Many studies have demonstrated the effectiveness of machine learning for solving vehicle routing problems (VRPs) and other combinatorial optimization tasks. Because exact solutions are often intractable, the design of strong heuristics is essential, and machine learning provides a promising tool for this purpose. Current research has concentrated mainly on relatively simple and well-studied cases, such as the Traveling Salesman Problem (TSP) and the Capacitated Vehicle Routing Problem (CVRP). Our goal is to extend machine learning approaches to more realistic VRP settings. In particular, we focus on hierarchical routing problems, such as the Multi-Depot Vehicle Routing Problem (MDVRP) and the Capacitated Location Routing Problem, where customer-to-depot assignment precedes route optimization. We also consider energy-constrained VRPs, such as the Electric CVRP (ECVRP), which require explicit modeling of charging decisions within routes.