UniPhys-40K
收藏资源简介:
UniPhys-40K是由浙江大学与Manycore Tech Inc.等机构联合构建的大规模物理基础数据集,旨在为具身智能和机器人仿真提供统一物理语义的仿真就绪三维资产。该数据集汇集了来自Objaverse、HSSD、3D-FUTURE等异构三维资源库的逾四万个三维模型,涵盖了84个物体类别,并标注了关节运动语义与材料、密度等内在物理属性,形成了从日常家居到大型基础设施的广泛物体尺度与质量分布。其构建过程依托于自动化物理基础管线UniPhys,通过感知引导的结构分解、几何感知的关节基础以及仿真驱动的一致性验证等模块,实现了从原始三维网格到物理一致资产的规模化转换。该数据集主要应用于机器人仿真、物理交互建模以及三维生成等领域,致力于弥合视觉逼真资产与仿真就绪资产之间的语义鸿沟,为物理基础模型的训练与评估提供关键支撑。
UniPhys-40K is a large-scale physics foundational dataset jointly constructed by Zhejiang University, Manycore Tech Inc., and other institutions. It aims to provide simulation-ready 3D assets with unified physical semantics for embodied intelligence and robotic simulation. This dataset aggregates over 40,000 3D models from heterogeneous 3D repositories including Objaverse, HSSD, and 3D-FUTURE, covering 84 object categories. It is annotated with joint motion semantics and intrinsic physical properties such as material and density, forming a wide range of object scales and mass distributions spanning from daily household items to large-scale infrastructure. Its construction relies on the automated physics foundational pipeline UniPhys, which enables large-scale conversion from raw 3D meshes to physics-consistent assets through modules such as perception-guided structural decomposition, geometry-aware joint foundation, and simulation-driven consistency verification. This dataset is mainly applied in fields such as robotic simulation, physical interaction modeling, and 3D generation. It is committed to bridging the semantic gap between visually realistic assets and simulation-ready assets, providing critical support for the training and evaluation of physics foundational models.
数据集概述:UniPhysGen
地址:https://github.com/breezexian/UniPhysGen
核心目标:UniPhysGen 是一个致力于为仿真就绪的3D资产提供统一物理基础的数据集项目。其全称为“Unified Physical Grounding for Simulation-Ready 3D Assets”,即“面向仿真就绪3D资产的统一物理基础”。
主要特征:
- 专注于3D资产的物理属性统一化,旨在使3D模型具备仿真环境中所需的物理交互能力。
- 强调“仿真就绪”(Simulation-Ready),即数据集中的3D资产可直接用于物理仿真场景。
- 项目名称中的“UniPhysGen”暗示了其生成(Gen)与统一(Uni)物理基础的双重功能。
适用场景:该项目适用于需要物理真实感的3D仿真环境,如机器人模拟、自动驾驶仿真、游戏物理引擎集成等领域。




