Automated Valet Parking using Double Deep Q Learning

An important parameter in the field of Autonomous Vehicles (AV) is the ability of the vehicle to park itself. Parking is a analysis and intelligence in that task which requires the AV to take in information about the availability of a spot, its dimensions, the dimensions of the vehicle and how to maneuver the vehicle in order to fit the car into the parking space without major accidents and collisions. But when reduced to the most basic form it is a control task. Reinforcement learning has been gaining a lot of interest in recent times in its ability to optimize control tasks in real world environment. Therefore, in this paper we apply Deep reinforcement learning algorithm to design an agent that would be able to park a car in a given parking environment whose information is already present with the AV. Double Deep Reinforcement learning has been successfully applied to the problem and the AV was able to park successfully 95% of the times and was able to learn this in 24 hours