Informed versus non-informed taxi drivers: agent-based simulation framework for assessing their performance

Abstract

Data driven research is becoming a standard in Transport. Recent advances in the Artificial Intelligence and Machine Learning related areas enable the possibility of automatically generating highly-accurate predictive analytics frameworks, under any context. Such frameworks can potentially provide unprecedented levels of information to all mobility actors regarding not only the current but also the future status of network variables – such as Origin-Destination flows. This fact elevates the decision support to a new standard, where operations can be optimized in real-time and in near-autonomous fashion. However, such advances also bring new questions: How much can a transport operator benefit from this? Is there a limit for the amount of information that all actors should have? This paper aims to answer such questions by introducing an agent based model able to simulate the behavior of individual taxi drivers on their passenger-finding strategies. Multiple strategies are proposed and compared through exhaustive computer-aided simulations. The goal is to find how different drivers will benefit from the availability of accurate information about the future spatiotemporal demand distribution. The experiments were conducted using real-world operational data collected from a large scale taxi fleet operating in Thessaloniki, Greece. The obtained results illustrate different perspectives of the cost-benefit tradeoff on disseminating future demand-related information at different scales and ratios.