Walk the Talk: Bridging the Reasoning-Action Gap for Thinking with Images via Multimodal Agentic Policy Optimization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Yang, Wenhao, Xia, Yu, Huang, Jinlong, Lu, Shiyin, Chen, Qing-Guo, Xu, Zhao, Luo, Weihua, Zhang, Kaifu, Zhou, Yuchen, Xia, Xiaobo, Wan, Yuanyu, Zhang, Lijun, Chua, Tat-Seng
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911575841964032
author Yang, Wenhao
Xia, Yu
Huang, Jinlong
Lu, Shiyin
Chen, Qing-Guo
Xu, Zhao
Luo, Weihua
Zhang, Kaifu
Zhou, Yuchen
Xia, Xiaobo
Wan, Yuanyu
Zhang, Lijun
Chua, Tat-Seng
author_facet Yang, Wenhao
Xia, Yu
Huang, Jinlong
Lu, Shiyin
Chen, Qing-Guo
Xu, Zhao
Luo, Weihua
Zhang, Kaifu
Zhou, Yuchen
Xia, Xiaobo
Wan, Yuanyu
Zhang, Lijun
Chua, Tat-Seng
contents Recent advancements in Multimodal Large Language Models (MLLMs) have incentivized models to ``think with images'' by actively invoking visual tools during multi-turn reasoning. The common Reinforcement Learning (RL) practice of relying on outcome-based rewards ignores the fact that textual plausibility often masks executive failure, meaning that models may exhibit intuitive textual reasoning while executing imprecise or irrelevant visual actions within their agentic reasoning trajectories. This reasoning-action discrepancy introduces noise that accumulates throughout the multi-turn reasoning process, severely degrading the model's multimodal reasoning capabilities and potentially leading to training collapse. In this paper, we introduce Multimodal Agentic Policy Optimization (MAPO), bridging the gap between textual reasoning and visual actions generated by models within their Multimodal Chain-of-Thought (MCoT). Specifically, MAPO mandates the model to generate explicit textual descriptions for the visual content obtained via tool usage. We then employ a novel advantage estimation that couples the semantic alignment between these descriptions and the actual observations with the task reward. Theoretical findings are provided to justify the rationale behind MAPO, which inherently reduces the variance of gradients, and extensive experiments demonstrate that our method achieves superior performance across multiple visual reasoning benchmarks.
format Preprint
id arxiv_https___arxiv_org_abs_2604_06777
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Walk the Talk: Bridging the Reasoning-Action Gap for Thinking with Images via Multimodal Agentic Policy Optimization
Yang, Wenhao
Xia, Yu
Huang, Jinlong
Lu, Shiyin
Chen, Qing-Guo
Xu, Zhao
Luo, Weihua
Zhang, Kaifu
Zhou, Yuchen
Xia, Xiaobo
Wan, Yuanyu
Zhang, Lijun
Chua, Tat-Seng
Computer Vision and Pattern Recognition
Recent advancements in Multimodal Large Language Models (MLLMs) have incentivized models to ``think with images'' by actively invoking visual tools during multi-turn reasoning. The common Reinforcement Learning (RL) practice of relying on outcome-based rewards ignores the fact that textual plausibility often masks executive failure, meaning that models may exhibit intuitive textual reasoning while executing imprecise or irrelevant visual actions within their agentic reasoning trajectories. This reasoning-action discrepancy introduces noise that accumulates throughout the multi-turn reasoning process, severely degrading the model's multimodal reasoning capabilities and potentially leading to training collapse. In this paper, we introduce Multimodal Agentic Policy Optimization (MAPO), bridging the gap between textual reasoning and visual actions generated by models within their Multimodal Chain-of-Thought (MCoT). Specifically, MAPO mandates the model to generate explicit textual descriptions for the visual content obtained via tool usage. We then employ a novel advantage estimation that couples the semantic alignment between these descriptions and the actual observations with the task reward. Theoretical findings are provided to justify the rationale behind MAPO, which inherently reduces the variance of gradients, and extensive experiments demonstrate that our method achieves superior performance across multiple visual reasoning benchmarks.
title Walk the Talk: Bridging the Reasoning-Action Gap for Thinking with Images via Multimodal Agentic Policy Optimization
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2604.06777