Freiburg Franka Play

这个数据集是用于机器人无监督强化学习研究的，包含了7自由度机械臂和并行夹爪在虚拟现实控制器操作下执行各种非定向操作任务的多个片段。其中约1%的数据带有自然语言嵌入标注，每个状态-动作对都以Numpy npz文件形式存储，包含静态相机和夹爪相机的RGB-D图像、本体感知状态以及对应的未来动作。

This dataset is intended for unsupervised reinforcement learning research in robotics. It includes multiple segments of 7-degree-of-freedom (7-DOF) robotic arms and parallel grippers performing various non-directional manipulation tasks when operated via virtual reality controllers. Approximately 1% of the data carries natural language embedding annotations, and each state-action pair is stored as a Numpy npz file, containing RGB-D images from static cameras and gripper-mounted cameras, proprioceptive states, as well as the corresponding future actions.

搜集汇总

背景与挑战

背景概述

该数据集专为机器人无监督强化学习设计，包含7自由度机械臂在虚拟现实操作下执行非定向任务的多个片段，其中约1%数据带有自然语言标注。数据以Numpy npz文件存储，整合了RGB-D图像、本体状态和未来动作，适用于多模态学习研究。