Reducing urban private vehicle dependency with shared autonomous vehicles: Agent-based simulation as a digital testbed for transport policy evaluations

Conventional Private Vehicles (CPVs) have long been a cornerstone of modern society by offering con-
venience and flexibility for personal mobility. However, the overreliance on CPV has given rise to many
problems and challenges to urban transport networks, such as air pollution, transport noise, traffic conges-
tion, traffic accidents and parking shortages. In response, many cities have implemented policies aiming
to reducing CPV usage. Yet, the current alternative modes, such as Public Transit (PT) and cycling, often
cannot offer the same level of comfort, convenience and flexibility as CPVs. Consequently, despite these
efforts, congestion and pollution persist in most metropolitan regions.
With the developments of self-driving technology, Shared Autonomous Vehicles (SAVs) have emerged
as a paradigm shift in urban mobility by providing on-demand and door-to-door transportation services.
Since SAV could provide urban commuters with comfort and convenience comparable to CPVs but at a
significantly lower cost than CPV ownerships, it holds great promise for reducing urban CPV reliances
towards cities that are more accessible, inclusive and sustainable. With SAVs being tested in various
pilot sites worldwide by 2023, investigating the impacts of SAVs under different operational strategies
is crucial for urban planners and policymakers to prepare for such a transformative technology. There-
fore, this research opportunity gave rise to the research objective of this thesis, which aims to answer the
following questions: How to generate a digital testbed for different SAV operational strategies and
policy evaluations on the macroscopic urban level? Based on the testbed, to what extent can SAVs
be integrated into current metropolitan regions’ mobility systems to eliminate the usage of CPVs,
and what will be the impact on society as a whole?
First, in order to generate the digital testbed that aligns with this thesis’ research objective, a sys-
tematic literature review is conducted for comprehensively assessing the state-of-the-art developments in
the research landscapes and detailed modelling specifications (Chapter 2). The outcome suggests that
Agent-Based Model (ABM) is more suitable for becoming a digital testbed of AV policy evaluation than
other conventional modelling algorithms by simulating the dynamic interactions between AVs and other
mobility stakeholders. Regarding the simulation outcomes, AV will bring significant opportunities and
potential challenges that may mitigate the advantages. In general, carpooled SAV combined with intel-
ligent operational strategies is expected to bring the most benefits to the overall society. Still, studies
in different regions demonstrate that the impacts of SAVs vary between areas, so a simulation based on
the city’s own characteristics will provide a more comprehensive estimation of the impacts of SAVs on
respective networks.
Then, in Chapter 3, we develop an ABM generation pipeline that generates high-quality ABM that is
more representative of real-life travel patterns, including the incorporation of travel demand by external
travellers who hold trips that originate from or head towards locations external to the study area, plus
allowing multiple destinations for the same kind of activity in activity chaining. Moreover, the input
of our pipeline contains social-spatial data and cellphone origin-destination matrices, which adhere to
individual privacy regulations and show significant transferability of our generation pipeline due to the
ubiquitous nature of mobile phones being adopted globally. Consequently, our pipeline provides an
effective alternative for ABM generation in regions facing data scarcity. With the proposed pipeline, a
representative ABM can be built much more efficiently compared to the present. Meanwhile, thanks to
the modular design of our pipeline, the open-source code holds vast flexibility for module replacements
or standalone adoption to other research questions. With this research, we wish to decrease the entry
level of future studies using ABMs and contribute to research in transport simulation and relevant fields
towards more accessible and reproducible modelling.
Next, based on the research objectives, Brussels, the capital of Belgium, is selected as the case study
region due to its recent efforts to eliminate urban CPV usage towards more sustainable mobility networks.
Consequently, a MATSim model for Brussels is generated in Chapter 3 based on the proposed generation
method. The MATSim Brussels scenario shows a plausible match with the real-world traffic data and
serves as the foundation of this thesis for SAV policy evaluations to eliminate urban CPV reliance.
Finally, SAVs are introduced as a new mode in the generated MATSim Brussels scenario. The service
is operated only within the Brussels Capital Region. To test the impacts of substituting CPV trips, we cat-
egorise current CPV trips on the Brussels network into two types: trips that origins and destinations both
within Brussels are named ‘internal trips’, whereas the remaining are ‘external trips’. Substituting these
trips towards more sustainable modes (SAVs or PT) follows a progressive order: We start by replacing
the trips of agents who do not hold any external CPV trips during the simulated day (Chapter 4). The
outcome suggests that under the 10% simulation scenario, 1,000 SAVs will be sufficient to serve these
35,993 CPV trips from 16,163 original CPV drivers with satisfying service levels. Then, the remain-
ing agents currently with external CPV trips are also shifted through the Park-and-Ride (PnR) initiative
(Chapter 5): CPVs are obligated to be parked at the designated PnR facilities at the city outskirt for
switching when entering Brussels so the remaining trips within Brussels are all carried out with SAVs or
PT based on trip-specific utilities. The outcome reveals the significant impacts of PnR market penetra-
tion and SAV pricing strategies on urban mobility. The proposed policies for eliminating current CPV
trips bring notable benefits, including a substantial CPV replacement ratio, increased PT utilisation, re-
duced congestion in the city centre and significant transport emission reductions. However, there are also
drawbacks with the increased PnR demand, such as lower SAV service levels, longer travel time for PnR
travellers and increased congestion in regions with high PnR demand. Consequently, our results suggest
that a PnR market penetration between 40% to 60% represents a feasible range under the current Brussels
mobility network. Furthermore, SAVs should be seen as a complement to PT rather than with a low fare
structure, as low pricing will lead to excessive reliance on SAVs, further reducing overall urban mobil-
ity efficiency. It is important to emphasise that SAV technology itself is not enough to reduce current
reliance on CPVs. Collaborative efforts and coordination among various stakeholders are necessary to
truly achieve the full potential of SAVs towards more sustainable, accessible and efficient future mobility
systems.
Overall, our thesis presents a benchmark from social-spatial and cellphone data towards the digital
testbed for SAV policy evaluations for reducing CPV trips in metropolitan regions. Chapter 6 provides a
general conclusion of the conducted research, followed by several recommendations to different mobility
stakeholders for maximising the benefits of SAV technology based on our findings. Additionally, we
discuss the transferability of our algorithm and some research opportunities for future researchers.