Multi-role agents improve collaboration for multimodal tool use
MRC-MMAgent: Reinforcement Learning for Multi-Role Collaboration in Multimodal Tool-Using Agents
Computer Vision and Pattern Recognition
Summary
Some AI models can combine pictures and words but still struggle with hard questions needing lots of knowledge and checking. This paper introduces a new way for AI agents to work together like humans do—looking up information, double-checking it, and correcting themselves. The authors show that their approach helps these agents make better decisions and answers by letting multiple roles cooperate and learn from the whole process. Their method works better than previous ones on several tests without needing more training data.
What this means in practice
- •For software engineers: Develop AI assistants that can access external tools and improve answers by revising their reasoning steps collaboratively.
- •For automated customer support teams: Enhance support bots to better retrieve and verify information from multiple sources before responding to users' complex queries.
Authors
Shengqin Wang, Jie Jin, Xiaofan Li, Wentao Yan, Yu Cheng, Qingsong HuYihang Chen, Weilin Luo, Jingyu Gong, Xin Tan, Yuan Xie, Zhizhong Zhang
Abstract
Multimodal tool-using agents augment vision-language models with access to external information and capabilities, enabling them to solve knowledge-intensive and visually complex tasks that cannot be addressed using visual context alone. However, existing reinforcement learning pipelines largely train single-role agents to search and answer using terminal rewards, overlooking how humans inspect retrieved evidence, reformulate queries, and revise conclusions during search. Consequently, they fail to optimize this self-corrective process over complete interaction trajectories. We propose MRC-MMAgent, a multi-role collaborative reinforcement learning framework that incorporates trajectory assessment and continuation directly into the executable agent loop. First, the framework establishes a multi-role collaborative loop that integrates solving, inspection, and continuation, enabling the agent to acquire evidence through tools, evaluate the preceding trajectory, and resume problem solving when the current answer is insufficiently supported. Second, we introduce a role-coupled learning objective that combines task success, interaction validity, and successful correction with full-trajectory policy optimization across the decisions generated by all roles. Extensive experiments show that, using only the original baseline training set without constructing additional datasets, our method consistently outperforms the corresponding base models and achieves leading or highly competitive performance across benchmarks.