WOMD-Normal

该数据集专为强化学习（RL）代理策略的闭环训练而设计，涉及自我代理和对抗代理。在训练过程中，采用了课程训练方法；代理通过模拟的激光雷达返回观察环境。该数据集的训练规模达到一百万时间步，任务是进行RL代理的闭环训练。

This dataset is specifically designed for closed-loop training of reinforcement learning (RL) agent policies, involving self-play agents and adversarial agents. A curriculum training approach is adopted during the training process, where agents observe the environment through simulated LiDAR returns. The training scale of this dataset reaches one million time steps, and the core task is closed-loop training for RL agents.