rollout_test_20260625_130518
收藏资源简介:
该数据集是一个机器人控制数据集,使用LeRobot工具创建,专门用于机器人模仿学习或强化学习研究。数据集包含一个完整的机器人操作回合,总时长约36秒(1088帧,30fps)。数据采用多模态形式:既包含机器人的低维状态信息(6个关节的位置),也包含两个视角的高维视觉观测——左手腕摄像头和顶部摄像头拍摄的480x640 RGB视频。每个数据样本包含机器人在特定时刻的动作指令(6维关节位置控制)、观测状态(6维关节位置反馈)、双视角图像、时间戳以及帧、回合、任务等索引信息。数据集以Parquet格式存储,总大小约300MB,适用于训练机器人策略模型,特别是需要结合视觉与状态信息的控制任务。
This dataset is a robot control dataset created using the LeRobot tool, specifically designed for robot imitation learning or reinforcement learning research. It contains a complete robot operation episode with a total duration of approximately 36 seconds (1088 frames at 30fps). The data is multimodal: it includes low-dimensional state information of the robot (positions of 6 joints) and high-dimensional visual observations from two perspectives—480x640 RGB videos captured by a left-wrist camera and a top camera. Each data sample includes the robots action command at a specific moment (6-dimensional joint position control), observation state (6-dimensional joint position feedback), dual-perspective images, timestamps, and index information such as frame, episode, and task. The dataset is stored in Parquet format with a total size of about 300MB and is suitable for training robot policy models, particularly for control tasks that require combining visual and state information.




