Navigation in a simplified Urban Flow through Deep Reinforcement Learning