Available via license: CC BY-NC-ND 4.0
Content may be subject to copyright.
Accepted Manuscript
Assessing the impacts of deploying a shared self-driving urban mobility system:
an agent-based model applied to the city of Lisbon, Portugal
Luis M. Martinez, José Manuel Viegas
PII: S2046-0430(16)30044-2
DOI: http://dx.doi.org/10.1016/j.ijtst.2017.05.005
Reference: IJTST 34
To appear in: International Journal of Transportation Science and
Technology
Received Date: 25 October 2016
Revised Date: 8 May 2017
Accepted Date: 8 May 2017
Please cite this article as: L.M. Martinez, J.M. Viegas, Assessing the impacts of deploying a shared self-driving
urban mobility system: an agent-based model applied to the city of Lisbon, Portugal, International Journal of
Transportation Science and Technology (2017), doi: http://dx.doi.org/10.1016/j.ijtst.2017.05.005
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
1
Assessing the impacts of deploying a shared self-driving
urban mobility system: an agent-based model applied to
the city of Lisbon, Portugal
Luis M. Martinez
International Transport Forum
2 Rue André Pascal, 75016 Paris.
Email: Luis.MARTINEZ@itf-oecd.org
José Manuel Viegas
International Transport Forum
2 Rue André Pascal, 75016 Paris.
Email: Jose.VIEGAS@itf-oecd.org
2
Abstract:
This paper examines the changes that might result from the large-scale uptake of a shared
and self-driving fleet of vehicles in a mid-sized European city. The work explores two
different self-driving vehicle concepts – a ridesharing system (Shared Taxi), which
emulates a taxi-like system where customers accept small detours from their original direct
path and share part of their ride with others and a dynamic bus-like service with minibuses
(Taxi-Bus), where customers pre-book their service at least 30 minutes in advance
(permanent bookings for regular trips should represent most requests) and walk short
distances to a designated stop. Under the premise that the “upgraded” system should as
much as possible deliver the same trips as today in terms of origin, destination and timing,
and that it should also replace all car and bus trips, it looks at impacts on car fleet size,
volume of travel and parking requirements. Mobility output and CO2 emissions are also
detailed in two different time scales (24 hr. average and peak-hour only). The obtained
results suggest that a full implementation scenario where the existing metro service is kept
and private car, bus and taxi mobility would be replaced by shared modes would
significantly reduce travelled vehicle.kilometres and CO2 emissions.
3
1. Introduction
Mobility is an important component of all human activity, ensuring the access of citizens to
exercise their social rights and the capacity to partake in productive activities. In an urban
environment, with higher density of population and economic activity, mobility plays an
even more relevant role, being a key component to warrant economic development and
social equity.
Yet, it is also seen as one of the major problems of urban areas due to the associated
externalities (congestion, greenhouse gases (GHG)), especially in highly motorised and car
dependent regions [1]. The rapid development of technology and increasing purchasing
power, which in many countries has surpassed the increase of transportation costs [2],
promoted more disposable income to buy cars [3]. Even in developing countries, where the
income levels are lower, cars are becoming the dominant motor vehicle, although mode
shares are still favourable to collective transportation [4].
The supremacy of the car allied with low occupancy rates on daily trips [5] is producing
severe impacts on the shape of societies, leading to harsh environmental and climate
changes; serious loss of efficiency of the transport system caused by congestion; promotion
of social asymmetry and exclusion from activities that can be only accessed by car; road
accidents and the strong dependence on fossil fuels [6,7].
Within this context, it is fundamental to understand the underlying factors that drive modal
choice and the reasons behind the dominance of the private car, creating solutions that will
ally individual satisfaction with societal welfare. According to [8], the private car presents a
clear advantage over the other transport options in three key attributes for mode choice
selection: flexibility, comfort and availability. These characteristics of the private car may
significantly blur the perception of other attributes of each mode, leading to a bias in favour
of the private car and to lexicographical choice processes [9].
Increasing environmental concerns and the high dependence on fossil fuels are pressuring
the transportation sector. In order to attain the ambitious environmental objectives in sight,
Governments will have to make an extra effort, which may speed up the convergence to
more efficient urban mobility solutions.
Even with the promotion of public transport (PT) networks, with the expansion of several
subway systems and the introduction of new Bus Rapid Transit (BRT) and Light Rail
Transit (LRT) systems, public transport continues to be losing market share to private
vehicles in most developed economies [10].
Three main approaches to mitigating urban mobility problems have been proposed:
influence demand to reduce travel needs (avoid), promote more sustainable transport
options (shift) and deploy better technology and reorganise supply (improve).
From a technological perspective (improve), the efforts in the last decades have been
concentrating on the development of cleaner energies and more efficient vehicle engines.
Other significant developments were obtained through the reduction of the environmental
impact of transportation infrastructure. These measures tend to be effective in the short
term; however, the overall impact on the system in the long run might be negligible if
transport demand continues to increase at the same pace [1].
