Visually-Guided Policy Optimization for Multimodal Reasoning
This work addresses a specific bottleneck in multimodal reasoning for vision-language models, offering an incremental improvement over existing reinforcement learning methods.
The paper tackles the problem of insufficient visual faithfulness in vision-language models during multimodal reasoning, where sparse attention to visual tokens and temporal visual forgetting degrade performance. The proposed Visually-Guided Policy Optimization (VGPO) framework improves visual activation and achieves superior results in mathematical multimodal reasoning and visual-dependent tasks.
Reinforcement learning with verifiable rewards (RLVR) has significantly advanced the reasoning ability of vision-language models (VLMs). However, the inherent text-dominated nature of VLMs often leads to insufficient visual faithfulness, characterized by sparse attention activation to visual tokens. More importantly, our empirical analysis reveals that temporal visual forgetting along reasoning steps exacerbates this deficiency. To bridge this gap, we propose Visually-Guided Policy Optimization (VGPO), a novel framework to reinforce visual focus during policy optimization. Specifically, VGPO initially introduces a Visual Attention Compensation mechanism that leverages visual similarity to localize and amplify visual cues, while progressively elevating visual expectations in later steps to counteract visual forgetting. Building on this mechanism, we implement a dual-grained advantage re-weighting strategy: the intra-trajectory level highlights tokens exhibiting relatively high visual activation, while the inter-trajectory level prioritizes trajectories demonstrating superior visual accumulation. Extensive experiments demonstrate that VGPO achieves better visual activation and superior performance in mathematical multimodal reasoning and visual-dependent tasks.