Developing self-driving vehicles is one of the most challenging tasks in machine learning nowadays. Current supervised learning approaches can achieve state-of-the-art results with an adequate dataset. However, collecting a representative dataset for a self-driving solution is a challenging task. Therefore, the possibility of training the self-driving models in a simulator and applying the trained model in the real-world is a promising approach. The task is to train a deep reinforcement learning or a deep imitation learning agent in the simulator of the Duckietown self-driving environment (https://github.com/duckietown/gym-duckietown), and evaluate its behaviour in both the simulator and the real world.