A general modeling and simulation framework for dynamic vehicle routing

TL;DR

This paper addresses the need for flexible evaluation tools in dynamic vehicle routing problems (DVRPs) by introducing a general modeling framework that supports online decision making and a discrete-event based framework that blends simulation with policy execution. It presents and open-sources dvrpsim, a Python-based DVRP simulator that is easily customizable to model diverse problems such as dynamic pickup-and-delivery, same-day delivery, and restaurant meal delivery, with support for postponement, decision points, and en-route constraints. The contributions include a literature-driven formalization of DVRP aspects, a general modeling framework that can represent DPDPs, SD-DPs, and VRPDSRs, and a practical tool for dynamic testing of routing policies. The work enables researchers to rigorously compare policies under realistic, dynamic conditions and provides a community resource for reproducible DVRP experimentation with broad applicability to logistics and service systems.

Abstract

In dynamic vehicle routing problems (DVRPs), some part of the information is revealed or changed on the fly, and the decision maker has the opportunity to re-plan the vehicle routes during their execution, reflecting on the changes. Accordingly, the solution to a DVRP is a flexible policy rather than a set of fixed routes. A policy is basically a problem-specific algorithm that is invoked at various decision points in the planning horizon and returns a decision according to the current state. Since DVRPs involve dynamic decision making, a simulator is an essential tool for dynamically testing and evaluating the policies. Despite this, there are few tools available that are specifically designed for this purpose. To fill this gap, we have developed a simulation framework that is suitable for a wide range of dynamic vehicle routing problems and allows to dynamically test different policies for the given problem. In this paper, we present the background of this simulation tool, for which we proposed a general modeling framework suitable for formalizing DVRPs independently of simulation purposes. Our open source simulation tool is already available, easy to use, and easily customizable, making it a useful tool for the research community.

Figures (7)

• Figure 1: Differences between the sequential and the discrete-event based decision process. Circles refer to distinct events (e.g., order requests, vehicle arrival). Black circles refer to decision points. Squares refer to states. Black squares refer to post-decision states.
• Figure 2: Vehicle operations between two consecutive departures.
• Figure 3: Sketch of the discrete-event based decision process.
• Figure 4: Events are inductive, meaning that processing one event can cause several new events to be added to or removed from the event queue.
• Figure 5: Selected states from the following scenario: ($s_0$) Vehicle $v$ is located at location $l_1$. ($s_1$) Order $o_1$ is requested. ($s_2$) Decision point is imposed. ($s_3$) Decision maker creates the route plan. ($s_4$) Order $o_2$ is requested. ($s_5$) Decision point is imposed. ($s_6$) Decision maker updates the route plan. ($s_7$) Vehicle is departed. ($s_8$) Order $o_2$ is requested. ($s_9$) Decision point is imposed. ($s_{10}$) Decision maker updates the route plan.