HyeonseokE/SO101-DistributeChoco_Ours_60epi
收藏资源简介:
该数据集是一个用于机器人技能学习的多模态数据集,由LeRobot框架创建。数据集包含60个训练集片段,总计29525帧,数据文件大小为100MB,视频文件大小为200MB,帧率为10fps。数据特征包括:机器人状态观测(6维关节位置)、动作(6维关节位置)、两个视角的图像观测(顶部和左腕摄像头,均为480x640分辨率RGB视频)、末端执行器位置(6维XYZRPY坐标)、抓取器二进制状态、技能和子任务的自然语言描述、技能类型、进度、目标位置(关节空间、笛卡尔空间和抓取器位置)、时间戳和索引等。数据集适用于机器人模仿学习或强化学习任务,支持多模态输入(如视觉和状态)到动作的映射。
This dataset is a multimodal dataset for robot skill learning, created using the LeRobot framework. It contains 60 training episodes, totaling 29,525 frames, with data files of 100MB and video files of 200MB, at a frame rate of 10fps. Features include: robot state observations (6-dimensional joint positions), actions (6-dimensional joint positions), image observations from two viewpoints (top and left wrist cameras, both 480x640 resolution RGB videos), end-effector position (6-dimensional XYZRPY coordinates), gripper binary state, natural language descriptions for skills and subtasks, skill type, progress, goal positions (joint space, Cartesian space, and gripper position), timestamps, and indices. The dataset is suitable for robot imitation learning or reinforcement learning tasks, supporting multimodal input (e.g., vision and state) to action mapping.




