TynClause/cvit-task-clean-table-100E

该数据集是一个机器人控制数据集，使用LeRobot框架创建，专门用于机器人学习任务。数据集包含114个episodes，总计110,817帧，涉及2个不同的任务。数据以30帧每秒（fps）的速率采集，特征包括动作（action）和观测（observation）数据：动作由6个关节位置（如肩部平移、肩部提升、肘部弯曲、手腕弯曲、手腕旋转、夹爪位置）组成，观测状态包含相同的6个关节位置。此外，观测还包括两个摄像头图像：正面摄像头（front）和手腕摄像头（wrist），每个图像分辨率为480x640像素，3个通道（RGB），以AV1编码的MP4视频格式存储。数据集还包含时间戳、帧索引、episode索引、任务索引等元数据。机器人类型为so_follower，数据以parquet文件（用于结构化数据）和mp4视频文件（用于图像数据）分块存储，总数据大小约为100MB（结构化数据）和200MB（视频数据）。数据集适用于机器人控制、强化学习或计算机视觉任务的研究和开发。

This dataset is a robotics control dataset created using the LeRobot framework, specifically designed for robot learning tasks. It contains 114 episodes, totaling 110,817 frames, and involves 2 distinct tasks. Data is collected at a rate of 30 frames per second (fps), with features including action and observation data: actions consist of 6 joint positions (e.g., shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, gripper position), and observation states include the same 6 joint positions. Additionally, observations include two camera images: a front camera and a wrist camera, each with a resolution of 480x640 pixels, 3 channels (RGB), stored as MP4 videos encoded with AV1. The dataset also includes metadata such as timestamps, frame indices, episode indices, and task indices. The robot type is so_follower, and data is stored in chunked parquet files (for structured data) and MP4 video files (for image data), with total sizes of approximately 100MB (structured data) and 200MB (video data). This dataset is suitable for research and development in robotics control, reinforcement learning, or computer vision tasks.