PhysX-Mobility
收藏资源简介:
PhysX-Mobility是由南洋理工大学与上海人工智能实验室联合构建的物理3D资产数据集,旨在突破现有物理数据集多样性不足的局限。该数据集包含2000余个常见现实物体,涵盖47个对象类别,通过系统化采集PartNet-Mobility资源并精细标注物理属性而构建。数据集通过严谨的物理标注流程,为每个对象提供完整的运动学参数与材料特性,主要应用于具身智能与机器人策略学习领域,为物理仿真系统提供高质量训练基础。
PhysX-Mobility is a physical 3D asset dataset co-developed by Nanyang Technological University (NTU) and Shanghai AI Laboratory (SAIL), aiming to address the limitation of insufficient diversity in existing physical datasets. This dataset contains over 2000 common real-world objects spanning 47 object categories, and is constructed by systematically collecting resources from PartNet-Mobility and performing fine-grained annotations of physical properties. Through a rigorous physical annotation workflow, it provides complete kinematic parameters and material properties for each object. Primarily applied in the fields of embodied intelligence and robot policy learning, this dataset offers high-quality training foundations for physical simulation systems.
PhysX-Mobility 数据集概述
基本信息
- 数据集名称: PhysX-Mobility
- 发布者: Caoza
- 访问地址: https://hf-mirror.com/datasets/Caoza/PhysX-Mobility
- 点赞数: 5
- 上月下载量: 3
任务类型
- Image-to-3D
技术特征
- 语言: English
- 标签:
- Physical 3D Generation
- 3D Vision
- 3D
学术信息
- ArXiv: arxiv: 2511.13648
- 许可证: cc-by-nc-4.0
数据集描述
该数据集旨在填补物理标注3D数据集的关键空白,是首个在五个基础维度上系统标注的物理基础3D数据集:
- 绝对尺度
- 材质
- 功能可供性
- 运动学
- 功能描述
技术细节
数据集整体结构与PhysXNet相同
引用信息
bibtex @article{physxanything, title={PhysX-Anything: Simulation-Ready Physical 3D Assets from Single Image}, author={Cao, Ziang and Hong, Fangzhou and Chen, Zhaoxi and Pan, Liang and Liu, Ziwei}, journal={arXiv preprint arXiv:2511.13648}, year={2025} }
致谢
基于PartNet-Mobility构建,向贡献者表示感谢

- 1通过南洋理工大学S实验室、上海人工智能实验室 · 2025年