4
From the avoid perspective, some policies have been recently designed to act on the supply
side, not only promoting tele-activities but also providing more efficient infrastructure and
land-use distribution. This concept emerged in the US during the last few decades under the
designation of Transit Oriented Development (TOD), where the objective is to develop
“Smart Growth” areas with mix land-uses and that are compact and walkable, usually
around rail stations. This new urban development paradigm aims to promote accessibility to
a wider variety of activities, encouraging walking and the use of more sustainable transport
options against the private car [11].
Shifting demand towards more sustainable transport options by providing the right
incentives and penalties has showed some efficacy in the last decades. Intervening on the
demand side by promoting more efficient and rational use of the existent transport system
has been pursued in several cities since the 90’s. This concept has been denominated as
Travel Demand Management (TDM) [12], gathering a set of measures that range from the
introduction or promotion of more efficient transport options or the creation of financial
incentives towards more efficient mobility.
There are many different measures that can be used in TDM and some of them have already
been successfully implemented: moral campaigns (e.g. eco-driving); the promotion of non-
motorised transport modes [13]; or financial incentives as parking policies aiming at
reducing cars in city centres [14-18], congestion charges (e.g. Singapore [19], London [20]
and Stockholm [21]); or the promotion of Intelligent Transport Systems (ITS) that may
enhance the efficiency of the transport system; among other policies.
The promotion and integration of shared transport options, within a so-called shared
economy paradigm has emerged. This new mobility paradigm has been found as a very
interesting option to divert car users to public transport options [22]. This concept has been
in discussion for several decades but only in the last decade, technology has evolved
sufficiently and its key instrument - the smartphone - spread enough across the population
to allow for the shared mobility market to gain some scale and become more viable.
This new approach to demand management aims at exploring mobility resources more
efficiently, while preserving good levels of comfort and flexibility normally associated with
the private vehicle [23]. The proposed shared modes tend to explore the low levels of
private vehicle usage both in (internal) space and time, vehicles being active mainly during
peak hours and rarely for more than 10% of the day, as well as the low level of occupancy
in each trip. Despite this, they are highly valued assets – so highly valued that households
put up with such levels of cost and low usage in order to derive specific benefits relating to
comfortable, door-to-door and schedule-less travel. This low efficiency at the personal level
is replicated at the social level with the congestion and emissions caused by the quite low
occupancy levels of those private cars. Could this inefficiency be reduced while retaining
these benefits?
The traditional shared mobility market tries to explore these two dimensions (sequential or
parallel share of vehicles) and segmenting supply to improve demand satisfaction. Two
main transport options have been widely explored: carpooling (space sharing among a
group of friends) and carsharing (time sharing). Additionally, we can consider two other
shared transport alternatives that further explore this spectrum of shared mobility
efficiency: ridesharing or Shared Taxis, which represent an expansion to the existing taxi
system where different passengers or parties share the same vehicle for parts of their rides,
and on demand minibus services, that expand or replace the regular bus concept beyond
5
fixed routes and fixed schedules to improve public transport provision efficiency and
efficacy. Both alternatives explore time and space sharing solutions.
With the arrival of ubiquitous internet access and dedicated app-based services, carsharing
has quickly grown in popularity and sophistication and numerous successful services have
been deployed around the world [24]. At the same time, there has been an analogous
development in terms of technological sophistication with ridesharing services – especially
for app-based on-demand services. These can take the form of taxi-like services or peer-to-
peer real-time ridesharing. As with app-based carsharing, these forms of ridesharing have
proven to be tremendously popular and pioneering companies in this field have been very
successful.
Several studies have explored through simulations the role of shared mobility services
either as additional service in the mobility market [25], or by fully deployed systems that
would replace all motorised mobility in a city [26,27]. In the later paper, the authors present
a model of a fleet of Shared Autonomous Vehicles (SAV) in a city of a size similar to that
of Austin, Texas. The model has the following characteristics: each SAV travels
autonomously (no human intervention) with at least one passenger to its final destination.
In this model, there are no stops between origin and destination to board other passengers
and no deviation occurs from the initial trip.
After each trip, the SAV moves on to the next traveller or relocates itself to a more
favourable location for lower cost parking and faster future passenger service. This implies
that there are no fixed stands that travellers must reach to start their trips since the SAV
comes to them. The fleet is comprised of traditional petrol-fuelled SAV sedans (no hybrid,
electric or alternative fuel vehicles were modelled). Finally, the authors consider a small
share of all trips, only 3.5%, of all trips are made using the SAV network, the rest being
made with conventional human-driven vehicles.
The modelling results suggest that each SAV would serve 31 to 41 persons per day, with an
average waiting time below 20 seconds. Each SAV would replace nearly 12 conventional
vehicles, and would lead to the elimination to 11 parking spaces per SAV in operation.
