Bperju/so101_cube_pick_place_50_20260530_161044
收藏资源简介:
该数据集是一个机器人操作数据集,专注于立方体拾取和放置任务。它包含17个episodes,总计6883帧数据,采样频率为30帧/秒。数据集记录了机械臂(机器人类型为so_follower)的动作和观察状态,动作包括6个关节位置(如肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置),观察状态包括相同的关节位置信息。此外,数据集提供了来自顶部摄像头和腕部摄像头的视频观察,视频分辨率为480x640像素,编码为AV1格式,无音频。数据以Parquet文件格式存储,并包含时间戳、帧索引、episode索引、索引和任务索引等元数据。数据集用于训练和评估机器人学习模型,适用于强化学习或模仿学习任务。
This dataset is a robotic manipulation dataset focused on cube pick-and-place tasks. It includes 17 episodes with a total of 6883 frames, sampled at 30 frames per second (FPS). The dataset records the actions and observation states of the robotic arm, where the robot type is so_follower. The actions cover 6 joint positions: shoulder translation, shoulder lift, elbow flexion, wrist flexion, wrist rotation, and gripper position, while the observation states contain the same joint position information. Furthermore, the dataset provides video observations from the top-down camera and the wrist-mounted camera, with a resolution of 480×640 pixels, encoded in AV1 format, and no audio tracks. The data is stored in Parquet file format, and contains metadata including timestamps, frame indices, episode indices, indices, and task indices. This dataset is used for training and evaluating robotic learning models, and is suitable for reinforcement learning or imitation learning tasks.




