HyeonseokE/SO101-Openlid_Ours_80epi
收藏资源简介:
该数据集是一个机器人控制数据集,使用LeRobot创建,专门针对so101_follower机器人类型。数据集包含80个episodes,总帧数为26880,帧率为10fps。数据以parquet文件格式存储,视频文件以mp4格式存储。数据集特征丰富,包括观察状态(如关节位置:肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部滚动和夹爪位置)、动作(相同关节位置控制)、图像观察(顶部和左腕摄像头的视频,分辨率为480x640,RGB三通道)、末端执行器位置(XYZ坐标和欧拉角)、夹爪二进制状态、技能信息(自然语言描述、验证问题、类型、进度、目标位置关节和末端执行器坐标)、子任务信息(自然语言描述、对象名称、目标位置)、以及时间戳、帧索引、episode索引等元数据。数据集设计用于机器人学习任务,如模仿学习或强化学习,支持多模态输入(状态和图像)和高级技能表示。
This dataset is a robotics control dataset created using LeRobot, specifically for the so101_follower robot type. It contains 80 episodes with a total of 26,880 frames at a frame rate of 10fps. The data is stored in parquet file format, and video files are stored in mp4 format. The dataset features include observation states (e.g., joint positions: shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions), actions (same joint position controls), image observations (videos from top and left wrist cameras with 480x640 resolution and RGB channels), end-effector positions (XYZ coordinates and Euler angles), gripper binary state, skill information (natural language descriptions, verification questions, types, progress, goal positions in joint and end-effector coordinates), subtask information (natural language descriptions, object names, target positions), and metadata such as timestamps, frame indices, episode indices. The dataset is designed for robotics learning tasks, such as imitation learning or reinforcement learning, supporting multimodal inputs (state and images) and high-level skill representations.