Overall distance travelled increased by 11% compared to a traditional human-driven self-
owned fleet. This increase in travel distance was largely due to the relocation of the SAVs
and the distance travelled to collect the next passenger. However, environmental impacts of
the implementation of such a fleet are positive, with 5.6% less greenhouse gas emissions,
34% less carbon monoxide emitted, as well as a 49% reduction in volatile organic
compound emissions, among others, compared to the traditional US light duty fleet.
Emission reductions could be further reduced considering the more intensive use of the
SAV which leads to a shorter life cycle for each vehicle (-1.5 to 2 years), hence an earlier
replacement by more recent and less polluting vehicles. The use of an electric fleet could
even further reduce emissions as well.
Among the identified limitations to this modelling exercise is the lack of a real-world
context, as the capturing of heterogeneous land use and travel patterns, seasonality and
weekends. Other changes could include incorporating car-pooling options to improve the
use of the SAVs as well as reduce overall distances travelled and associated environmental
impacts.
Other studies have explored the potential of autonomous shared fleets. Spieser, Treleaven et
al. [28] explore the effect of a complete removal of the entire private vehicle fleet in
Singapore, and its replacement by a self-driving and shared fleet. The findings suggest that
such a fleet could remove two thirds of the vehicles currently operating in Singapore while
6
still delivering all of the trips currently made by private vehicles. The authors note several
benefits of autonomous driving, i.e. better safety performance, an increase in the
convenience and optimisation of trips, a decrease in the congestion, lower overall costs,
lower parking space requirements, etc. The case study presented is that of self-driving and
shared vehicles, the authors note the findings could be extended to more general situations,
such as shared vehicles with human drivers. However, the paper concludes that the most
cost and time effective option, thus the most viable one among those presented would be
that of an Automated Mobility-On-Demand (AMoD) system: the Shared Self-Driving
Model appears almost 50% cheaper than the Human-Driven Cars model. However, such
system increases the overall distance travelled, as well as vehicle use intensity, which may
erode benefits linked to travel times and congestion.
Other interesting study on the same topic has been conducted by [27]. This study models
the implementation of a fleet of autonomous taxis (ATaxis) in New Jersey, based on origin-
destination travel survey derived trips. These trips approximate the real trips made by
people in New Jersey every day. Passengers would go to a station, take an ATaxi and it
would then bring them to the nearest station to their final destination. Other passengers can
join the ride, provided that their destinations are not too far from the destination of the first
passenger.
Results suggest that there is significant rideshare potential. This potential is sensitive to
relaxing the travel scheduling constraint away from the original trip. Average vehicle
occupancy increases along with the increase of the waiting time at the station (to increase
the chance that another passenger joins). It increases as well when destinations of
passengers are close to each other. The simulation shows that demand varies temporarily
and spatially -- the potential for ridesharing increases during peak hours for example, and in
given locations such as railway stations. Taking that into account could make it possible for
such system to contribute to significant reductions of congestion in trafficked areas,
alongside a corresponding reduction in pollution.
More exercises that attempt to measure the impact of SAV in transport mobility and
emissions can be found in [29]. The results also suggest some significant reduction of GHG
emissions with a full implementation of SAV fleets.
In this work, we examine the potential outcomes of a radical change in urban mobility
configuration that would result from the implementation of a shared and fully autonomous
vehicle fleet. This study was developed at the International Transport Forum (ITF), using a
simulation framework previously developed at the University of Lisbon to explore
carsharing business models for the city of Lisbon [30].
Our work investigates the convergence of shared transport services, including ridesharing
and self-driving vehicle technology (SAV). The former has traditionally concerned largely
informal and ad-hoc sharing (household car-sharing, car-pooling, etc.) but, starting in the
1980s, new models of co-operative-based and commercial carsharing emerged. These
forms of carsharing allowed individuals to subscribe to shared fleets whose vehicles they
reserve, access and use only when they need them. Pricing for these services is typically
calculated on a per-hour or per kilometre basis (or both). These services are situated
somewhere between traditional car rental services and taxis and have proven popular in
many urban areas since they allow individuals to have access to cars without necessarily
owning one. With the arrival of ubiquitous internet access and dedicated app-based
services, carsharing has quickly grown in popularity and sophistication and numerous
successful services have been deployed around the world. At the same time, there has been
7
an analogous development in terms of technological sophistication with ridesharing
services – especially for app-based on-demand services. These can take the form of taxi-
like services or peer-to-peer real-time ride sharing. As with app-based car-sharing, these
forms of ridesharing have proven to be very popular and pioneering companies in this field
have generated billions of dollars in market capitalisation.
These services currently require a driver and so it seems interesting to examine what might
be the next step in these services’ evolution, namely, their integration with self-driving
technology. This is not necessarily just a theoretical exercise – both Google and Uber have
signalled both explicitly and implicitly that they see great potential for shared and
autonomous vehicle fleets in both the carsharing and ridesharing modes. Several
researchers have also examined the comprehensive impacts of the deployment of shared
and self-driving vehicle fleets in various contexts.
