Vision-Language-Action Models

Most robots today don't use language models. They use computer vision and narrower AI trained for one job, like spotting objects or planning a path.

But researchers are now building robots that combine three things in one system: seeing (vision), understanding language, and acting.

These are called Vision-Language-Action models, or VLA models.

Instead of writing exact rules for every task, you can just tell the robot what to do: "pick up the red block" and it works out how.

They learn from huge amounts of examples (videos, images, instructions), and the robot can understand what you say, not just what it sees.

These are still early and they can still fail on simple tasks.