Dynamic Attention Allocation Environment

该数据集模拟了一个环境，其中多个代理在不同阶段（正常、风险、事故）监控多个仓库，以预防潜在的事故。这一设计灵感来自于食品检查和害虫控制等应用。该环境支持对处于不同阶段的仓库进行监控，并根据代理定义的奖励函数对事故进行处罚。规模上，包括5个仓库，每个仓库有3种状态，以及多个代理。任务是在不同投票规则下的策略聚合与评估。

This dataset simulates an environment where multiple agents monitor multiple warehouses across three distinct stages: normal, risk, and accident to prevent potential accidents. This design is inspired by applications such as food inspection and pest control. The environment supports monitoring of warehouses in different stages and imposes penalties for accidents based on reward functions defined by the agents. In terms of scale, it includes 5 warehouses, each with 3 possible states, plus multiple agents. The core task is policy aggregation and evaluation under different voting rules.