Would you care to elaborate? Surely most of the lines of code are not directly implementing theoretical things, but I would say the meat of the work is without a doubt theoretical.

I don't know about self driving cars, but I have known people who did PhD level research in control engineering and then went into industry (real, heavy metal industry, like offshore oil installations in the North Sea) implementing advanced control systems.

The estimate they gave me of the contribution of their control algorithms (the "theoretical" part of the project) to the over amount of effort getting the thing working was less than 1%.

Theory is thinking about how to build a self-driving car, and the algorithms and considerations that would be required to do so, and perhaps writing a paper about how best to go about doing it. Practical is writing code, running wire, and testing the car. Surely they've done both, but now that the car exists they're into the practical territory. As they adventure through the practical part, surely new theoreticals are discovered and published.

Just because nobody has built a self-driving car before does not automatically make actually building one a theoretical exercise. The process is applying the sciences to making a car do something, which is very practical. Your awesome sort algorithm and the paper explaining it is theoretical. My implementation of it is practical.

That is the differentiation that most people can't see.