New planning benchmarks: vehicle routing problem variants

The Vehicle Routing Problem (VRP) is a classical problem in Operations Research. It is a generalisation of the Travelling Salesman Problem in which a set of locations must be visted by one of a number of vehicles. All vehicles start and end at a depot location. The aim of the problem is to minimise the total distance covered by the trucks. There are many different variants of the VRP. For example, Capacitated VRP, VRP with Time Windows, VRP with Multiple Depots and VRP with Pickup and Delivery. All of these variants are interesting planning problems, as they deal with resource management, time windows, routing, etc. In this work, we model some of the variants of VRP in PDDL. We provide this as an intersting new benchmark set for planning. Already, many planning benchmarks are based in some way on transportation. Modelling VRP directly allows access to a great quantity of existing benchmark instances. This has the benefit of posing difficult and realistic challenges to existing planners.