Embodied AI

Embodied AI (or Physical AI) is AI that lives in robots, drones or other devices with bodies so it can see, move and act in the real world.

AI changes traditional robots from following fixed rules, like assembling cars in a factory, to being able to learn from experience and adapt to new situations.

Most embodied AI today uses computer vision and AI models trained for specific tasks, not LLMs.

How does a robot see?

A robot's camera captures an image as a grid of tiny coloured dots called pixels. On its own, a robot has no idea what those pixels mean.

Computer vision analyses patterns across those pixels, such as colours, edges and shapes, to identify objects and understand what is happening in an image.

It learns these patterns by training on large numbers of labelled images.

This allows it to recognise similar objects in images it has never seen before, helping self-driving cars detect pedestrians and warehouse robots avoid obstacles.