Summary
PhysForge is a two-stage framework for generating physics-grounded, simulation-ready 3D assets, addressing a key bottleneck for interactive virtual worlds and embodied AI. Unlike prior methods that focus only on static geometry, PhysForge is rooted in functional logic and hierarchical physics. It is supported by PhysDB, a large-scale dataset of 150,000 assets with four-tier physical annotations. In the first stage, a vision-language model (VLM) acts as a physical architect, planning a Hierarchical Physical Blueprint that defines material, functional, and kinematic constraints. In the second stage, a physics-grounded diffusion model realizes this blueprint, synthesizing high-fidelity geometry together with precise kinematic parameters through a novel KineVoxel Injection (KVI) mechanism. Experiments show that PhysForge produces functionally plausible assets that can be directly used in simulation, serving as a robust data engine for interactive 3D content generation and embodied agents. Paper: arXiv 2605.05163.
Overview
- Field: Computer Vision
- Authors: Yunhan Yang, Chunshi Wang, Junliang Ye, Yang Li, Zanxin Chen, Zehuan Huang, Yao Mu, Zhuo Chen, Chunchao Guo, Xihui Liu
- Published: 2026-05-06
- arXiv: 2605.05163
Abstract
Synthesizing physics-grounded 3D assets is a critical bottleneck for interactive virtual worlds and embodied AI. Existing methods predominantly focus on static geometry, overlooking the functional properties essential for interaction. The authors propose that interactive asset generation must be rooted in functional logic and hierarchical physics.
To bridge this gap, they introduce PhysForge, a decoupled two-stage framework supported by PhysDB, a large-scale dataset of 150,000 assets with four-tier physical annotations:
1. A VLM acts as a physical architect to plan a Hierarchical Physical Blueprint defining material, functional, and kinematic constraints.
2. A physics-grounded diffusion model realizes this blueprint by synthesizing high-fidelity geometry alongside precise kinematic parameters via a novel KineVoxel Injection (KVI) mechanism.
Experiments demonstrate that PhysForge produces functionally plausible, simulation-ready assets, providing a robust data engine for interactive 3D content and embodied agents.
---
*Auto-collected on 2026-05-08.*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/177619595