yinongh/adaptation_1
收藏资源简介:
该数据集是一个机器人操作数据集,使用LeRobot工具创建。数据集包含20个训练集,总帧数2473,涉及1个任务(可能为插头插入操作),包含60个视频,帧率为30fps。数据以parquet格式存储,结构包括动作特征(如末端执行器的位置和旋转变化)和丰富的观测特征,例如末端执行器姿态、深度图像(如对齐的插座深度图像和初始插头深度图像)、点云数据(初始手腕点云)、手腕相机和辅助相机的RGB与深度图像,以及时间戳、帧索引等元数据。这些多模态数据适用于机器人学习任务,如模仿学习或强化学习。
This dataset is a robotics manipulation dataset created using the LeRobot tool. It contains 20 training episodes, with a total of 2473 frames, involving 1 task (likely a plug insertion operation), and includes 60 videos at 30fps. The data is stored in parquet format, featuring action data (e.g., end-effector position and rotation changes) and rich observation features, such as end-effector pose, depth images (e.g., aligned socket depth images and initial plug depth images), point cloud data (initial wrist points), RGB and depth images from wrist and auxiliary cameras, along with metadata like timestamps and frame indices. This multimodal data is suitable for robotics learning tasks, such as imitation learning or reinforcement learning.




