HuggingFaceVLA/libero

LIBERO是一个针对机器人操作任务的自然语言指令的全面基准测试数据集。该数据集包括四个子数据集：Libero-Spatial、Libero-Object、Libero-Goal和Libero-10，并经过预处理以适应LeRobot框架。它包含了1693个操作片段，40种独特的操作任务，使用Franka Panda 7-DOF机械臂，双摄像头设置，以及10Hz的帧率。数据集以LeRobot v2.1格式存储，总大小约为35GB。

LIBERO is a comprehensive benchmark for robot manipulation tasks with natural language instructions. This dataset includes four sub-datasets: Libero-Spatial, Libero-Object, Libero-Goal, and Libero-10, and has been preprocessed to fit the LeRobot framework. It contains 1,693 manipulation episodes, 40 unique manipulation tasks, uses a Franka Panda 7-DOF arm, a dual camera setup, and operates at a frame rate of 10 Hz. The dataset is stored in the LeRobot v2.1 format and is approximately 35GB in size.

搜集汇总

背景与挑战

背景概述

该数据集是一个用于机器人技术研究的大规模数据集，包含273,465行数据，涵盖图像和时间序列模态，采用parquet格式存储，适用于视觉语言动作模型的研究和开发。