Flatland3 Challenge

该数据集专为多代理路径查找（MAPF）问题设计，包含了多个测试阶段。在这些阶段中，智能体需要穿越环境复杂度和规模逐渐增加的多种场景。该数据集支持课程学习，并允许在各种场景和规模下评估强化学习方法。数据集的规模范围从30x30地图上的7个代理，到158x158地图上的425个代理，其任务旨在解决多代理路径查找问题。

This dataset is specifically designed for the multi-agent path finding (MAPF) problem and includes multiple test phases. In these phases, agents are required to navigate through various scenarios with gradually increasing environmental complexity and scale. This dataset supports curriculum learning and enables the evaluation of reinforcement learning methods across diverse scenarios and scales. The scale of the dataset ranges from 7 agents on a 30×30 map to 425 agents on a 158×158 map, with the tasks aiming to solve the multi-agent path finding problem.