Design and Realization of Autonomous Cars Using Deep Q Learning

Danish Ahmad Khan, Sameena Naaz and Farheen Siddiqui

Self driving cars, one of the most acclaimed technologies of the 21st century after the internet, are actually a bone of contention amongst the orthodox drivers and the iconoclastic spearheads of innovation. With the evolution of advanced algorithms such as reinforcement learning algorithm and Q-learning the world of artificial intelligence has taken a big leap forward. Just as a child learns what is right and what is wrong this algorithm also imitates that way of nature to categorize things that should be done and that should not through a system of reward points and negative points. This paper focusses on how to implement this reinforcement algorithm in a real world like scenario in the self-driving car. In this work the bellman equation is used for giving rewards for certain action and the Markov decision processes is used for the decision-making process which includes a certain degree of randomness in the self-driving car. As we humans make mistakes and learn from them, similarly the car here also learns from the wrong decisions made earlier. This paper also discusses the compromises the car has to make to reach the goal, just as we humans do and it's adeptness at that.

Volume 11 | 04-Special Issue

Pages: 1808-1813