Robotic Manipulation Task Suite

该数据集包含了700个不同任务中成功的人类演示和失败的自主收集的环节，总计约58,000个训练用的环节。每个环节的平均时长为35个时间步长，且该数据集是通过13台机器人收集而成的。规模上，大约有58,000个环节，其中包括38,000个成功的演示和20,000个失败的环节。这项工作的任务是进行机器人操作的多任务强化学习。

This dataset contains successful human demonstrations and autonomously collected failed episodes across 700 distinct tasks, with a total of approximately 58,000 training episodes. The average duration of each episode is 35 timesteps, and the dataset was collected using 13 robots. Breaking down the scale, the 58,000 total training episodes consist of 38,000 successful demonstrations and 20,000 failed episodes. The core task of this work is multi-task reinforcement learning for robotic manipulation.