ember-lab-berkeley/robocasa365-pretrain-mg
收藏资源简介:
该数据集名为Pretraining (MimicGen) — atomic,是RoboCasa365集合的一部分,主要用于机器人领域的预训练任务。数据集包含60个原子任务的模拟生成滚动数据,每个任务大约有10,000个演示,总计1,615小时的数据。这些数据是通过脚本增强从人类演示中生成的。数据集的具体统计信息包括536,030个片段,116,246,439帧(20 fps),720个自然语言表述的任务(基于60个RoboCasa任务类)。数据集还详细描述了摄像头的配置、机器人的类型以及加载数据集的方法。
This dataset, named Pretraining (MimicGen) — atomic, is part of the RoboCasa365 collection, and is primarily used for pre-training tasks in the robotics field. It contains simulated generated rollout data for 60 atomic tasks, with approximately 10,000 demonstrations per task, totaling 1,615 hours of data. These data are generated from human demonstrations via script augmentation. Specific statistical information of the dataset includes 536,030 segments, 116,246,439 frames (20 fps), and 720 natural language task descriptions based on 60 RoboCasa task categories. The dataset also provides detailed descriptions of camera configurations, robot types, and the method for loading the dataset.




