harvitronix / reinforcement-learning-car

Using reinforcement learning to teach a car to avoid obstacles.
MIT License
488 stars 202 forks source link

Sonar with cat #4

Closed harvitronix closed 8 years ago

harvitronix commented 8 years ago

This PR changes the main behavior of the program from reading pixel colors to simulating a "sonar" experience, where the state is actually three distance readings, rather than a bunch of 0s and 1s.

It also adds some enhancements like moving obstacles (and a fast moving "cat") to combat over-fitting.