To perform this assessment, we developed a new agent-based model to simulate the
behaviour of all players of this system: the travellers, potential users of the shared mobility
system; the cars, which are dynamically routed on the road network to pick-up and drop-off
clients or move to/from/between stations; and a mobility dispatcher that efficiently assigns
cars to clients, respecting the defined service quality standards (waiting time, detour time).
We chose to develop such model under the principles of agent-based techniques given the
difficulty in representing the complex spatial-temporal relation between vehicle supply and
passenger demand by other types of aggregated and disaggregated simulation methods. The
usefulness of agent-based models has been well demonstrated in several areas of
transportation analysis [31-35]. These models allow a detailed representation of the
interactions of multiple agents in a realistic synthetic environment where the intent is to re-
create and predict the appearance of a complex phenomenon, which is the case of the
mobility market. The objective is to obtain high level indicators from describing accurately
the lower level reality of shared mobility supply interacting with the current mobility
demand. There are already available several platforms or software that have developed
agent-based models to transport modelling, even with shared mobility components (i.e.
SimMobility or MATSim) [36,37]. Nonetheless, a full design of the platform and the
design of the algorithms and shared mobility solutions was required to explore different
transport market supply organisation structures.
We base this analysis on a real urban context: the city of Lisbon, Portugal. We selected
Lisbon as a case study due to the availability of data required to develop an agent-based
simulation and because of its relative comparability with other European contexts.
This document is organised as follows: after this introduction, we describe the methodology
of the simulation model and the mobility scenarios that were tested. Next, we present a set
of key results obtained, and we finish with a discussion of the policy implications that
derive from those results.
8
2. Case Study: the city of Lisbon
Having a detailed view of our test city’s existing mobility patterns along with its specific
topology and other characteristics is necessary in order to provide a frame of reference for
our modelling exercise. This of course constrains the direct transferability of the results of
the modelling exercise as they are firmly embedded in a unique local context. Nonetheless,
insights derived from this exercise can provide an indication of the scale and direction of
the expected impacts of the deployment of such a mobility system.
Our simulation exercise was based on data from the municipality of Lisbon, Portugal.
Lisbon is the capital city of Portugal and is the largest city in the country with
approximately 565,000 inhabitants in an area of 84.6 km². The city is the centre of the
Lisbon Metropolitan Area (LMA), which has approximately 2.8 million inhabitants,
representing roughly 25% of the Portuguese population, with an area of about 3000 km2
formed by other 18 municipalities.
The metropolitan area generates over 5 million person-trips each day, of which 55% are
commuting trips to go to work or study. Of this total activity in the LMA, about 1.2 million
trips take place within the administrative boundaries of the Lisbon municipality. It is this
set of trips that are examined in our simulation model.
The inhabitants of the Lisbon municipality display a relatively low car ownership rate (217
cars/1000 inhabitants) and a low number of daily trips per inhabitant (1.9). As parking
space is very scarce in the old heritage areas of the city, this has acted as a deterrent to car
ownership to residents. The low daily trip rate is related to the demographic profile of the
city centre population which is skewed towards an ageing population, in some traditional
city boroughs, more than 50% of the population is aged 65 or older.
Lisbon’s main characteristics with regards to transport infrastructure and public transport
services provision are outlined in Table 1.
Lisbon has a well-established underground network that plays in important role in the daily
mobility of inhabitants and workers of the Metropolitan Area, and especially within the
municipality of Lisbon. The underground system was inaugurated in 1959 and has been
growing since, covering currently a significant part of the Lisbon area, with some short
extensions to suburban areas.
The Lisbon underground is a medium-sized network comprised of four lines (covering
about 43 kilometres of linear distance) and 52 stations transporting a total of 176.7 million
passengers per year. There are also four main commuter rail lines connecting the greater
region with the city centre. The relevance of these lines for the present study is limited
since only 13 stations are located within the city centre.
Recent data indicates that over 60 000 cars, 400 buses and 2 000 taxis circulate
simultaneously during peak traffic periods resulting in an average density of 60 vehicles per
road kilometre [25]. This value is relatively high, especially when considering that a
significant part of the road network is comprised of narrow streets.
9
Table 1: Infrastructure and public transport provision in Lisbon
Infrastructure provision
Road
Road network density (km road / km2) 12.58
Motorway network density (km road / km2) 0.24
Public transport
Underground network density (km line / km2) 0.51
Underground stations density (stations / km2) 0.65
Priority lanes for public transport (km road / km2) 0.22
Parking
Street parking capacity 153 000
Off-street parking capacity 50 000
Parking availability (parking space/inhabitant) 0.38
Public transport services provision
Bus (thousands vehicles-km) 329.7
Trains (thousands vehicles-km) 27.0
Underground (thousands vehicles-km) 6 495.7
Source: Câmara Municipal de Lisboa, 2005.
Latest values estimated for the city indicate a maximum of 160000 cars parked
simultaneously, resulting in a very high utilisation rate of the available capacity (78%) [38].
