Teaching a Smartphone App to Drive your car

Elon Musk sort of made waves insisting vision, e.g. Cameras, must be the way to go to solve full self-driving A.I.

Not Lidar, nor radar. Just Camera vision, with some sort of AI making choices based on what it sees, similar to how a human drives.

George Hotz in this interview believes in an even more minimal model; that most phones already have enough processing power to do everything we need a fully self-driving AI to do. Just hook them up to the sensors already in most vehicles, and it will drive it better than the Wyamos or Ubers of the world spending $$$ millions on self-driving Lidar demos.

Hotz lays down the progress, the scope of the effort, what he would call “the correct” , or holistic way of approaching the problem, and how the industry has changed its approach over time.

In short, he believes that when his approach is proven successful it will have downstream implications for things beyond self-driving. It’ll be a step towards the AGI holy grail; robots doing your cooking and cleaning. As who has time for that?

I have little interest in this concept.

I have a Cadillac CT-6. The year after the model I have, Cadillac offered “smart cruise control” which is supposed to work on the Interstate Highway system. The trouble is we have so much traffic in Florida that there would be no place to use it.

My automobile has cameras that take pictures around the whole car. It also has sonar based sensors all around it. That’s a lot more than one darn cell phone. Still Cadillac did not feel that this is enough to have “smart cruise control” work anywhere except on the Interstate highways.

The cell phone would plug into those sensors and make use of them. The phone is not doing the task alone, the point is that it’s processor, which are mostly in the Megahertz scale, is enough to intake the data, and run an AI suitable enough to achieve full self-driving.

Tesla uses far more expensive, Gighertz scale GPUs to do the same thing. And even that’s a bargain next to the far more custom/ proprietary processors used by Waymo and Uber.