Simulating sonar

This version strays from the first version in a couple ways:

1. Instead of a bunch of 0/1 pixel readings, the state is instead 3 distances, which is a step towards a real-world scenario where we'll use sonar to detect the state of the room.
2. Obstacles are no longer randomly created, but they now move in an effort to combat over-fitting.

This is a bit of a "step" version, as in the next version, we'll be focused entirely on trying to apply this simulation to a real RC car.

Original version of Robocar

This is the code used in parts 1 and 2 of the series of blog posts on Medium.

Part 1: https://medium.com/@harvitronix/using-reinforcement-learning-in-python-to-teach-a-virtual-car-to-avoid-obstacles-6e782cc7d4c6

Part 2: https://medium.com/@harvitronix/reinforcement-learning-in-python-to-teach-a-virtual-car-to-avoid-obstacles-part-2-93e614fcd238#.vbakopk4o