CarlosMunoz0/vla-mujoco-so101-merged-v1

该数据集是使用LeRobot框架创建的机器人领域数据集，专门针对so101_mujoco类型的机器人。它包含305个episodes，总计297016帧，覆盖4个不同任务。数据以parquet文件格式存储，总数据文件大小为100MB，视频文件大小为200MB。数据集特征包括：动作（action）为6维浮点数组，表示机器人的关节位置（如肩部平移、肩部抬升、肘部弯曲、手腕弯曲、手腕旋转和夹爪位置）；状态观测（observation.state）同样为6维浮点数组，反映机器人的实时关节位置；图像观测来自两个摄像头：realsense和wrist_cam，均提供480x640分辨率、3通道的RGB视频，帧率为30fps，使用AV1编解码器；此外，还包括时间戳、帧索引、episode索引、索引和任务索引等元数据。所有数据仅用于训练分割，适用于机器人控制、视觉-语言-动作（VLA）等研究任务。This dataset is created using the LeRobot framework in the robotics domain, specifically for the so101_mujoco robot type. It contains 305 episodes with a total of 297016 frames, covering 4 distinct tasks. The data is stored in parquet format, with a total data file size of 100MB and video file size of 200MB. Dataset features include: action as a 6-dimensional float32 array representing robot joint positions (e.g., shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper positions); observation.state as a 6-dimensional float32 array reflecting real-time joint positions; image observations from two cameras: realsense and wrist_cam, both providing RGB video at 480x640 resolution with 3 channels, a frame rate of 30fps, using AV1 codec; additionally, metadata such as timestamp, frame index, episode index, index, and task index are included. All data is allocated for training splits, suitable for research tasks like robot control and vision-language-action (VLA) applications.