Autonomous Driving System based on Deep Q Learnig

This paper deals with the simulation results of an autonomous car learning to drive in a simplified environment containing only lane markings and static obstacles. Learning is performed using the Deep Q Network. For a given input image of the street captured by the car front camera, the Deep Q Network computes the Q values (rewards) corresponding to the actions available to the autonomous driving car. These actions are discrete angles through which the car can steer for a fixed speed. The autonomous driving system in the car enforces the action that has the highest reward. Our simulation results show high accuracy in learning to drive by observing the lanes and bypassing obstacles.