guloyy/SO101-eval2-grb-expanded-v3-256

该数据集是一个用于机器人任务的数据集，专门针对SO101评估2 GRB扩展版本v3-256。它包含300个完整的情节，总帧数为135,640帧，覆盖164个不同任务。数据以30帧每秒（fps）的帧率采集，总数据文件大小为100MB，视频文件大小为200MB。数据集采用Apache 2.0许可证。数据集结构包括训练分割（0:300），数据以Parquet格式存储，视频以MP4格式存储。特征包括动作（6个浮点型关节位置：肩部平移、肩部提升、肘部弯曲、腕部弯曲、腕部旋转和夹爪位置）、观测状态（与动作相同的6个关节位置）、观测图像（来自前摄像头的256x256x3视频，使用AV1编解码器）、时间戳、帧索引、集索引、索引和任务索引。该数据集适用于机器人控制、模仿学习或强化学习等任务。This dataset is designed for robotics tasks, specifically for the SO101 evaluation 2 GRB expanded version v3-256. It includes 300 total episodes, with a total of 135,640 frames, covering 164 distinct tasks. The data is collected at a frame rate of 30 frames per second (fps), with a total data file size of 100MB and video file size of 200MB. The dataset is licensed under Apache 2.0. The dataset structure includes a training split (0:300), with data stored in Parquet format and videos in MP4 format. Features include action (6 float32 joint positions: shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper position), observation state (same 6 joint positions as action), observation images (front camera video of size 256x256x3, using AV1 codec), timestamp, frame index, episode index, index, and task index. This dataset is suitable for tasks such as robot control, imitation learning, or reinforcement learning.