A vehicle routing problem with dynamic demands and restricted failures solved using stochastic predictive control

In this paper, we describe a vehicle routing problem representing real-time route scheduling for an agent delivering services to customers where the service demands are dynamically evolving. A failure is defined as an occurrence of service demand overload at a customer location. A stochastic predictive control algorithm applied to a partially observable Markov decision process model is used to solve the routing problem, minimizing the service agent's travel effort while maintaining the occurrence rate of failures to be low.