Abstract

In the last few decades, the continuous growth of travel demand increased urban traffic congestion. The situation demands an efficient way of handling resources like roads and vehicles. Parallelly, the advancement of technology proposed more innovative forms of shared mobility, such as flexible pods. These autonomous, modular vehicles possess the unique capability to autonomously merge and split during their journeys, facilitating en-route transfer of passengers. Formulating a precise mathematical model that comprehensively captures the intricacies of such a complex system is quite challenging. This paper introduces an alternative approach rooted in Reinforcement Learning that circumvents the need for a predetermined mathematical model. Instead, it enables the system to learn its dynamics through interactions with the environment. The primary goal of this research is to acquire an optimized passenger transfer policy. The experimental findings indicate that the transfer-based pod service outperforms private taxi services and Demand Responsive Transport services, boosting vehicle utility by 18% and 8%, respectively. Remarkably, the pod service achieves these gains with only one-third of the vehicles required for private taxis and 13% fewer vehicles than the Demand Responsive Transport service. However, it's important to note that the transfer-based pod service does result in a modest increase in passenger travel time, ranging from 2% to 13%. Therefore, the transfer-based pod service offers a promising opportunity to alleviate traffic congestion while accepting a compromise on passenger travel times.

Full Text
Published version (Free)

Talk to us

Join us for a 30 min session where you can share your feedback and ask us any queries you have

Schedule a call