antmaze-large

该数据集对不使用任何奖励信息的交互式算法提出了挑战，它要求学习者控制一个四足机器人，在迷宫环境中导航至目标位置。该数据集已经用于评估HyPE和HyPER算法与其他逆强化学习方法的样本效率，其规模较大，任务是在迷宫环境中进行逆强化学习。

This dataset poses a challenge to interactive algorithms that do not utilize any reward information. It requires the learning agent to control a quadruped robot to navigate to a target position within a maze environment. This large-scale dataset has been employed to evaluate the sample efficiency of HyPE, HyPER and other inverse reinforcement learning methods, with its core task focusing on inverse reinforcement learning in maze environments.