Saved in:
Bibliographic Details
Main Authors: Lu, Quanfeng, Ma, Zhantao, Zhong, Shuai, Wang, Jin, Yu, Dahai, Ng, Michael K., Luo, Ping
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.20018
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866912556546785280
author Lu, Quanfeng
Ma, Zhantao
Zhong, Shuai
Wang, Jin
Yu, Dahai
Ng, Michael K.
Luo, Ping
author_facet Lu, Quanfeng
Ma, Zhantao
Zhong, Shuai
Wang, Jin
Yu, Dahai
Ng, Michael K.
Luo, Ping
contents The rapid advancement of large vision language models (LVLMs) and agent systems has heightened interest in mobile GUI agents that can reliably translate natural language into interface operations. Existing single-agent approaches, however, remain limited by structural constraints. Although multi-agent systems naturally decouple different competencies, recent progress in multi-agent reinforcement learning (MARL) has often been hindered by inefficiency and remains incompatible with current LVLM architectures. To address these challenges, we introduce SWIRL, a staged workflow for interleaved reinforcement learning designed for multi-agent systems. SWIRL reformulates MARL into a sequence of single-agent reinforcement learning tasks, updating one agent at a time while keeping the others fixed. This formulation enables stable training and promotes efficient coordination across agents. Theoretically, we provide a stepwise safety bound, a cross-round monotonic improvement theorem, and convergence guarantees on return, ensuring robust and principled optimization. In application to mobile GUI control, SWIRL instantiates a Navigator that converts language and screen context into structured plans, and an Interactor that grounds these plans into executable atomic actions. Extensive experiments demonstrate superior performance on both high-level and low-level GUI benchmarks. Beyond GUI tasks, SWIRL also demonstrates strong capability in multi-agent mathematical reasoning, underscoring its potential as a general framework for developing efficient and robust multi-agent systems.
format Preprint
id arxiv_https___arxiv_org_abs_2508_20018
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle SWIRL: A Staged Workflow for Interleaved Reinforcement Learning in Mobile GUI Control
Lu, Quanfeng
Ma, Zhantao
Zhong, Shuai
Wang, Jin
Yu, Dahai
Ng, Michael K.
Luo, Ping
Artificial Intelligence
Computation and Language
Computer Vision and Pattern Recognition
Multiagent Systems
The rapid advancement of large vision language models (LVLMs) and agent systems has heightened interest in mobile GUI agents that can reliably translate natural language into interface operations. Existing single-agent approaches, however, remain limited by structural constraints. Although multi-agent systems naturally decouple different competencies, recent progress in multi-agent reinforcement learning (MARL) has often been hindered by inefficiency and remains incompatible with current LVLM architectures. To address these challenges, we introduce SWIRL, a staged workflow for interleaved reinforcement learning designed for multi-agent systems. SWIRL reformulates MARL into a sequence of single-agent reinforcement learning tasks, updating one agent at a time while keeping the others fixed. This formulation enables stable training and promotes efficient coordination across agents. Theoretically, we provide a stepwise safety bound, a cross-round monotonic improvement theorem, and convergence guarantees on return, ensuring robust and principled optimization. In application to mobile GUI control, SWIRL instantiates a Navigator that converts language and screen context into structured plans, and an Interactor that grounds these plans into executable atomic actions. Extensive experiments demonstrate superior performance on both high-level and low-level GUI benchmarks. Beyond GUI tasks, SWIRL also demonstrates strong capability in multi-agent mathematical reasoning, underscoring its potential as a general framework for developing efficient and robust multi-agent systems.
title SWIRL: A Staged Workflow for Interleaved Reinforcement Learning in Mobile GUI Control
topic Artificial Intelligence
Computation and Language
Computer Vision and Pattern Recognition
Multiagent Systems
url https://arxiv.org/abs/2508.20018