2026-04-27direction_have_cap_water-lerobot-without-rinse
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,专门用于机器人学习任务。数据集包含一个双机械臂机器人(类型为aloha)执行操作任务时记录的多模态数据,涵盖34个完整的操作序列,总计39,366个数据帧,所有数据均用于训练。数据以50fps的帧率记录,总数据量约为200MB。特征包括:观测状态为14维关节位置向量(对应左、右机械臂的7个关节:腰部、肩部、肘部、前臂旋转、腕部角度、腕部旋转、夹爪);动作指令为14维关节控制向量,结构与观测状态相同;此外还有14维关节速度观测和14维关节力矩观测。视觉观测部分包含来自四个摄像头的视频流:高位摄像头、低位摄像头、左腕部摄像头和右腕部摄像头,所有视频均为RGB格式(3通道),分辨率480x640,使用AV1编码。元数据包括任务索引、子任务描述字符串、训练标志、时间戳、帧索引、序列索引和全局索引。该数据集适用于机器人模仿学习、强化学习、多模态感知与控制策略学习等研究。
This dataset is a robot manipulation dataset created using the LeRobot tool, specifically designed for robot learning tasks. It contains multimodal data recorded during operation tasks performed by a dual-arm robot (type aloha), comprising 34 complete episodes with a total of 39,366 data frames, all used for training. The data is recorded at 50fps, with a total data volume of approximately 200MB. Features include: observation states as 14-dimensional joint position vectors (corresponding to 7 joints per arm: waist, shoulder, elbow, forearm rotation, wrist angle, wrist rotation, gripper); action commands as 14-dimensional joint control vectors with the same structure; additionally, 14-dimensional joint velocity observations and 14-dimensional joint torque observations. Visual observations include video streams from four cameras: a high camera, a low camera, a left wrist camera, and a right wrist camera, all in RGB format (3 channels) with 480x640 resolution and AV1 encoding. Metadata includes task index, subtask description string, training flag, timestamps, frame index, episode index, and global index. This dataset is suitable for research in robot imitation learning, reinforcement learning, multimodal perception, and control policy learning.




