My research aims at modeling and testing algorithms that make use of anticipation in solving dynamic and stochastic vehicle routing problems.
By dynamic, I mean vehicle routing problems in which new customers can (and are likely to) appear and request for a service, even though the vehicles already started their routes and possibly serviced some requests. As an example, we could consider an on-demand public transportation system: provided a fleet of minibuses capable to transport a limited number of passengers at any moment, which are the vehicle actions that maximize the expected number of satisfied requests by the end of the operational day, knowing that every passenger's request comes with a timing constraint?
By stochastic, I mean problems for which we are given a probabilistic knowledge concerning the uncertain part of the data. In our former example, the uncertainty concerns on the presence or not of each customer, and on the moment at which he or she makes a request (if it does). Fortunately, every potential customer comes with a probability distribution on the moment at which its request, or presence, is revealed. By taking that stochastic information into account in the mathematical models we optimize, we're in theory (but also in practice) able to create anticipation.