HyeonseokE/SO101-SortBlock_Ours_40epi
收藏资源简介:
该数据集是使用LeRobot框架创建的机器人操作数据集,专门用于SO101机器人(so101_follower类型)执行排序块任务。数据集包含40个训练集片段,总计37972帧,数据以parquet格式存储,视频以MP4格式存储(总数据大小100MB,视频大小200MB),帧率为10fps。特征包括:观察状态(6个关节位置:肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转、夹爪位置)、动作(相同6个关节位置)、图像观察(顶部摄像头和左腕摄像头,分辨率480x640,RGB三通道)、末端执行器位置(机器人坐标系下的XYZRPY坐标)、夹爪二进制状态、技能信息(自然语言描述、验证问题、技能类型、进度、目标关节位置、目标末端执行器位置、目标夹爪位置)、子任务信息(自然语言描述、对象名称、目标位置XYZ坐标)、以及时间戳、帧索引、片段索引、任务索引等元数据。数据集适用于机器人学习、模仿学习或强化学习任务,特别是涉及多模态感知和操作的任务。
This dataset was created using LeRobot and contains robotic manipulation data for the SO101 robot (so101_follower type) performing a block sorting task. It includes 40 training episodes with a total of 37,972 frames, stored in parquet format for data and MP4 for videos (total data size 100MB, video size 200MB), at 10fps. Features encompass observation states (6 joint positions: shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, gripper), actions (same 6 joint positions), image observations (top and left wrist cameras, 480x640 resolution, RGB channels), end-effector position (XYZRPY in robot coordinates), gripper binary state, skill information (natural language description, verification question, skill type, progress, goal joint position, goal end-effector position, goal gripper position), subtask information (natural language description, object name, target XYZ position), and metadata such as timestamp, frame index, episode index, and task index. The dataset is suitable for robotics learning, imitation learning, or reinforcement learning tasks, particularly those involving multimodal perception and manipulation.




