Decentralized Combinatorial Auctions for Dynamic and Large-Scale Collaborative Vehicle Routing

While collaborative vehicle routing has a significant potential to reduce transportation costs and emissions, current approaches are limited in terms of applicability, unrealistic assumptions, and low scalability. Centralized planning generally assumes full information and full control, which is often unacceptable for individual carriers. Combinatorial auctions with one central auctioneer overcome this problem and provide good results, but are limited to small static problems. Multi-agent approaches have been proposed for large dynamic problems, but do not directly take the advantages of bundling into account. We propose an approach where participants can individually outsource orders, while a platform can suggest bundles of the offered requests to improve solutions. We consider bundles of size 2 and 3 and show that travel costs can be decreased with 1.7% compared to the scenario with only single order auctions. Moreover, experiments on data from a Dutch transportation platform company show that large-scale collaboration through a platform results in system-wide savings of up to 79% for 1000 carriers.