ABSTRACT

The transportation sector is witnessing rapid development of autonomous vehicle (AV) technology. While an AV can be more energy efficient than a conventional human-driven vehicle, their environmental impacts at the fleet and city level could be either significantly better or worse than the traditional systems, depending on how people use them " adopting AVs as privately-owned AVs (PAV) or centrally-managed shared AVs (SAV) will result in very different fleet size, vehicle-miles-travelled (VMT), and carbon emissions. To understand the environmental impacts of AVs at the city level, it is critical to consider who are likely to adopt which types of AVs, their travel demands, and the associated AV operation. Previous studies evaluating the potential impacts of AVs on transportation system and the environment are limited by the existing travel demand models, which do not have sociodemographic information linked to the travel demands to support adoption modeling or only generate trip origin and destination at the zonal level that is insufficient to support shared AV modeling. Additionally, existing research mainly focused on SAV systems and did not consider the potential competition between SAV and PAV. It is necessary to compare the system performance between the privately-owned AV system and the centrally-managed shared AV system and under the scenarios that both systems co-exist to inform AV system development. Furthermore, although AVs can help reduce fleet size, each vehicle will be used more intensively and accelerate vehicle replacement. To fully quantify the environmental impacts of a citys AV system, it is important to take a life-cycle perspective, considering not only vehicle use but also upstream vehicle manufacturing, downstream vehicle disposal, and fleet replacement.

To address these gaps, this work proposed an integrated agent-based model to quantify the environmental impacts of private and shared autonomous vehicles. The integrated model includes four key components: 1) a travel demand generation model that links high resolution individual and household travel demand with social-demographic information, 2) an AV adoption model that evaluates individuals and households likelihood for AV adoption and preferences for AV use, 3) an AV operation model to simulate the operation needs for different AV fleets, and 4) an AV life cycle model that assesses the simulated AV systems emissions considering vehicle replacement. Applying this integrated model to a case study of Miami City, the results have showed that 1) existing travel demand generation methods cannot fully meet the AV modeling needs and may overestimate the environmental benefits of AV systems, 2) integrating heterogeneous individual adoption behaviors with AV operation models is necessary to evaluate the system performance of AVs without overestimates, 3) AVs can reduce the system-level life-cycle emissions by a quarter to nearly half, where the market penetration of AVs plays an important role, and 4) compared with PAVs, SAVs are more environmentally beneficial but less likely to be adopted. The proposed models and findings of this work can inform decision making for SAV operators, policy makers, and transportation planners.