Research Advances in AI-Assisted Material Generation for Physical AI

Подробнее о видео

Digital twins are vital for the training and testing of physical AI that can understand and perform complex actions in the real world. But building digital twins is slow and tedious. NVIDIA’s latest generative AI and rendering research showcases the potential to dramatically accelerate and scale these workflows.

These breakthroughs combine NVIDIA Cosmos World Foundation Models and advanced rendering technologies to allow developers to pair AI assistants with experts at the heart of digital twin and physical AI development. In these future workflows, experts will engage with AI assistants using simple language to quickly generate realistic, physically accurate base materials for 3D models. They can then fine-tune generated materials, adjusting properties such as roughness and texture to better reflect the real world.

This research, along with the workflows it enables, showcases the potential to quickly transform 3D models into large-scale, simulation-ready digital twins that can be used to train safe and efficient physical AI.

Read the research paper:

#OpenUSD
#DigitalTwin
#NVIDIAOmniverse
#IndustrialAI
#SIGGRAPH2025