Solving OpenAI's reinforcement learning CarRacing environment
In this project, a python based car racing environment is trained using a deep reinforcement learning algorithm to perform efficient self driving on a racing track. A deep Q learning algorithm is developed and then used to train an autonomous driver agent. Different configurations in the deep Q learning algorithm parameters and in the neural network architecture are then tested and compared in order to obtain the best racing car average score over a period of 100 races. According to OpenAI Gym, this environment is considered solved when the agent successfully reaches an average score of 900 on the last 100 runs.
A video with the final car's performance can be seen here: https://www.youtube.com/watch?v=jbdjhoDT41M
A video of the car training can be seen here: https://youtu.be/C9CZpbuOz04