WHY IS ROBOTICS SUCH A HOT TOPIC?

If you follow the ever-advancing field of AI, then you’ll probably have noticed the increasing mention of robotic systems - often termed physical AI - not least by Jensen Huang from NVIDIA at recent GTC events. In the same way that about ten years ago, advances in GPUs combined with the ready availability of large datasets to fuel AI model development; the recent advances in large language models (LLMs) and agentic AI models capable of reasoning has ushered in a new dawn of physical systems that can interact with the real-world and ourselves.

These physical systems - often either robotic arms or entire independent quadruped or humanoid robots - require huge amounts of data from sensors and actuators, plus the ability to perceive their environments to effectively respond to stimuli and interact. So complex are these data and processing demands that even the most powerful GPU-accelerated systems were not capable of AI at this scale. NVIDIA’s solution to this is three computers, the first a DGX appliance to train the AI model that will be deployed on the robot; the second an Omniverse platform to simulate how the robot will move and react in the real world; and third a Jetson system running the model on the robot.

AI robotic systems are much more complicated than other types of AI model. Imagine an algorithm designed to recognise five types of cooking utensil - this requires training on cooking utensils and cameras to see the various types. Now add in the ability to pick up these utensils and use them - the cameras now need robotic arms and the training of how each utensil is used. Now add mobility so the robot can recognise, pick up and use the utensil, but only in the correct kitchen location - this adds training on processes and a spatial understanding of its environment. Finally make the robot work in a crowded kitchen where it may be obstructed or knocked off balance mid task - training now needs to include how to negotiate obstacles and how to recover an interrupted task, without causing injury as some collisions may be with softer humans rather than harder surfaces.

All these additional inputs and parameters make for an increasingly complex AI model, however when physical AI systems are capable of such flexibility, customisation and nuanced human interaction, then we could be on the verge of a whole new era of where robots can be deployed. This is why there so much talk about AI and robotics.

The range of robotic applications is potentially huge, covering all manner of scenarios from manufacturing to healthcare to engineering to logistics.

There are different ways that this training can be achieved. The kitchen scenario mentioned above would seem to involve a lot of AI training on large datasets and many trial and error processes that would likely be costly and potentially dangerous. This is known as reinforcement learning and is very time consuming. However, the latest GPUs also enable the advent of digital twin environments. A digital twin is a virtual world identical to our own - obeying all the same rules - so that robots can be designed and tested in a cloud platform such as NVIDIA Omniverse using a robotic development application such as NVIDIA Isaac Lab. This allows robots to be fully trained prior to being deployed in the real world - thus removing many of these development obstacles and making robot deployment much faster and simpler.

You can learn more about robotics and other applications by reading our Omniverse and Digital Twin blog posts. Alternatively explore human-robotic interaction further by reading our Oxford Robotics Institute case study.