VoMP: Predicting Volumetric Mechanical Property Fields

Abstract

Physical simulation relies on spatially-varying mechanical properties, typically laboriously hand-crafted. We present the first feed-forward model to predict fine-grained mechanical properties, Young's modulus ($E$), Poisson's ratio ($\nu$), and density ($\rho$), throughout the volume of 3D objects. Our model supports any 3D representation that can be rendered and voxelized, including Signed Distance Fields (SDFs), Gaussian Splats and Neural Radiance Fields (NeRFs). To achieve this, we aggregate per-voxel multi-view features for any input, which are passed to our trained Geometry Transformer to predict per-voxel material latent codes. These latents reside on the trained manifold of physically plausible materials, which we train on a real-world dataset, guaranteeing the validity of decoded per-voxel materials. To obtain object-level training data, we propose an annotation pipeline combining knowledge from segmented 3D datasets, material databases, and a vision-language model. Experiments show that VoMP estimates accurate volumetric properties and can convert 3D objects into simulation-ready assets, resulting in realistic deformable simulations and far outperforming prior art.

Method

---

Our pipeline first trains a MatVAE, a VAE that learns a 2D latent space of the triplets ($E$, $\nu$, $\rho$) trained on a dataset of 100K physically-valid triplets. We take a 3D representation, render it from different views, voxelize it, aggregate image features across views and reconstruct them onto the voxels. These voxels with image features are passed to our trained Geometry Transformer to predict per-voxel material latent codes which lie in the latent space of the MatVAE. These latents are then decoded to get the per-voxel material properties.

Robot Simulation (Gaussian Splat)

Using VoMP

Step 1: Capture a Splat 📷

Drag to rotate • Scroll to zoom • Right click to pan

Step 2: VoMP predicts properties 🚀

Young's Modulus

Poisson's Ratio

Density

Step 3: Use a simulator to produce realistic interactions 🔥

Next Step

Simulating Gaussian Splats

Simulating Meshes

Mechanical Property Fields

• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 
• 

VoMP produces correct materials for internal volumetric structures. We show all the voxels (leftmost) and then cluster material predictions to produce segments.

VoMP predicts fine-grained mechanical property fields throughout the volume of 3D objects. We show (top) the predicted fields and (bottom) a slice plane through the fields.

Comparing Simulations to Baselines

Comparing Property Fields with Baselines

• 
• 
• 
• 

	NeRF2Physics	PUGS	Phys4DGen	VoMP
Young's Modulus
Poisson's Ratio
Density

	NeRF2Physics	PUGS	Phys4DGen	VoMP
Young's Modulus
Poisson's Ratio
Density

	NeRF2Physics	PUGS	Phys4DGen	VoMP
Young's Modulus
Poisson's Ratio
Density

	NeRF2Physics	PUGS	Phys4DGen	VoMP
Young's Modulus
Poisson's Ratio
Density

We comapre VoMP with baseline methods and show estimated mechanical property fields for 3D objects. VoMP achieves significantly better mechanical volumetric property fields.

Citation

    @article{dagli2025vomp,
        title={VoMP: Predicting Volumetric Mechanical Property Fields},
        author={Dagli, Rishit and Xiang, Donglai and Modi, Vismay and Loop, Charles and Tsang, Clement Fuji and 
                Chen, Anka He and Hu, Anita and State, Gavriel and Levin, David I.W. and Shugrina, Maria},
        journal={arXiv preprint},
        year={2025}
    }