资源简介:
该数据集是基于LeRobot框架创建的机器人操作数据集,记录了so_follower型机器人在单一任务上的演示数据。数据集包含13个episode,共计13419帧图像和对应的状态动作序列。每个时间步提供以下特征:从外部摄像头采集的RGB图像(480x640x3)和深度图像(480x640x1,单位毫米),以及从腕部摄像头采集的RGB图像和深度图像。同时记录机器人6维关节状态(包括肩部、肘部、腕部、夹爪位置)和对应的6维动作指令。数据以parquet文件格式存储,附带时间戳、帧索引、episode索引和任务索引。该数据集适用于机器人模仿学习、行为克隆、离线强化学习等研究任务。
This dataset is a robot manipulation dataset created based on the LeRobot framework, recording demonstration data of the so_follower robot on a single task. The dataset contains 13 episodes, totaling 13,419 frames of images and corresponding state-action sequences. Each timestep provides the following features: RGB images (480x640x3) and depth images (480x640x1, in millimeters) from an external camera, as well as RGB and depth images from a wrist camera. It also records the robots 6-dimensional joint states (including shoulder, elbow, wrist, and gripper positions) and corresponding 6-dimensional action commands. The data is stored in parquet file format, accompanied by timestamps, frame indices, episode indices, and task indices. This dataset is suitable for research tasks such as robot imitation learning, behavioral cloning, and offline reinforcement learning.