Ant Colony Optimization for a stochastic vehicle routing problem with driver learning

Motivated by an industry project with a small package shipping company in France, we study a vehicle routing problem with stochastic travel and service times that considers the influence of driver familiarity with routes and customers on routing efficiency. Our approach forgoes any fixing of delivery areas thus maintaining routing flexibility. Driver specific travel and service times give drivers incentives to stay in familiar areas. Following common practice, we consider delivery deadlines instead of time windows. To solve the routing problem, we develop an Ant Colony Optimization (ACO) method due to its robustness when dealing with stochastic problem parameters. Our ACO approach includes a new indicator value to deal with customers that are hard to integrate into tours because of delivery restrictions. Numerical studies show that our algorithm is able to trade off between driver learning and routing flexibility and performs strongly for most types of test instances.