This parking constraint has suppressed car ownership rates in the heart of the city resulting
in a more balanced mode share distribution in the city centre than in the whole of the
metropolitan area (Table 2). While the latest census data reveals that 60% of all trips within
the greater Lisbon municipality are undertaken in a private car, this percentage drops to
40% in the city centre with 20% of city centre trips taking place by non-motorised means
(mainly walking), although just representing 11% of the passenger.kilometres (pkm) .
Table 2: Mode share assessment (in Pass.km) (2011)
Modes City of Lisbon LMA
Private car 35.56% 59.26%
Motorcycle 2.18% 1.23%
Taxi 1.62% 0.42%
Bus 25.07% 13.76%
Walking/cycling 11.12% 3.11%
High-capacity public transport (Underground or rail) 19.80% 10.56%
Car + High-capacity public transport 1.11% 4.00%
Bus+ High-capacity public transport 3.54% 7.66%
10
3. Methodology
3.1. General framework
In this work we developed an agent-based model that simulates the daily mobility of
Lisbon, Portugal. The model is formed by three main agents that interact in a common
environment: users, vehicles, and dispatcher. The model is based on real trip-taking activity
replicating the patterns and schedules observed in the survey. Transport infrastructure and
supply set the environment, where its road network, i.e., a set of nodes and arcs,
characterizes the city, while public transport supply is represented by the available routes
(bus, subway, and rail). Our model addresses the interaction between clients and vehicles,
simulating their connection and how, in terms of timing and location, the services are
performed. The model presents a dynamic representation of traffic environment using an
update of traffic assignment conditions every 5 simulation minutes. Travel time is
attributed to each arc and intersection based flow capacity ratio and free-flow speed, not
considering any change on traffic density parametres resulting from self-driving vehicles.
When shared alternatives are set as available, the dispatcher system manages the centralised
task of assigning mobility requests to vehicles using the location of shared vehicles, their
current occupancy level and the location of clients as its main inputs. The model estimates
trip routing based on an algorithm that generates the lowest time path between any pair of
nodes of the network.
The city is divided in a homogeneous grid of 200mx200m cells, being used as spatial
resolution of the model for the selection of allowed passenger groups and allocation of taxis
to clients as well as the assignment to current public transport routes. All trip origins and
destinations are linked to the closest node of the road network.
The ridesharing system, either for Shared Taxis or Taxi-Buses, presents a pre-optimised
location set of stations where vehicles are parked when idle. For that purpose, a minimum
set cover algorithm was programmed, considering as potential location of stations all the
existing off-street parking facilities of Lisbon and constraining the travel time of all grid
cells to the close4st station of 5 minutes using the morning peak hour link and intersection
travel times.
The next sections describe the shared mobility option designed and simulated for the city of
Lisbon, how demand is generated in the model, the behaviour of the user agent, the
definition of operation of shared vehicles and how the centralised dispatcher manages the
interaction between demand and supply. The simulation model workflow for mobility is
presented in Figure 1.
11
Figure 1: Simulation model workflow
3.2. Setting new shared mobility alternatives in the market
This section presents the conceptual design of the envisaged complementary shared
services that are then tested on the simulation platform. Two market segments were
addressed in this model: a self-driving ridesharing system (Shared Taxi), which emulates a
taxi-like system where customers accept small detours from their original direct path and
share part of their ride with others, and a dynamic bus-like service with self-driving
minibuses (Taxi-Bus), where customers pre-book their service at least 30 minutes in
advance (permanent bookings for regular trips should represent most requests) and walk
short distances to a designated stop. In all cases travellers get a transfer-free trip from origin
to destination. The following table presents a brief characterization of these two modes.
According to the definition of shred mobility services given by [39], the Shared Taxi
services is provided by “Ridesourcing” services and by having multiple passenger it
designated as “Ridesliptting”. Yet, the envisaged model has a cost as function of linear
distance from origin to destination and not producing a split of fare but just a cost
estimation to provide a service for each origin-destination pair. Taxi-Bus in [39] is
considered to be an “Alternative Transit Service”, which does not fit also into the concept
explored as both services designed interact closely to provide services to customers.
Table 1: Shared mobility services specification adopted
Mode Booking Access time
Max. Waiting
time Max. Total Time loss
Vehicle type
Shared
Taxi
Real time
Door
-
to
-
door
5 minutes (<= 3
km), up to 10
minutes (>= 12
km)
(detour time + waiting
time) from 7 minutes
(<= 3 km), up to 15
minutes (>=12 km)
minivan currently
seating 8 rearranged to
seat only 6, providing
easy entry and exit
Taxi
-
Bus
30 minutes
in advance
Boarding and alighting
up to 300 m away
from door, at points
designated in real time
Tolerance of 10
minutes from
preferred
boarding time
Set by the minimum
linear speed from
origin to destination
(15 km/h)
Minibuses with 8 and
16 seats. No standing
places
12
The specifications for these services were designed for a high level of acceptance of modal
transfer by current car drivers, by providing them with the three quality attributes of
flexibility, comfort and availability mentioned in [8].
