creating a system that checks if each step the AI takes is correct.
What's the problem?
Training AI to use computers is hard because it requires a lot of high-quality data showing the AI how to do things step-by-step.
What's the solution?
The researchers created STEVE, a system that uses another AI to check if each step in the AI's training data is correct, allowing them to use more data and train the AI more effectively.
Why it matters?
This work matters because it makes it easier and cheaper to train AI to use computers, which could lead to AI assistants that can help people with everyday tasks.
Abstract
Developing AI agents to autonomously manipulate graphical user interfaces is a long challenging task. Recent advances in data scaling law inspire us to train computer-use agents with a scaled instruction set, yet using behavior cloning to train agents still requires immense high-quality trajectories. To meet the scalability need, we designed STEVE, a step verification pipeline for computer-use agent training. First, we establish a large instruction set for computer-use agents and collect trajectory data with some suboptimal agents. GPT-4o is used to verify the correctness of each step in the trajectories based on the screens before and after the action execution, assigning each step with a binary label. Last, we adopt the Kahneman and Tversky Optimization to optimize the agent from the binary stepwise labels. Extensive experiments manifest that our agent outperforms supervised finetuning by leveraging both positive and negative actions within a trajectory. Also, STEVE enables us to train a 7B vision-language model as a computer-use agent, achieving leading performance in the challenging live desktop environment WinAgentArena with great efficiency at a reduced cost. Code and data: https://github.com/FanbinLu/STEVE.