California: Nvidia CEO Jensen Huang believes the next major breakthrough in artificial intelligence will move the technology beyond computer screens and into the physical world, enabling machines to understand their surroundings, make decisions and perform tasks with greater independence.
Huang describes this emerging technology as Physical AI, which combines artificial intelligence with cameras, sensors and robotic systems to help machines respond to real-world situations without requiring instructions for every movement.
Unlike conventional AI tools such as ChatGPT, which can answer questions, write emails and analyze information, Physical AI could enable a robot to recognize a fragile cup, pick it up carefully and place it on a table without breaking it.
The technology is not limited to humanoid robots. It could also be used in self-driving vehicles to anticipate when pedestrians may cross the road, while warehouse robots could navigate around people and collect items without relying on predetermined routes.
Hospitals and factories are among the other sectors exploring potential applications of the technology, which could automate tasks and help machines operate in complex environments.
However, teaching machines to understand the physical world remains a major challenge. While an incorrect response from a chatbot may provide misleading information, a similar mistake by a robot could damage equipment or cause injury.
To address this challenge, Nvidia is developing simulation technologies that allow robots to practise tasks in virtual environments before carrying them out in the real world.
The company’s Cosmos AI models are designed to generate realistic scenarios that can help train machines to understand how objects move and interact with their surroundings.
Huang acknowledged that technical challenges, safety concerns and costs remain obstacles to the widespread adoption of Physical AI. However, he suggested that robotics is entering a phase comparable to the early development of ChatGPT.
The technology could eventually change how people interact with machines. Instead of asking artificial intelligence how to perform a task, users may simply instruct it to carry out the task itself.