Selfdriving car

A neuroevolutionary self driving (software) car.

Components

Car

• Direction
• Speed (constant?)
• Sensors
• Connected to an ANN
• Every step that it makes while driving flows trough the ANN

Road

• Show cars
• Detect collisions
• Indicate distance traveled by each car

Artificial Neural Network

• Nodes and weights.
• Variable weights
• Input: data from car sensors
• Output: value between -1 (60 degrees to the left) and 1 (60 degrees to the right)

Evolver

• Gets the ANN, tests and reproduces them:
  • Makes 50 cars with random weights
  • Applies the ANN to the cars and measure their distance
  • Selects the 25 best cars
  • Makes 25 new cars with weights inherited from the top 25
  • Applies a mutation
  • Replaces bad 25 with good 25
  • Repeat