Towards a Generative 3D World Engine for Embodied Intelligence

Xinjie Wang¹, Liu Liu¹, Yu Cao², Ruiqi Wu^5,1, Wenkang Qin²,

Dehui Wang^4,3, Wei Sui³, Zhizhong Su¹

¹Horizon Robotics, ²GigaAI, ³D-Robotics, ⁴Shanghai Jiao Tong University, ⁵VCIP, CS, Nankai University

arXiv Video Code Hugging Face

EmbodiedGen generates interactive 3D worlds with real-world scale and physical realism at low cost.

Framework

EmbodiedGen is a toolkit to generate diverse and interactive 3D worlds composed of generative 3D assets with plausible physics, leveraging generative AI to address the challenges of generalization in embodied intelligence related research. EmbodiedGen composed of six key modules: Image-to-3D, Text-to-3D, Texture Generation, Articulated Object Generation, Scene Generation and Layout Generation.

Results

Demo

Image-to-3D

Demo

Text-to-3D

"Antique brass key, intricate filigree"

"Rusty old wrench, peeling paint"

"Sleek black drone, red sensors"

"Miniature screwdriver with bright orange handle"

"European style wooden dressing table"

Demo

Texture Generation

Articulated Object Generation

Scene Generation

BibTeX

@misc{xinjie2025embodiedgengenerative3dworld,
      title={EmbodiedGen: Towards a Generative 3D World Engine for Embodied Intelligence}, 
      author={Wang Xinjie and Liu Liu and Cao Yu and Wu Ruiqi and Qin Wenkang and Wang Dehui and Sui Wei and Su Zhizhong},
      year={2025},
      eprint={2506.10600},
      archivePrefix={arXiv},
      primaryClass={cs.RO},
      url={https://arxiv.org/abs/2506.10600}, 
}