This is a neural network on Scratch! Click the green flag and it should do the rest. (no need for Turbo Mode unless you really want to. They usually get it pretty quick. This should work on any map, feel free to remix and draw a new one, just don't make the turns too sharp.
HOW IT WORKS: (kinda long) A neural network is kind of like a "brain simulator". There are 3 layers: Inputs, Processing, and Outputs. In this case, there are 5 Inputs that the car is given. All it knows is how far away it is from the green at 5 different directions: straight forward, left, right, forward-right, and forward-left. Then, there are 5 neurons for Processing. These multiply the weights of each Input by a certain value to help decide whether to go right or left. Finally, in Output, the sum of each Input multiplied by each neuron's individual weight value adds up to decide whether the car turns right, left, or keeps facing straight. In each generation, 5 cars are generated. The first 5 have completely random neuron weights, and the one that makes it the farthest is declared to have the "best" neurons. The next generation is based on those "best" values, with a 30% chance of a complete mutation on each neuron (completely randomizing its value). This allows for progress to be made by randomly mutating certain neurons. The other 70% of the time, the weights follow the "best" neuron values with very small deviations. That's pretty much it!