Humanoid-X
收藏资源简介:
Humanoid-X是由南加州大学等机构创建的大规模人形机器人数据集,旨在通过大量的人类视频数据促进人形机器人的学习。该数据集包含163,800个动作样本,涵盖多种动作类别,每个样本包含视频、文本描述、3D人体姿态、人形机器人关键点和机器人动作序列。数据集通过从互联网和学术数据集中挖掘视频,经过视频字幕生成、人体姿态估计、动作重定向等步骤创建。Humanoid-X的应用领域主要集中在通过自然语言指令实现人形机器人的通用姿态控制,旨在提高机器人在日常任务中的通用性和可扩展性。
Humanoid-X is a large-scale humanoid robot dataset developed by institutions including the University of Southern California, designed to facilitate humanoid robot learning using massive volumes of human video data. This dataset contains 163,800 motion samples spanning diverse motion categories, where each sample includes videos, text descriptions, 3D human poses, humanoid robot keypoints, and robot motion sequences. The dataset is constructed by mining videos from the Internet and academic datasets, followed by key procedures such as video caption generation, human pose estimation, and motion retargeting. The primary application scenarios of Humanoid-X center on general pose control of humanoid robots through natural language instructions, with the ultimate goal of improving the generality and scalability of robots in daily tasks.




