Saved in:
Bibliographic Details
Main Authors: Yan, Wentao, Wang, Shengqin, Zhou, Huichi, Chen, Yihang, Shao, Kun, Xie, Yuan, Zhang, Zhizhong
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2604.20486
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914498824110080
author Yan, Wentao
Wang, Shengqin
Zhou, Huichi
Chen, Yihang
Shao, Kun
Xie, Yuan
Zhang, Zhizhong
author_facet Yan, Wentao
Wang, Shengqin
Zhou, Huichi
Chen, Yihang
Shao, Kun
Xie, Yuan
Zhang, Zhizhong
contents Training multimodal agents via reinforcement learning for knowledge-intensive visual reasoning is fundamentally hindered by the extreme sparsity of outcome-based supervision and the unpredictability of live web environments. To resolve these algorithmic and environmental bottlenecks, we introduce ProMMSearchAgent, establishing a novel Sim-to-Real training paradigm for multimodal search. We decouple policy learning into a deterministic, local static sandbox. Crucially, to learn effectively within this constrained environment, we propose an introspective process-oriented reward. By probing the agent's own parametric knowledge boundaries, we generate dense behavioral metadata that explicitly rewards the correct cognitive decision, initiating a multimodal or text search only when visually or factually uncertain. Extensive experiments demonstrate that our locally-trained policy transfers zero-shot to the live Google Search API. ProMMSearchAgent achieves new SOTA performance, outperforming MMSearch-R1 by +5.1% on FVQA-test, +6.3% on InfoSeek, and +11.3% on MMSearch.
format Preprint
id arxiv_https___arxiv_org_abs_2604_20486
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle ProMMSearchAgent: A Generalizable Multimodal Search Agent Trained with Process-Oriented Rewards
Yan, Wentao
Wang, Shengqin
Zhou, Huichi
Chen, Yihang
Shao, Kun
Xie, Yuan
Zhang, Zhizhong
Computer Vision and Pattern Recognition
Training multimodal agents via reinforcement learning for knowledge-intensive visual reasoning is fundamentally hindered by the extreme sparsity of outcome-based supervision and the unpredictability of live web environments. To resolve these algorithmic and environmental bottlenecks, we introduce ProMMSearchAgent, establishing a novel Sim-to-Real training paradigm for multimodal search. We decouple policy learning into a deterministic, local static sandbox. Crucially, to learn effectively within this constrained environment, we propose an introspective process-oriented reward. By probing the agent's own parametric knowledge boundaries, we generate dense behavioral metadata that explicitly rewards the correct cognitive decision, initiating a multimodal or text search only when visually or factually uncertain. Extensive experiments demonstrate that our locally-trained policy transfers zero-shot to the live Google Search API. ProMMSearchAgent achieves new SOTA performance, outperforming MMSearch-R1 by +5.1% on FVQA-test, +6.3% on InfoSeek, and +11.3% on MMSearch.
title ProMMSearchAgent: A Generalizable Multimodal Search Agent Trained with Process-Oriented Rewards
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.20486