AI-Based Approaches for Autonomous Vehicle Fleet Optimization and Management

Abstract

The general goal of intelligent transportation systems (ITS) is to enhance transportation network efficiency through congestion reduction, accident prevention, fuel consumption reduction, and carbon dioxide emission reduction. Traditionally, ITS have been designed to address single problems: real-time traffic-density-based signal control to reduce congestion, incident detection to reduce accidents, fare optimization to increase efficiency of public transportation, and so forth. This modus operandi is slowly being replaced by a holistic view of the overall system—as machine learning and AI methodologies prove capable of handling multi-scale, multimodal, stochastic and non-stationary data. Fleet management is a system of systems, and autonomy and intelligence can be weaved into every module of the system: Vehicles must be recharged, parked, and cleaned; they should be reallocated to maximize utility given the stochastic distribution of end-users and to reduce idle time that could be used for system maintenance or energy storage. Autonomous fleet management is already a cutting-edge and increasingly fertile spot for AI and ITS research interplay [1].

Autonomous vehicles (AVs) are a catalyst for major changes in transportation. The removal of driving constraints, such as human resources, opens up multiple new business opportunities around the existing business models of public or freight transportation. In the case of an on-demand taxi service, the autonomous fleet owner maximizes profit by carefully selecting the demand segment to pursue: Urban taxi markets might be more dynamic compared to long-haul haul freight transportation with more predictable, repetitive patterns of demand. Therefore, while the problem is the same, the models of demand, costs, and constraints might vary for the two markets [2].