🚗🚗 https://g3pgames.itch.io/self-driving 🚕🚕
Practice: Example Of Using Neural Networks and Genetic Algorithms On Self Driving Cars
Neural network takes in the inputs and gives out the outputs. Genetic Algorithm trains the neural network on how to get better outputs. Car has 3 major sensors (in our case): three extruded raycasts in each direction - to hit obsticles. The output of these sensors is the distance of the wall, which is fed into the neural network. The neural network manipulates these values and does a bunch of functions on them.
Two outputs are given out:
- STEERING,
- ACCELARATION -> Which controls the car.
In order to train the car better we need the GENETIC ALGORITHM.
Program Logic:
50 random cars are generated that has a random neural network (each interprete the sensory data in a different way and do random stuff) When a car "dies" meaning in hit an obsticle, we select the ones that did the best and we combine (morph) them together in order to create a slightly more efficient car.
Example:
- We pick parent A and parent B
- We take a few weights from both and merged them together to create a child.
- The child "should" then do better the the generation before, so we pick 20 best cars from 50 randomly generated ones.
- Then we generate around 50 more cars using the top 20 parents.
Other factors that influence outcome:
- Mutation Rates
- Selecting the worst performers
