Live Website : train-your-ai-driver
demo.mp4
-
Neural Network trains itself through parallel cars trying to manoeuvre through the traffic.
-
The car with the sensors is the current best car in the simulation i.e. the car that has made the most progress through the traffic.
-
Restart the training simulation with parallel cars mutated from your best car.
-
Dump the current best car if you think its next generation models cannot successfully manoeuvre through the traffic.
-
The 5 sensors are the inputs to the neural network. Their activations can be seen in the bottom-most layer of the neural network animation.
-
The controls of the car (up, down, left, right) are the outputs of the neural network. The triggering of the controls can be seen in the top-most layer of the neural network animation.
-
Modify number of lanes on the road
-
Modify the hidden layers in the neural network and watch the cells in the layers of the network get triggered in the animation
-
Add dummy traffic cars wherever you want to
To start the app in localhost (default port is 3000) :
npm start