The mobility alternatives were designed to fully replace current motorised road transport
alternatives (car, motorcycle, taxi and bus), implying a modal shift of users to the new
transport alternatives or to the previous options that remain available (walking/biking,
subway and rail).
Model diversion assessment is presented in the next sections.
3.3. Demand and mode choice
Based on an extensive mobility survey conducted in Lisbon Metropolitan Area (LMA) [1],
we created a synthetic population of trips within the city, aggregated by the aforementioned
grids. The used synthetic travel simulation model was developed and calibrated for the
LMA in previous studies [40]. The model output contains all the trip extremes not only
discretized in space (at the census block level) but also in time (presenting different trip
departure and arrival times) for a synthetic week day in the reference year of 2010 [40].
The model is stochastic because the number of trips that actually occur between two grid-
cells at each hour was generated using a Poisson distribution with λ equal to the average
number of hourly trips.
Each trip is characterized not only by its time of occurrence, origin and destination, but also
by trip purpose, traveller’s age and by whether the traveller has a public transport pass.
Additionally, based on Census data and other mobility surveys [41-43], each trip is further
characterized by the traveller’s gender, income and by whether the traveller has a driving
license or not, car, motorcycle, parking spaces at home and at work. The purpose of the trip
determines the activity duration, which can be used to calculate parking costs.
The destination grids are characterized by parking cost and parking pressure (ratio of
demand to supply), both depending on time of day, linking the available demand data with
the statistics and pricing of Lisbon’s parking.
For all modes currently available – car, motorcycle, taxi, walking, bus or tram, subway or
suburban train, combination between light and heavy transport modes – the trip is
characterized by access time, waiting time, travel time, cost and number of transfers (if
applicable) [44].
To determine the modal choice of the users in the reference configuration (current
mobility), the Agent-Based model (ABM) incorporates the discrete choice model (DCM)
described in [44]. The model specification and calibration results are presented in Table 2,
and nested-logit aggregates transport alternatives into:
• "Motorised Private transport” nest: private car (PC), motorcycle (MT) and taxi
(TX);
• "Public transport plus walking” nest: bus (BS), walk (WK), heavy public transport –
subway and rail (HV) and combination of bus + heavy public transport (CB).
13
Table 2: Coefficients of the obtained discrete choice model
Attributes
Transport Alternatives
Motorised Private transport (MP)
Public transport plus walking (PT)
PC
MT
TX
BS
WK
HV
CB
ASC
NA
NA
-
4.230***
-
0.218*
-
1.140***
-
0.270*
-
0.706***
Socio
-
demographic attributes
Age [25
-
35]
NA
NA
NA
NA
NA
0.559***
NA
Age [35
-
65]
NA
NA
NA
NA
-
0.308*
NA
-
Age [+65]
NA
NA
NA
0.195*
NA
NA
-
Income (thousand
€)
NA
NA
NA
0.007***
NA
NA
NA
Land use, car
and public transport monthly pass availability
No parking at home
NA
NA
0.237***
NA
NA
NA
NA
No parking at
destination NA NA 0.237***
NA NA NA NA
Own car
NA
NA
NA
NA
NA
NA
NA
Public transport pass
NA
NA
NA
0.562***
-
0.766***
0.562***
NA
Parking
Pressure [0
-
1]
-
0.131***
NA
NA
NA
NA
NA
NA
Transport operation attributes
Fuel cost (1/
€)
-
0.326***
-
0.326***
NA
NA
NA
NA
NA
NA
Toll (1/
€)
-
0.221***
-
0.221***
-
NA
NA
NA
NA
NA
Parking cost (1/
€)
-
0.221***
NA
NA
NA
NA
NA
NA
NA
Travel time (1/min)
-
0.026***
-
0.026***
-
0.026***
-
0.024***
-
0.005***
-
0.015***
-
0.015***
Access time (1/min)
NA
NA
NA
-
0.051***
NA
-
0.053***
-
0.051***
Tariff (1/
€)
NA
NA
-
0.115***
-
0.498***
NA
-
0.498***
-
0.498***
Transfers
NA
NA
NA
-
0.199***
NA
-
0.150**
-
0.150**
Waiting time (1/
min) NA NA -0.028***
-0.028***
NA -0.045***
-0.028***
Nested scale (η)
1.000
1.951
0
-
test significance
NA
***
1
-
test significance
NA
***
NA not available, ***significant at the 99% level; **significant at the 95% level; *significant at the 90% level.
†Parking pressure defined as the ratio between estimated demand and supply of parking in a specific area and
time period of the day.
The model proved to have an adequate specification: ρ2=0.37 and the utility function of
each transport alternative included socio-demographic variables of the user, land use, car
and transit availability and instrumental attributes such as travel time and cost, all being
statistically significant at a 90% confidence level.
