Synthetic Data from Action-less Human Videos
收藏arXiv2025-09-30 收录
下载链接：
https://portal-cornell.github.io/X-Sim/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从RGBD人类视频生成的合成数据，用于训练高性能的机器人操作策略。它包含了物体姿态和摄像机视角的变化，以增强对测试时变化的鲁棒性。该数据集在5个操作任务上进行了评估，使用的是7自由度的Franka机械臂。这些操作任务包括捡取放置、非抓握操作以及精确插入等。
This dataset is synthetic data generated from RGBD human videos, designed for training high-performance robotic manipulation policies. It incorporates variations in object poses and camera viewpoints to enhance robustness against changes encountered during testing. Evaluations of this dataset are conducted across five manipulation tasks using a 7-degree-of-freedom (7-DOF) Franka robotic arm. These tasks include pick-and-place, non-prehensile manipulation, and precision insertion, among others.
提供机构：
X-Sim Research Team
