Using Reinforcement Learning for the Three-Dimensional Loading Capacitated Vehicle Routing Problem

The 3L-CVRP is a long-standing problem in the operations research literature. However, these existing approaches have two shortcomings. First, existing approaches are based on heuristics that are limited by the quality of the chosen rules and need to be adapted to new scenarios by domain experts. Second, existing approaches require convergence towards near-optimal solutions, which is computationally expensive.