Paper Overview
Field: Computer Vision (CV) Authors: Trong-Tung Nguyen, Anand Bhattad Published: 2026-09-29 arXiv: 2609.38153
Abstract (translated from Chinese summary)
The authors present PowerSim, a method that brings physically grounded, differentiable dynamics to PowerFoam's power diagram based 3D representation. PowerSim directly couples a pre-trained PowerFoam scene to the Material Point Method (MPM) by exploiting a natural alignment between the two: the geometric and appearance properties of each primitive correspond closely to the quantities MPM already tracks as an object deforms. As a result, simulated motion can directly drive the scene's geometry and appearance, without an intermediate auxiliary representation.
Built on this framework, PowerSim enables a range of applications on real and synthetic scenes:
1. Simulating a static scene's response under user interaction 2. Recovering spatially varying material fields 3. Compositing primitives from independently captured scenes into a single simulatable scene 4. Ray-traced reflections that update consistently as objects deform
Results
The experiments show that PowerSim outperforms previous frameworks on physics-based dynamics while unlocking unique advantages — such as secondary-ray lighting effects on dynamic scenes. The authors note that results are best viewed on the project website.
Original Abstract (English)
> We introduce PowerSim, a method to bring physically grounded, differentiable dynamics to PowerFoam's power diagram based 3D representation. PowerSim directly couples a pre-trained PowerFoam scene to the Material Point Method (MPM) by exploiting a natural alignment between the two: the geometric and appearance properties of each primitive correspond closely to the quantities MPM already tracks as an object deforms. Consequently, simulated motion can drive the scene's geometry and appearance directly, without an auxiliary representation in between. Built on this framework, we enable a range of applications on real and synthetic scenes: (1) simulating a static scene under user interaction, (2) recovering spatially varying material fields, (3) compositing primitives from independently captured scenes...
---
*Auto-collected on 2026-10-01*