tersoohaan/eval_pi0_3D_infe_light
收藏资源简介:
该数据集是使用LeRobot平台创建的机器人学数据集,专注于机器人控制与感知任务。数据集包含15个完整的情节(episodes),总计9597帧数据,采样率为30帧/秒。数据以parquet文件格式存储,视频文件为mp4格式。数据集记录了机器人(型号为so101_follower)的6个关节(包括肩部平移、肩部升降、肘部弯曲、腕部弯曲、腕部旋转和夹爪)的位置作为动作和观测状态。同时,数据集提供了两个摄像头的观测信息:一个前置摄像头和一个顶部摄像头,分别输出480x640分辨率的RGB视频和深度图(前置摄像头包含深度信息)。此外,数据集还包含时间戳、帧索引、情节索引、任务索引等元数据。该数据集适用于机器人模仿学习、强化学习或视觉控制等研究任务。
This dataset is a robotics dataset created using the LeRobot platform, focusing on robot control and perception tasks. It contains 15 complete episodes with a total of 9597 frames at a sampling rate of 30 fps. The data is stored in parquet format, and video files are in mp4 format. The dataset records the positions of 6 robot joints (including shoulder pan, shoulder lift, elbow flex, wrist flex, wrist roll, and gripper) as both actions and observation states. It also provides observations from two cameras: a front camera and a top camera, each outputting 480x640 resolution RGB videos and depth maps (the front camera includes depth information). Additionally, the dataset includes metadata such as timestamps, frame indices, episode indices, and task indices. This dataset is suitable for research tasks like imitation learning, reinforcement learning, or visual control in robotics.




