多模态智能体推理的智能体探索性策略优化

摘要

具备扩展推理能力的视觉-语言模型能够解决复杂问题，但许多现实世界的问题需要依赖外部工具，而仅靠内部推理往往无法解决。因此，智能体推理将两种行为以结构性不对称的方式交织在一起：思考（默认的自包含行为）与工具使用（一种高方差辅助性行为）。我们将这种不对称称为“思考-行动鸿沟”。在诸如GRPO等标准强化学习策略下，这一鸿沟会在训练中表现为两个诊断性症状：工具使用仅出现在约30%的展开中，而当尝试使用时，分组中约40%的问题对应的工具使用展开全部错误，从而抑制了本应作用于工具调用的学习信号。我们提出AXPO（智能体探索式策略优化）：对于每个全部错误的工具使用子组，AXPO固定思考前缀，重新采样工具调用及其后续内容，并辅以基于不确定性的前缀选择。在九个多模态基准测试和三个规模的Qwen3-VL-Thinking模型上，SFT+AXPO在平均性能上优于SFT+GRPO（8B模型平均Pass@1提升1.8个百分点，Pass@4提升1.8个百分点），并且8B规模的SFT+AXPO在Pass@4上超越了32B基础模型，参数量仅为后者的四分之一。

English

Vision-language models with extended reasoning succeed on complex problems, but many real-world problems require external tools that internal reasoning alone often cannot resolve. Agentic reasoning therefore interleaves two behaviors with a structural asymmetry: thinking (the self-contained default) and tool use (a high-variance auxiliary acting). We refer to this asymmetry as the Thinking-Acting Gap. Under standard RL recipes like GRPO, the gap manifests as two diagnostic symptoms during training: tool use is attempted on only ~30% of rollouts, and when attempted, the tool-using rollouts within a group are all-wrong on ~40% of questions, suppressing the learning signal at the tool calls that needed it. We propose AXPO (Agent eXplorative Policy Optimization): for each all-wrong tool-using subgroup, AXPO fixes the thinking prefix and resamples the tool call and its continuation, paired with uncertainty-based prefix selection. Across nine multimodal benchmarks and three scales of Qwen3-VL-Thinking, SFT+AXPO outperforms SFT+GRPO at average (+1.8pp Pass@1 and +1.8pp Pass@4 at 8B on average) and 8B with SFT+AXPO surpasses the 32B Base on Pass@4 with 4 times fewer parameters.