walleed_hg_double_fold
收藏资源简介:
该机器人学数据集使用LeRobot工具创建,旨在支持机器人控制或模仿学习相关的研究与应用。数据集采用多模态形式,核心内容包括机器人的动作指令、状态观测以及来自前摄像头的视觉观测。具体而言,动作和状态观测均包含六个关节的位置信息:肩部平移、肩部抬升、肘部弯曲、腕部弯曲、腕部旋转以及夹爪位置。视觉观测为来自前摄像头的RGB视频流,分辨率为640x480,帧率为30fps。此外,数据集还提供了时间戳以及用于数据组织的多个索引字段,包括帧索引、片段索引和任务索引。数据集规模较大,共包含286个独立的数据片段,总计332,942帧数据,但仅对应一个任务。数据以分块形式存储,主数据文件为Parquet格式,视频文件为MP4格式。数据集适用于机器人策略学习、行为克隆、离线强化学习等任务。
This robotics dataset is created using the LeRobot toolkit, aiming to support research and applications related to robot control or imitation learning. It is a multimodal dataset, whose core contents include robot action instructions, state observations, and visual observations from the front-facing camera. Specifically, both the action and state observations contain position information of six joints: shoulder translation, shoulder elevation, elbow flexion, wrist flexion, wrist rotation, and gripper position. The visual observations are RGB video streams from the front camera, with a resolution of 640x480 and a frame rate of 30 fps. In addition, the dataset provides timestamps and multiple index fields for data organization, including frame index, episode index, and task index. The dataset has a large scale, containing a total of 286 independent data episodes with 332,942 frames in total, but only corresponding to one single task. The data is stored in chunks, with the main data files in Parquet format and video files in MP4 format. This dataset is applicable to tasks such as robot policy learning, behavioral cloning, and offline reinforcement learning.




