This grant will support research investigating how a real-time marketplace might be constructed to match freight shipments with available carriers. In such a system, freight shippers make a request to the system manager, and an automated service will match shipping jobs to carriers who are likely to accept them. The benefits of such a system include more efficient use of freight capacity (fewer empty miles) and reduced shipping prices, resulting in economic benefits, reduced fuel consumption and emissions, and reduced congestion. These benefits will directly contribute to economic prosperity in the nation. This award supports the basic research needed to design and implement such a system, particularly focusing on the matching mechanism between shippers and carriers, and methods for quickly estimating the additional cost carriers will incur should they agree to a match. More specifically, the research work involves developing new algorithms in these areas, and testing and validating these algorithms with real-world freight data. In addition to benefits in the trucking industry, broader economy, and reduced urban congestion, the education and outreach activities will provide benefits to diverse K-12 and university student populations.
The research approach further extends traditional algorithms and heuristics for stochastic multi-armed bandit problems and the vehicle routing problem. A freight match problem is characterized by the arrival of stochastic demands that are queued for service and the need to estimate the costs a match would impose on a carrier, which serves as reserve prices for carriers for particular requests. The model represents this by extending the traditional multi-armed bandit problem to allow the set of "arms" to vary over time, as carriers enter and leave the system. Different from a traditional vehicle routing problem, efficiently and robustly estimating changes in routing costs from incorporating freight jobs is more important in the freight matching problem than deriving detailed routing plans at a fine-grained level. The research team will investigate new algorithms in both of these areas, implementing them in a simulation framework and testing them with real-world data, to test the hypothesis that the queueing bandit approach can outperform auctions and alternative market mechanisms.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
