Urban consolidation centers (UCCs) have a key role in many initiatives in urban logistics, yet few of them are successful in the long run. The high costs often prevent attracting a sufficient number of UCC users. In this paper, we study sustainable business models and the supporting role of administrative policies. We perform an agent-based simulation applied to the city of Copenhagen and collect data from a variety of sources to model the agents. Both the data and case setup are validated by means of expert interviews. We test 1,458 schemes that combine several administrative measures and cost settings. Most schemes yield significant environmental benefits; many of them reduce the truck kilometers driven by about 65% and emissions by about 70%. The key challenge is to identify schemes that are also financially sustainable. We show the importance of committing carriers to the UCC as soon as possible, as carriers potentially generate the bulk of the revenue. Subsequent revenues may be generated by offering value-adding services to receivers. Based on the numerical experiments, we pose various propositions that aid in providing favorable conditions for a UCC, improving its chances of long-term success.

Smart modular freight containers -as propagated in the Physical Internet paradigm -are equipped with sensors, data storage capability and intelligence that enable them to route themselves from origin to destination without manual intervention or central governance. In this self-organizing setting, containers may autonomously place bids on transport services in a spot market setting. However, for individual containers it might be difficult to learn good bidding policies due to their short lifespan. By sharing information and costs between one another, smart containers can jointly learn bidding policies, even though simultaneously competing for the same transport capacity. We replicate this behavior by learning stochastic bidding policies in a semi-cooperative multi-agent setting. To this end, we develop a reinforcement learning algorithm based on the policy gradient framework. Numerical experiments show that sharing solely bids and acceptance decisions leads to stable bidding policies. Real-time system information only marginally improves performance; individual job properties suffice to place appropriate bids. Furthermore, we find that carriers may have incentives not to share information with the smart containers. The experiments give rise to several directions for follow-up research, particularly addressing the interaction between smart containers and transport services in selforganizing logistics.

Abstract. We study a dispatch problem with uncontrolled batch arrivals of LTL orders at an urban consolidation center. These arrivals reflect the delivery of goods by independent carriers. The specific order properties (e.g., destination, size, delivery window) may be highly varying in city logistics, and directly distributing an incoming batch may yield high costs. Instead, the hub operator may decide to wait for incoming batches that allow for more efficient distribution. A waiting policy is required to decide which orders to ship and which orders to hold. We model the dispatching problem as a Markov decision problem. Dynamic Programming (DP) is applied to solve toy-sized instances. To solve realistic instances, we propose an Approximate Dynamic Programming (ADP) approach. Through numerical experiments, we show that the ADP approach closely approximates the optimal values of DP for small instances, and outperforms two benchmark policies for larger instances.

This paper addresses the planning of freight dispatch in flexible transport networks featuring multiple carriers. To deal with the computational challenges of the planning problem, we develop an Approximate Dynamic Programming (ADP) algorithm that utilizes neural network techniques to learn dispatch policies. We test whether dispatch policies learned autonomously by carrier agents (based on local information) match the quality of policies learned by a central planner (based on full network information). Numerical experiments show that the policies yield solutions of comparable quality for small instances, yet the decentralized approach is capable to scale to larger instances. Finally, the ADP policies are compared to four benchmark policies, which are all significantly outperformed.

No abstract