Agent S2: A Compositional Generalist-Specialist Framework for Computer Use Agents
Saaket Agashe, Kyle Wong, Vincent Tu, Jiachen Yang, Ang Li, Xin Eric Wang
2025-04-02
Summary
This paper is about creating AI agents that can use computers like humans, by clicking buttons and navigating menus.
What's the problem?
Current AI agents struggle to accurately identify objects on the screen, plan for long tasks, and perform well on different types of tasks.
What's the solution?
The researchers developed Agent S2, which uses different AI models for different parts of the task, like finding objects and planning actions, to improve accuracy and performance.
Why it matters?
This work matters because it can lead to AI assistants that can automate tasks on computers and mobile devices, making people more productive.
Abstract
Computer use agents automate digital tasks by directly interacting with graphical user interfaces (GUIs) on computers and mobile devices, offering significant potential to enhance human productivity by completing an open-ended space of user queries. However, current agents face significant challenges: imprecise grounding of GUI elements, difficulties with long-horizon task planning, and performance bottlenecks from relying on single generalist models for diverse cognitive tasks. To this end, we introduce Agent S2, a novel compositional framework that delegates cognitive responsibilities across various generalist and specialist models. We propose a novel Mixture-of-Grounding technique to achieve precise GUI localization and introduce Proactive Hierarchical Planning, dynamically refining action plans at multiple temporal scales in response to evolving observations. Evaluations demonstrate that Agent S2 establishes new state-of-the-art (SOTA) performance on three prominent computer use benchmarks. Specifically, Agent S2 achieves 18.9% and 32.7% relative improvements over leading baseline agents such as Claude Computer Use and UI-TARS on the OSWorld 15-step and 50-step evaluation. Moreover, Agent S2 generalizes effectively to other operating systems and applications, surpassing previous best methods by 52.8% on WindowsAgentArena and by 16.52% on AndroidWorld relatively. Code available at https://github.com/simular-ai/Agent-S.