Multi-Agent Scheduler for Rent-a-car companies Slava Andreev Magenta Technology 349 N-Sadovaya St. Samara 443125 Ru +7846 342 51 74 Andreev@magenta- technology.ru George Rzevski Magenta Technology 130 Shaftesbury Aven London W1D 5EU, UK +44 (0) 20 71 49 37 50 george.rzevski@ma genta- technology.com Peter Shveykin Magenta Technology 349 N-Sadovaya St. Samara 443125 Ru +7846 342 51 74 Shveykin@magenta- technology.ru Peter Skobelev Magenta Technology 349 N-Sadovaya St. Samara 443125 Ru +7846 342 51 74 Skobelev@magenta- technology.ru Igor Yankov Magenta Technology 349 N-Sadovaya St. Samara 443125 Ru +7846 342 51 74 Yankov@magenta- technology.ru ABSTRACT The paper gives overview of multi-agent dynamic scheduling solution for rent-a-car business domain. Specific requirements for cars and drivers scheduling, developed multi-agent and ontology- based approach for real time scheduling, systems architecture and performance measurements are presented. Key design decisions and results of first stage of developments are discussed. One of most important results is the system ability to schedule complex interdependent works of many resources and update generated schedules by events in real time. Developed multi-agent approach can be applied for solving complex real time scheduling and optimization problems for many applications Categories and Subject Descriptors H.4.2 [Information Systems Applications]: Type of Systems - decision support, logistics General Terms Management, Measurement, Design, Economics, Performance Keywords Dynamic scheduling, real-time planning, multi-agent systems, microeconomics, ongoing optimization, events, criteria of decision-making. 1. INTRODUCTION The problem of resources scheduling in real time is one of the most difficult in the modern theory of optimization [1-3]. In particular, when it is required to create schedules of many participants which are connected with each other, and changes in the schedule of one participant cause changes in the schedules of other participants. In the present paper we will consider an example of the scheduling solution for rent-a-car company based on multi-agent approach. In our approach the output schedule is the result of the distributed decision-making processes which are constantly running by agents of demands and resources with conflicting interests. The approach developed allows to solve a complex problem of cars distribution to rentals in real time for the whole network of stations under conditions of high uncertainty and dynamics of demand and supply and with other important advantages. Now the system developed is already delivered to rent-a-car company and successfully works in first pilot stations. It is expected to introduce the system in nearby 25 stations of this company till the end of the year and in the beginning of the next year. C G In the paper we will consider in more details specific requirements for solving rent-a-car scheduling problem, developed multi-agent approach to the solution of this problem, architecture of delivered system and main features of its implementation, including performance measuring. Also it will be shown that the developed approach can be also useful for a wide range of other applications in transport and production logistics, and other industrial applications. 2. THE PROBLEM DOMAIN OF RENT-A- CAR SCHEDULING At a glance rent-a-car business assumes simple scheduling process: the client specifies group of the desirable car, time and a place where he would like to take it, and also a place and time where he would like to leave it after the rent term expiry. However, to solve this problem it is required to make a number of decisions, from what station to take away the car looking on preferences of the client, distances and costs and presence of cars at stations, what driver to schedule to wash and deliver car accordingly with the business processes of the company, how much driver overtime it will require, etc. As a result to solve the problem just by taking available car from nearest station is not possible. And the problem becomes even more complex because it is also necessary to deliver allocated driver to the chosen car, and after delivery of the car this driver needs to be returned back on station or to direct to another rental or to home. Very often, in turn, for this purpose another car and another driver are required and so on. As a result to solve the problem is necessary to make drivers consolidation for such trips with optimization of routes. Besides, there are a number of unforeseen events in such networks: new rentals arrive and some of them vary or even cancel during the work, the car can break, the driver can become ill or takes day-off suddenly, driver can be late in a traffic jam, etc. Usually the territory of rent-a-car business breaks into set of small regions in which a number of stations are functioning where the rental information is arriving, where cars are accumulating and being served and drivers are beginning the working day. At each station there is a special personnel which is engaged in manual generation of the schedule for drivers and cars. Usually this work is carried out by skilled managers who, considering set of factors, define what car, to what client should be given, and also what driver when and where should take away or deliver the car etc. Besides, these managers can to co-operate both with clients, and with drivers, flexibly correcting schedules in process of situation changes at stations. ite as: Multi-Agent Scheduler for Rent-a-car companies, S.Andreev, .Rzevski, P.Shveikin, P.Skobelev,I.Yankov, Proc. of 8th Int. Conf. on Autonomous Agents and Multiagent Systems (AAMAS 2009), Decker, Sichman, Sierra, and Castelfranchi (eds.), May, 10–15, 2009, Budapest, Hungary, pp. XXX-XXX. Copyright © 2009, International Foundation for Autonomous Agents and Multiagent Systems (www.ifaamas.org). All rights reserved.