Last-mile delivery is increasingly a decisive step in optimizing the entire supply chain. The increasing prevalence of home deliveries, especially in urban areas, calls for greater efforts to improve logistics management and increase efficiency. In recent years various strategies and technologies are being studied such as the use of crowd logistics, autonomous unmanned aerial vehicles, autonomous delivery robots. In particular, the use of autonomous robots can be improved with route optimization; it is one of the processes that can improve performance in last mile delivery. The proposed study focuses on adopting autonomous vehicles in a context of pedestrian area with a methodology based in the combination of two approaches: use of simulation urban mobility software considering real traffic parameters and artificial intelligence through machine learning technique, a deep reinforcement learning to improve routing of autonomous robot routes in good delivery through autonomous learning. The methodology is tested by modeling a real urban pedestrian area in Italy through a simulated case study. The results show an improvement in logistics performance and future developments to analyze last-mile delivery with a fleet of autonomous robots.