AutoWebWorld: Synthesizing Infinite Verifiable Web Environments via Finite State Machines
Yifan Wu, Yiran Peng, Yiyu Chen, Jianhao Ruan, Zijie Zhuang, Cheng Yang, Jiayi Zhang, Man Chen, Yenchi Tseng, Zhaoyang Yu, Liang Chen, Yuyao Zhai, Bang Liu, Chenglin Wu, Yuyu Luo
2026-02-19
Summary
This paper introduces a new way to create realistic, but completely controlled, web environments for training AI agents to interact with websites.
What's the problem?
Training AI to use websites is hard because getting lots of examples of good website interactions is expensive and unreliable. When an AI tries something on a website, it's difficult to automatically check if it did the right thing because the website's inner workings are hidden. People have to manually verify each step, which takes time and money.
What's the solution?
The researchers built a system called AutoWebWorld that *creates* websites from simple blueprints. These blueprints, called Finite State Machines, clearly define every possible state of the website and what happens when you click a button or fill out a form. Because the rules are known, the system can automatically check if the AI is doing things correctly and if it's reaching the intended goal. They used this system to generate over 11,000 training examples very cheaply.
Why it matters?
This work is important because it allows for the creation of large amounts of high-quality training data for web AI agents without the need for expensive human verification. By training on this synthetic data, the AI agents become much better at interacting with *real* websites, outperforming previous methods and consistently improving as more training data is used.
Abstract
The performance of autonomous Web GUI agents heavily relies on the quality and quantity of their training data. However, a fundamental bottleneck persists: collecting interaction trajectories from real-world websites is expensive and difficult to verify. The underlying state transitions are hidden, leading to reliance on inconsistent and costly external verifiers to evaluate step-level correctness. To address this, we propose AutoWebWorld, a novel framework for synthesizing controllable and verifiable web environments by modeling them as Finite State Machines (FSMs) and use coding agents to translate FSMs into interactive websites. Unlike real websites, where state transitions are implicit, AutoWebWorld explicitly defines all states, actions, and transition rules. This enables programmatic verification: action correctness is checked against predefined rules, and task success is confirmed by reaching a goal state in the FSM graph. AutoWebWorld enables a fully automated search-and-verify pipeline, generating over 11,663 verified trajectories from 29 diverse web environments at only $0.04 per trajectory. Training on this synthetic data significantly boosts real-world performance. Our 7B Web GUI agent outperforms all baselines within 15 steps on WebVoyager. Furthermore, we observe a clear scaling law: as the synthetic data volume increases, performance on WebVoyager and Online-Mind2Web consistently improves.