Habitat-Lab

该数据集是一个为在3D环境中训练和评估深度强化学习模型而设计的高性能模拟器，特别适用于身体人工智能任务。数据集包含了在评估过程中观察到的轨迹，并识别了机器人代理在无碰撞情况下跟随人形生物时的不安全情况。其规模适用于真实的深度强化学习导航任务，具体任务为社会性导航。

This dataset is a high-performance simulator dedicated to training and evaluating deep reinforcement learning models within 3D environments, and it is specifically designed for embodied artificial intelligence tasks. It contains trajectories collected during evaluation processes, and identifies unsafe scenarios where robotic agents follow humanoid creatures without triggering collisions. The scale of this dataset is suitable for real-world deep reinforcement learning navigation tasks, with the specific target task being social navigation.