A dynamic routing protocol with payments for the Physical Internet: A simulation with learning agents

The Physical Internet is meant to dynamically route hundreds of thousands of loads a day all over the world. Each time a load arrives at a node that is not its destination, a next node, a carrier, and a price should be jointly selected to get the load closer to its destination.The dynamic nature of the routing system and the large number of loads to process in short periods of time are problematic for standard optimization processes. Fast and easy to run protocols, like those we see employed in the Digital Internet, must be designed. Using a game theoretic approach, we design a protocol whose core component is an auctioning process run each time a load arrives at a new node different from its destination. The protocol promises to be fast, reliable, and resilient, and aims to minimize the price of the total routing process by using concepts like administrative distance as used in Digital Internet routers.If the behaviors of the carriers are rational, then the proposed auction, when played only one time, is Incentive Compatible, Allocatively Efficient, Budget Balanced, and respects Individual Rationality. The final protocol, however, involves three intelligent agents - shippers, nodes, and carriers - whose action space and interactions are too complex for pen and paper analysis.As a first step to understand the performance of such a new system, we simulate an environment with only Full-Truck-Loads in a market composed of single truck companies. We assume simple behaviors for shippers, and we use a learning process for nodes and carriers. We show that the usage of such a protocol is extremely efficient for it significantly reduces total delivery costs as well as empty mileage.

Automated summary

The Physical Internet aims to dynamically route and auction loads to efficiently deliver them to their destinations while minimizing costs. However, due to the complexity of the system, simulations and learning processes are needed to evaluate its performance.