Metric-AI/so_101_chocolate_merged_dedup
收藏资源简介:
该数据集是一个机器人数据集,使用LeRobot工具创建,专门用于机器人任务。数据集包含70个总剧集和18849个总帧数,数据以parquet文件格式存储,总数据文件大小为100MB,视频文件大小为200MB。特征包括动作数据(如机器人关节位置:肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部滚动和夹爪位置)、观察状态(与动作相同的关节位置)、观察图像(包括固定摄像头和手眼摄像头的视频数据,分辨率为480x640,3通道,帧率为30fps,使用av1编解码器),以及时间戳、帧索引、剧集索引、索引和任务索引等元数据。机器人类型为“so_follower”,任务类别为robotics,适用于训练和评估机器人控制模型。数据集结构支持分块处理,块大小为1000,并包含训练分割(剧集0到70)。
This is a robotics dataset developed using the LeRobot toolkit, tailored specifically for robotics tasks. It comprises a total of 70 episodes and 18849 frames, stored in Parquet file format. The total size of the data files is 100 MB, while the total size of the video files is 200 MB. The dataset includes the following core features: 1. Action data: Robot joint position commands including shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist roll, and gripper position 2. Observation states: Identical joint position measurements as those in the action data 3. Observation images: Video streams captured by both fixed cameras and eye-in-hand cameras, with a resolution of 480×640, 3 RGB channels, a frame rate of 30 fps, and encoded with the AV1 codec 4. Metadata: Timestamps, frame indices, episode indices, sample indices, and task indices The target robot type for this dataset is "so_follower", with the task category classified as robotics. It is suitable for training and evaluating robot control models. The dataset structure supports chunked processing with a chunk size of 1000, and includes the training split covering episodes 0 to 70.




