Robots and Physical AI Shaped by NVIDIA Research at SIGGRAPH 2025

Physical AI is redefining the intersection of artificial intelligence and the physical world, equipping machines to perceive, reason, and interact autonomously in dynamic environments. This video is a concise visual summary of key NVIDIA research and innovations in physical AI featured at SIGGRAPH 2025. We demonstrate the integration of neural graphics, physics-based simulation, computer vision, synthetic data generation, and advanced AI reasoning—core pillars of physical AI that drive the next generation of robotics and intelligent systems. Physical AI, highlighted in this sample of SIGGRAPH research, extends traditional AI by teaching machines to operate in three-dimensional, real-world spaces. Leveraging highly accurate computer simulations, NVIDIA’s work enables autonomous agents—such as robots and self-driving vehicles—to safely practice skills virtually before deployment in unpredictable real-world settings. Environment perception is powered by advanced computer vision and neural graphics, helping AI not only see but truly comprehend the complex physical spaces in which it operates. Synthetic data generation, a crucial aspect in NVIDIA’s research, allows for scalable, diverse training data that accelerates the development and validation of physical AI models. Paired with physics-based simulation, these methods let AI systems repeatedly learn from trial and error, gaining robust capabilities like dexterous manipulation, adaptive navigation, and decision-making in real-time scenarios. The video also visualizes breakthroughs from technical papers—including novel neural rendering, efficient simulation techniques, and next-gen generative models that streamline everything from texture painting to 3D scene construction. By showcasing highlights from our papers presented at SIGGRAPH 2025, you now have a glimpse at how physical AI methods—spanning perception to action—are shaping robotics, digital twins, autonomous vehicles, and smart environments. See all of NVI