Is important to be aware of the quality of the mode choice estimation will depend greatly
also on the ability to forecast the choice set available for each customer. The most explicit
situation is motorcycle, which would suggest a very favourable utility function, yet its
mode choice is limited as the motorbike ownership is very low in the city of Lisbon.
Having the full characterization of the trip as input, the DCM calculates the probability of
choosing each mode. A mode is assigned to the traveller by Monte Carlo simulation where
modes with higher probability will be chosen more often.
Currently, one user is equivalent to one trip, i.e. users do not cluster in parties neither have
memory. Therefore, previous experience does not have an impact on future choices. It also
means that each decision is made individually and is not activity-based, nor the daily
routines of the user’s family are considered in the choice.
14
The absence of stated preferences devoted to the shared mobility solutions to be tested and
the lack of experience of users to these services forced the use of alternative approach to
estimate mode choice under an alternative radically new scenario. For this purpose, an
expert analysis procedure was used in the alternative shared mobility scenarios’, generating
a rule-based lexicographic choice process based on socio-demographic and mobility
attributes of the user. The envisaged selection process tries to incorporate the individual
rationale in mode choice while preserving some stochasticity. The mode choice process has
the following sequential rules:
• Walking:
1. trips with distance <= 300 metres
2. 300 metres<trips with distance <= 1500 metres
Stochastic choice following a walking acceptance distribution
calibrated for Lisbon in [45]
otherwise
• Subway or rail:
1. Number of transfers <=2 and stochastic choice following the
distribution of acceptance of total access time (origin+destination) to
stations, as calibrated for Lisbon in [45]
otherwise
• Shared-taxi:
1. People that own a car and are the most frequent driver / user and not
owning a public transport pass
Stochastic random number generator (<=0.7)
2. People that own a car and are not the most frequent driver / user and
not owning a public transport pass
Stochastic random number generator (<=0.5)
otherwise
• Taxi-Bus:
1. Remaining users that do not select the previous options
This procedure is implemented to each generated user, allowing estimation of modal
diversion in each shared mobility configuration. The Taxi-Bus selection is constrained to
the availability of services that aggregate at least 50% occupancy at the same time during
some section of the route. The users that select Taxi-Bus and do not have services available
are then diverted (upgraded) to the shared-taxi option at no extra cost for the user. The
following table presents the aggregate statistics of modal choice of users in the system for a
synthetic mobility set with 1,138,696 daily trips inside the core of the city of Lisbon (some
96 km2). The results show that individual motorised transport users (car, motorcycle and
taxi) switch mainly to the shared-taxi and Taxi-Bus alternative, while walking and subway
users mainly remain in the reference mode, except in cases of long walking distances or bad
subway connectivity were users divert to motorised shared alternatives. Users that
previously combined rail and bus modes mainly switch to shared alternatives, either due to
poor heavy transport connectivity or due to long access time to subway and rail.
The designed modes do not present an explicit integration between shared and heavy public
transport alternatives within the city. The concept of using shared services (mainly Taxi-
15
Bus) when moving upscale to the metropolitan area should be explored as a relevant option
either to Subway or rail + bus or to Subway or rail + car users.
Table 3: Analysis of mode diversion
Reference scenario
mode
Reference
modal choice
Mode in new scenarios
Walk
Subway or
rail Shared-taxi Taxi-Bus
Private car
45.50%
11.04%
6.21%
46.67%
36.08%
Motorcycle
1.25%
10.79%
9.16%
39.40%
40.65%
Taxi
2.25%
4.70%
11.57%
39.83%
43.91%
Bus
14.01%
2.27%
19.15%
13.20%
65.38%
Walk
21.21%
79.71%
1.81%
16.47%
2.01%
Subway or rail
13.89%
1.19%
93.14%
0.07%
5.60%
Subway or rail + bus
1.89%
0.14%
8.35%
36.39%
55.12%
Overall modal share in new
scenario
22.66%
19.36%
28.66%
29.32%
When the model chooses a shared option, a new user (agent) is generated in the simulation
environment, with a departure node, an arrival node and a starting time. Currently, one user
is equivalent to one trip, i.e. users do not cluster in parties at the outset of their trip (though
they do share vehicles whilst underway in the shared-taxi and Taxi-Bus simulations).
3.4. Users
In the simulation environment, a trip is generated when a user requests a departure from a
point towards another point. The model accounts for the simulation parameters (resulting
from the specification of each shared mode) and accounts for waiting time, detour time and
arrival time tolerances that are defined for the model run. The dispatcher then finds, in real
time or with the pre-booking, the best possible routing and assigns one of several available
vehicle types to carry out the trip in either a shared-taxi or Taxi-Bus mode.
The user then waits for the vehicle or walks to a specified pick-up location and boards the
vehicle. When the vehicle arrives at its destination, the user exits the system and a set of
indicators are generated in a trip log so that they can be used for ex-post system evaluation.
