so101_pick_place_items
收藏资源简介:
该数据集使用LeRobot平台创建,是一个机器人操作领域的演示数据集,包含机器人执行任务时的多模态记录。数据集由251个独立任务片段组成,总计142,681帧数据,涵盖3种不同的任务类型。数据采用结构化格式存储,主要字段包括:动作指令(控制机器人6个关节的位置,包括肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转和夹爪);观测数据(包括机器人相同的6关节状态,以及来自腕部摄像头和前置摄像头的视觉图像,图像分辨率为640x480,RGB三通道);此外还包含时间戳、帧索引、片段索引、全局索引和任务索引等元数据。所有数据以30帧/秒的速率采集,视频数据采用AV1编码。数据集专门用于训练和评估机器人模仿学习或强化学习模型,特别是涉及视觉感知和关节控制的抓取与放置类任务。数据集仅提供训练集划分,使用的机器人平台类型为so_follower。
This dataset, developed using the LeRobot platform, is a demonstration dataset for the field of robotic manipulation, containing multimodal recordings of robots performing tasks. It comprises 251 independent task segments, with a total of 142,681 frames of data covering 3 distinct task types. The data is stored in a structured format, with its core fields including: - Action commands: Used to control the positions of the robot's six joints, specifically shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and the gripper; - Observational data: Includes the robot's six-joint states, as well as visual images captured by the wrist camera and front-facing camera. The images have a resolution of 640×480 with three RGB channels; - Additional metadata such as timestamps, frame indices, segment indices, global indices, and task indices. All data is collected at a rate of 30 frames per second, and the video data is encoded using the AV1 codec. This dataset is specifically intended for training and evaluating robotic imitation learning or reinforcement learning models, particularly grasping and placing tasks that involve visual perception and joint control. The dataset only provides a training set split, and the robotic platform utilized is so_follower.




