kjh5678688/cup_and_cube

这是一个由LeRobot创建的机器人数据集，专为双手机器人（bi_so_follower）设计。数据集包含100个episodes，总计119,600帧数据，帧率为30fps。数据以parquet文件（总大小100MB）和mp4视频文件（总大小200MB）存储，涵盖训练任务（splits为train）。数据集特征包括：动作数据（action，12维浮点数组，表示左右机械臂的关节位置和夹爪位置）、状态观测（observation.state，12维浮点数组，与动作结构相同）、视频观测（来自三个摄像头的视频流：left_front_l、left_person_view和right_front_r，分辨率均为480x640，3通道RGB，编码为av1），以及时间戳、帧索引、episode索引、任务索引等元数据。该数据集适用于机器人控制、模仿学习或强化学习研究。This is a robotics dataset created using LeRobot, designed for a dual-arm robot (bi_so_follower). It contains 100 episodes with a total of 119,600 frames at 30fps. The data is stored in parquet files (total size 100MB) and mp4 video files (total size 200MB), covering training tasks (splits: train). Dataset features include: action data (12-dimensional float32 array representing joint positions and gripper positions for left and right arms), state observations (observation.state, 12-dimensional float32 array, same structure as action), video observations (from three camera streams: left_front_l, left_person_view, and right_front_r, all at 480x640 resolution, 3-channel RGB, encoded as av1), and metadata such as timestamp, frame index, episode index, and task index. This dataset is suitable for robotics control, imitation learning, or reinforcement learning research.