DAVIAN-Robotics/0528_bk_config2
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot框架创建。数据集包含30个训练片段(episodes),总计7942帧,覆盖9个不同的任务。数据以parquet文件和视频文件形式存储,总数据量约为100MB,视频文件约为200MB。数据集采集自Franka机器人,帧率为20 fps。观测特征包括:29维的机器人状态向量(包含7个关节位置、7个关节速度、末端执行器的3D位置、四元数朝向、线速度、角速度、夹爪位置和抓取状态);三个相机视角的图像观测(手腕、左后、右后),每个均为480x640分辨率的RGB视频;以及时间戳、帧索引、片段索引等元数据。动作空间为8维向量,表示末端执行器的目标位置、朝向和夹爪控制。数据集适用于机器人学习任务,如模仿学习或强化学习。
This dataset is a robotic manipulation dataset developed using the LeRobot framework. It consists of 30 training episodes, totaling 7942 frames, spanning 9 distinct tasks. The data is stored in both parquet files and video files, with a total volume of approximately 100MB for non-video data and ~200MB for video files. The dataset was collected using a Franka robot, with a capture frame rate of 20 fps. Observation features include: a 29-dimensional robot state vector, which encompasses 7 joint positions, 7 joint velocities, the 3D position and quaternion orientation of the end effector, linear velocity, angular velocity, gripper position and grasp status; image observations from three camera perspectives (wrist, left rear, right rear), all of which are RGB videos with a resolution of 480×640; as well as metadata such as timestamps, frame indices, and episode indices. The action space is an 8-dimensional vector that represents the target position, orientation of the end effector, and gripper control commands. This dataset is applicable to robotic learning tasks such as imitation learning and reinforcement learning.



