# Visual Para-Thinker++：一种用于视觉推理的单策略多智能体框架

- 来源：HuggingFace Daily Papers（社区热门论文）
- 发布时间：2026-06-08 08:00
- AIHOT 分数：55
- AIHOT 链接：https://aihot.virxact.com/items/cmqaef40q0kqkslldxox9eu6w
- 原文链接：https://arxiv.org/abs/2606.09290

## AI 摘要

Visual Para-Thinker++ 是一种单策略多智能体框架，将共享 MLLM 策略实例化为角色条件化的 Main、Worker 和 Summary Agent。Main Agent 按固定模式分解任务，Worker Agent 在上下文隔离下并行推理，Summary Agent 整合全部 Worker 推理轨迹而非对最终标签进行多数投票。共享策略通过多智能体能力注入和角色解耦多智能体优化训练，为对应 token 片段分配角色特定奖励和优势以减少梯度冲突。推理引擎通过共享视觉前缀和 KV cache 重用实现高效多智能体 rollout。在 V*、CountBench、RefCOCO 系列和 HallusionBench 上，该框架一致优于单轨迹和推理时并行基线，在幻觉敏感任务上增益尤为显著。

## 正文

Visual reasoning requires integrating evidence distributed across regions, attributes, and relations, making single-chain reasoning prone to early perceptual commitment and hallucination. We propose Visual Para-Thinker++, a single-policy multi-agent framework in which one shared MLLM policy is instantiated as role-conditioned Main, Worker, and Summary Agents. The Main Agent decomposes the task with fixed allocation patterns; Worker Agents reason in parallel under context isolation; and the Summary Agent reconciles full Worker reasoning traces rather than majority-voting on final labels. The shared policy is trained by Multi-Agent Capability Injection and Role-Decoupled Multi-Agent Optimization, which assign role-specific rewards and advantages to corresponding token segments to reduce gradient conflict among collaborative roles. A native inference engine enables efficient multi-agent rollout through shared visual prefix and KV cache reuse. Across V*, CountBench, the RefCOCO family, and HallusionBench, Visual Para-Thinker++ consistently outperforms single-trajectory and inference-time parallel baselines, with especially strong gains on hallucination-sensitive visual reasoning.