Abstract

Today, the world is captivated by cognitive AI applications such as large language models. But what will it take to bring the benefits of AI into the messy, diverse and safety-critical physical world? Robotics and autonomous systems must deal with open-ended environments, irreversible physical actions, and deployment economics that look very different from pure software.

In this talk, I will outline the frontier challenges and opportunities in embodying AI in the real world, drawing on our journey building Wayve. Originating from research on deep learning for scene understanding at the University of Cambridge, Wayve has spent the last decade developing Embodied Intelligence for autonomous driving. Our technology has been demonstrated across more than 500 cities in Europe, North America and Asia, and will soon be deployed with major automakers such as Nissan and fleet partners like Uber.

I will share the key technical ideas, system-level lessons, and open problems that must be solved to make Embodied AI a safe, scalable and economically viable reality.