VideoChat-M1：通过多智能体强化学习实现面向视频理解的协作策略规划
VideoChat-M1: Collaborative Policy Planning for Video Understanding via Multi-Agent Reinforcement Learning
把视频理解建模为多个策略智能体各自生成、执行并相互通信更新工具调用策略的过程，并用最终答案奖励与中间协作反馈做联合强化学习优化；对做长视频/复杂时序推理的视觉研究者，这是一套可复用的「多智能体+工具调用+MARL」框架，可替代静态固定的工具链。
原文依据 we adopt a distinct Collaborative Policy Planning (CPP) paradigm with multiple policy agents
摘要译文
由模型翻译 · 术语统一 · 请对照原文核对关键结论现有多数多智能体视频理解框架采用静态且不可学习的工具调用机制，这限制了对多样化线索的发现，而这些线索对于针对时序或空间复杂视频的鲁棒感知与推理至关重要。为应对这一挑战，我们提出了一种新的视频理解多智能体系统，即 VideoChat-M1。我们不使用单一或固定的策略，而是采用一种独特的协作策略规划（CPP）范式，其中包含多个策略智能体，该范式包括三个关键过程。(1) 策略生成：每个智能体针对用户查询生成其独有的工具调用策略。(2) 策略执行：每个智能体依次调用相关工具来执行其策略并探索视频内容。(3) 策略通信：在策略执行的中间阶段，智能体之间相互交互以更新各自的策略。通过这一协作框架，所有智能体协同工作，基于同伴提供的上下文洞见动态地精炼各自偏好的策略。此外，我们为该 CPP 范式配备了多智能体强化学习（MARL）。因此，策略智能体可以被联合优化以提升性能，其优化同时受最终答案奖励和中间协作过程反馈的引导。大量实验表明，VideoChat-M1 在四项任务的八个基准上取得了 SOTA 性能。值得注意的是，在 LongVideoBench 上，我们的方法比 Gemini 2.5 pro 高出 3.6%，比 GPT-4o 高出 15.6%。
英文原文摘要 1531 字符
Most of the multi-agent video understanding frameworks adopt static and non-learnable tool invocation mechanisms, which limit the discovery of diverse clues essential for robust perception and reasoning regarding temporally or spatially complex videos. To address this challenge, we propose a novel Multi-agent system for video understanding, namely VideoChat-M1. Instead of using a single or fixed policy, we adopt a distinct Collaborative Policy Planning (CPP) paradigm with multiple policy agents, which comprises three key processes. (1)Policy Generation: Each agent generates its unique tool invocation policy tailored to the user's query. (2) Policy Execution: Each agent sequentially invokes relevant tools to execute its policy and explore the video content. (3) Policy Communication: During the intermediate stages of policy execution, agents interact with one another to update their respective policies. Through this collaborative framework, all agents work in tandem, dynamically refining their preferred policies based on contextual insights from peers. Moreover, we equip our CPP paradigm with Multi-Agent Reinforcement Learning (MARL). Consequently, policy agents can be jointly optimized to enhance the performance, guided by both the final answer reward and intermediate collaborative process feedback. Extensive experiments demonstrate that VideoChat-M1 achieves SOTA performance across eight benchmarks on four tasks. Notably, on LongVideoBench, our method outperforms Gemini 2.5 pro by 3.6% and GPT-4o by 15.6%.