Vision-language-action model

A vision-language-action model (VLA) combines vision (what the system sees), language (an instruction or description) and action (a physical movement, usually by a robot) in one model. The model translates an observation and a command directly into a control signal, without a human programming the intermediate step.

What it means for the calculation

This is a different category from an estimating model. A VLA drives a physical movement; a model that understands a part and proposes a cost does not drive or move anything. Blake falls in the second category: it recognises and calculates, it does not control a robot or a machine.

Where this comes up

In robotics and autonomous systems that act directly and physically based on what they observe, not in systems that only produce a proposal or a number.

Common mistakes

Labelling any system that works with geometry or machines as "vision-language-action," even when it does not control anything physical. The term applies specifically to models that carry out an action, not to models that only recognise or calculate.

Is Blake a vision-language-action model?

No; Blake recognises geometry and requirements and proposes a calculation, but does not control a machine or a robot.

What are VLAs actually used for?

Mostly in robotics: a robot arm that grips, places or assembles based on what a camera sees.

What is the difference from physical AI in general?

Physical AI is the broader category; a VLA is specifically a model that translates observation directly into a physical action.