! Repository moved to new address !

Ekin Akyürek, Afra Feyza Akyürek (2018)

Koç University Spring 2018 | Senior Design Project

Best Senior Design Project Award

[Video] [Report]

Self-driving cars is one of the popular trends of the recent years. Current developments in deep learning have given away limitless opportunities to create artificially intelligent systems. Deep reinforcement learning is widely used to train computer agents to succeed in complicated tasks in simulated environments. The primary goal of this project is to create an AI agent that can learn to drive a car using sensor inputs without colliding other objects (or cars) in a simulated environment using deep reinforcement learning. A simulation environment was created in Unity3D along with a neural network toolbox developed by the team in native C#. Eventually, the car agent successfully completed tracks at different difficulty levels, avoided from randomly appearing objects that it had never seen before and finally demonstrated an outstanding performance in a multi-car environment.