DyCo-RL: Dynamic Cross-Modal Coordination for Visual Reasoning
About
Reinforcement Learning with Verifiable Rewards (RLVR) has emerged as a leading paradigm for enhancing visual reasoning in Multimodal Large Language Models (MLLMs). However, existing RLVR methods optimize primarily for the reasoning outcome, fundamentally overlooking the fine-grained cross-modal coordination required during the generation process. Through token-level analyses and controlled interventions, we reveal that during Chain-of-Thought (CoT) reasoning, MLLMs frequently fail to dynamically alternate between extracting visual evidence and synthesizing textual context-a coordination breakdown that is causally linked to reasoning failures. Motivated by these findings, we propose DyCo-RL, which integrates dynamic cross-modal coordination into RLVR optimization. Specifically, DyCo-RL uses the Fisher-Rao geodesic distance to measure within-modality attention shifts, assigning tokens to either visually-oriented or text-oriented functional roles. It then evaluates the alignment between a token's actual attention allocation and its assigned role, leveraging this score for alignment-guided advantage reweighting during policy optimization. Extensive experiments demonstrate that the algorithm-agnostic DyCo-RL, when applied to Qwen2.5-VL-3B/7B, consistently improves four representative RLVR algorithms across seven benchmarks spanning visual-centric and mathematical reasoning.
Related benchmarks
| Task | Dataset | Result | Rank | |
|---|---|---|---|---|
| Visual Question Answering | RealworldQA | Accuracy49.5 | 327 | |
| Mathematical Reasoning | WeMath | Accuracy40.7 | 317 | |
| Mathematical Reasoning | MathVerse | -- | 266 | |
| Visual Question Answering | A-OKVQA | Acc82.7 | 240 | |
| Visual Question Answering | MMStar | Accuracy53.2 | 151 | |
| Visual Question Answering | ChartQA (test) | Accuracy83.2 | 95 | |
| Visual Question Answering | SEED-Bench Image | Accuracy69 | 80 | |
| Visual Reasoning | MMBench | -- | 48 | |
| Visual Reasoning | HallusionBench | Accuracy73.2 | 43 | |
| Mathematical Reasoning | MathVision | Top-1 Accuracy29.3 | 27 |