Engineering Optimization Vol. 37, No. 1, January 2005, 1–27 An asexual genetic algorithm for the general single vehicle routing problem PARTHA CHAKROBORTY* andARIJIT MANDAL Department of Civil Engineering, Indian Institute of Technology, Kanpur 208 016, India (Received 30 September 2003; in final form 16 March 2004) This article proposes an optimization algorithm for the general vehicle routing problem. The algorithm uses mutation based genetic algorithms (GAs) (or asexual GAs). The algorithm is general, in that it can handle various types of the vehicle routing problem; namely, the traveling salesman problem, the single vehicle pick-up and delivery problem, and the single vehicle pick-up and delivery problem with time windows. The algorithm is fast and gives optimal/near-optimal solutions with minimal computation effort. The algorithm is simple and easy to implement. Results from forty-six problems (most of which are obtained from existing sources) are also presented. Keywords: Vehicle routing; Asexual genetic algorithms 1. Introduction The single vehicle routing problem (SVRP) is an umbrella description of three classic problems in operations research and transportation engineering; namely, (i) the traveling salesman prob- lem (TSP), (ii) the single vehicle pick-up and delivery problem (SVPDP), and (iii) the single vehicle pick-up and delivery problem with time windows (SVPDPTW). These problems have one thing in common—in each of them a route (which is optimal in some way) is to be determined such that a vehicle starting from a depot and following that route will touch each of the nodes (distributed in a two-dimensional space) once and return to the depot. The criteria defining optimality range from shortest route length to shortest riding time of pas- sengers and depend on the specific problem type. There are, of course, other characteristics that make each of the above problems different from one another. In the next section, these problems are described in detail. Traditionally, these problems have proven to be difficult and require large computation effort to give good solutions (which are not necessarily the optimal) for even moderate problem sizes (say, 50 or more nodes). Further, despite the basic similarity of the problems, there are no generalized algorithms which can solve all the three types of problems. The purpose of this * Corresponding author. Email: partha@iitk.ac.in Engineering Optimization ISSN 0305-215X print/ISSN 1029-0273 online © 2005 Taylor & Francis Ltd http://www.tandf.co.uk/journals DOI: 10.1080/03052150410001721468