evo-racing

This is a race simulation that uses genetic algorithm (GA) to improve the performance of the cars. A generation ends after 60 seconds or after no car has improved its progress in more than 5 seconds. Then the scores are computed and GA runs crossover and simulation to create a new generation of racers.

Cars have 5 sensors that detect walls and output a higher value the closer the wall is to the car. These outputs, along with the car linear and angular speed,  enter a shallow (3 layers) neural network that outputs the acceleration and steering inputs.

Controls:

• Move camera: WASD or arrows
• Zoom in/out: scroll wheel
• Change camera mode (free or follow first): C
• Pause/unpause: Space bar
• 1x speed: 1
• 3x speed (highly unstable): 2
• 10x speed (garanteed broken): 3