Why use Deep Learning instead of algorithms for decision making in self-driving cars?

Question

Let's say I want to build a self-driving car that uses cameras and ai to detect objects and driving lanes.

So if the camera detects e.g. a car in front, it calculates the distance (using a sensor) and passes that information to the algorithm which handles steering and acceleration.

Another example is parking. I could just write an algorithm that can park the car perfectly every time, instead of trying to train an ai, which would take a lot of time and computation.

Why would I use Deep Learning to make the car navigate instead of using a simple algorithm (like in video games) that can find the best path and steer accordingly?

Answer 1

The real world is harder than a video game.

I could just write an algorithm that can park the car perfectly every time.

No you cannot. You may believe that you can, but you cannot. Real world data is too imperfect and complex.

Let's say I want to build a self-driving car that uses cameras and ai to detect objects and driving lanes.

So if the camera detects e.g. a car in front, it calculates the distance (using a sensor) and passes that information to the algorithm which handles steering and acceleration.

How do you propose "passing that information" to the algorithm? Raw camera data are images with pixels. What sort of algorithm could you construct that could drive based on raw camera input?

In short a human could never construct such an algorithm. It is beyond our abilities.

I also want to point out, deep learning IS an algorithm, just one learned from data.