UE5.3 : AI Cars learns to drive with DeepLearning methods
Deep Learning: Simple neural network without hidden layers
There are ± 30 cars on the track. The cars have 5 points of vision to enable them to turn.
Race time: ± 60 seconds, timer common to AIs.
When a car passes a checkpoint, it gains 1 second on its timer to encourage it to make more points. In addition, a coefficient will be applied to checkpoints throughout the circuit to improve the AI's score as it improves.
The first generation has a random weight. The next generation focuses on the top 5 AIs. This means that generation n+1 will generate new AI cars with the old best data cars. The more points a car has, the more data the next generation will get.