Evolutionary Local Search in Vehicle Scheduling: Cost Architecture, Constraint Modeling, and the Divide Among Theory and Enterprise Practice

Abstract

Freight routing at enterprise scale is a combinatorial decision problem whose solution space is too large for human planners to search manually and too complex for simple heuristics to navigate reliably. This paper studies Vehicle Scheduling and Routing (VSR) optimization, an algorithm-based approach used in modern ERP-integrated transportation systems to automate the search for optimal solutions. The paper traces the problem from its roots in the Vehicle Routing Problem with Time Windows (VRPTW) through the mechanics of evolutionary local search, paying particular attention to how cost functions are constructed, how hard and soft constraints interact, and how the semantics of distance-cost calculation differ between route-based and destination-based models. The paper draws on operations research literature for theoretical grounding while concentrating on practical questions: why does the optimizer behave as it does, how does configuration drive that behavior, and what does a logistics professional need to understand to deploy VSR optimization effectively? Field observations show that miscalibrated cost parameters are the leading cause of suboptimal optimizer output in production deployments.