Unmanned warehouses are an important part of logistics, and improving their operational efficiency can effectively enhance service efficiency. However, due to the complexity of unmanned warehouse systems and their susceptibility to errors, incidents may occur during their operation, most often in inbound and outbound operations, which can decrease operational efficiency. Hence it is crucial to to improve the response to such incidents. This paper proposes a collaborative optimization algorithm for emergent incident response based on Safe-MADDPG. To meet safety requirements during emergent incident response, we investigated the intrinsic hidden relationships between various factors. By obtaining constraint information of agents during the emergent incident response process and of the dynamic environment of unmanned warehouses on agents, the algorithm reduces safety risks and avoids the occurrence of chain accidents; this enables an unmanned system to complete emergent incident response tasks and achieve its optimization objectives: (1) minimizing the losses caused by emergent incidents; and (2) maximizing the operational efficiency of inbound and outbound operations during the response process. A series of experiments conducted in a simulated unmanned warehouse scenario demonstrate the effectiveness of the proposed method.

翻译：无人仓库是物流的重要组成部分，提高其运行效率可有效提升服务效能。然而，由于无人仓库系统的复杂性和易错性，运行过程中可能出现异常事件，其中以出入库操作最为常见，这会降低运行效率。因此，改进对这类事件的响应至关重要。本文提出一种基于Safe-MADDPG的应急响应协同优化算法。为满足应急响应过程中的安全需求，我们研究了各因素之间的内在隐含关系。通过获取应急响应过程中智能体的约束信息以及无人仓库动态环境对智能体的约束，该算法降低了安全风险，避免了连锁事故的发生；这使得无人系统能够完成应急响应任务并实现其优化目标：（1）最小化应急事件造成的损失；（2）最大化响应过程中出入库操作的运行效率。在模拟无人仓库场景下进行的一系列实验验证了所提方法的有效性。