3.5. Cars and minibuses
The cars agent is formulated as a reactive agent that follows the instructions of the
dispatcher. Idle cars are in stations spread across the city (60 stations for the city of
Lisbon), and whenever the car is empty and not dispatched to a new trip, it relocates itself
to the nearest station. Active cars follow the shortest path and minimise travel time for its
route assignment taking into dynamically updated link travel times and intersection delays
(every 5 minutes).
Taxi-Buses are vehicles that are by default available at stations and relocate between
services if required or stop to idle to the closest station to the last service. The system
generated 320 potential stops in the city of Lisbon than can be activated during the day. The
location of these stops was constrained by minimum distance between stops (300 metres)
and the selection of the road node with greater connectivity in the neighbouring area in
16
order to ensure flexible routing for the vehicles (avoid streets with traffic only in one
direction or left turning blocking).
The fleet of cars and minibuses is an output of the simulation by measuring the number of
vehicles that are required in the simulation and their relocation dynamics between stations
during the day. The minibuses required are differentiated between 8 or 16 seated passenger
vans or minibuses.
3.6. The Dispatcher
The Dispatcher is an entity that defines a set of rules for matching cars to users, centralising
all real-time information required to produce and monitor these trips. The choice of which
car or minibus to match with a user’s request takes into account a time-minimisation
principle that applies not just to the requesting user but also to those already underway in
the same vehicle.
Several parametric constraints has been defined that must be satisfied for each trip route
solution proposed by the dispatching system as described in the service specification
section.
The model defines in parallel the dispatching of Taxi-Bus and shared-taxis when both
systems are operating. Users launch their requests and preferences that are recorded in the
system. In case of a Taxi-Bus request, they are processed 30 minutes in advance. The
dispatcher runs a local search algorithm that tries to maximise the number of passengers
assigned complying with the users’ constraints at each step (best match in minibus service
that warrant at least 50% occupancy at least in some part of the trip and an average (per
kilometre) occupancy rate greater than 25 percent of the vehicle capacity). Some users that
are not assigned to Taxi-Buses because of these constraints are then re-assigned (upgraded)
to the shared-taxi system as real time requests, following the shared-taxi real time booking
system automatically performing door-to-door services.
The shared-taxi dispatching services operate a real time optimisation model that tries to
minimise the additional vehicle kilometres generated by each additional user in the system,
estimating the minimum insertion Hamiltonian path for each operation vehicle that satisfies
all the constraints of passengers already assigned or on-board and the current request. This
is performed under a variable radius associated with the maximum waiting time acceptable
by the request client. Every request is dispatched using the time flexibility constraints
estimated for the trip. The client receives right away a notification of assignment with the
maximum waiting time from the initial model solution. Yet, the request may be pending on
the Shared Taxi to send for a few minutes. The dispatcher may try a more efficient
assignment to a vehicle approaching his location that produces less additional vehicle-
kilometres (and possibly additional vehicles) in the system. The shared-taxi plate is
communicated to the user at least 3 minutes ahead of the estimated boarding time. When 50
percent of the maximum waiting time of that user is reached and no convenient shared-taxi
with clients has been found, the dispatcher searches for available empty taxis nearby, and if
none is found, it generates an additional vehicle departing from the closest station to serve
the user.
The dispatcher also controls the vehicle movements when idle, ensuring efficient vehicle
movements to stations that may require additional fleet soon.
17
4. Testing new shared-mobility scenario configurations
4.1. Shared mobility scenarios tested
The model was tested for three main scenarios. In all tested scenarios, current car, taxi and
bus trips were replaced for shared mobility alternatives, preserving the walking and heavy
public transport available. The tested scenarios were:
• the current scenario, taken as reference to assess the impacts of the introduction of
shared mobility services (Baseline);
• one scenario that uses the lexicographic choice for mode diversion with only walking,
subway or rail and shared-taxis available (Two modes - 2M);
• and a scenario that uses the lexicographic choice and allows users to additionally
select Taxi-Bus as an alternative (Three modes - 3M).
In the next section we present the results for the Baseline scenario and the behaviour of the
alternative scenarios relative to this reference. These include a set of indicators related to
mode shares, vehicle-kilometres performed by all motorised road modes (vkm), CO2
emissions generated by road modes, total travel times, (including access, waiting and on-
board time for public transport and shared mobility options, and on-board and parking
search time for private car), plus occupancy levels of shared alternatives, and fleet sizes
required to provide the envisaged services.
The implemented fleets while been autonomous will be accounted from a GHG perspective
as the current average fleet composition of Lisbon for private vehicles and van and minibus
available in the market (diesel). This assumption tries to isolate the impacts of self-driving
and ridesharing component from the change of technology. Yet, it is expected that new self-
driving vehicles in the future may evolve fast to electric fleets with low CO2 emissions in
the energy production source and much less local pollutant impacts.
4.2. Model results
The model was computed over the same synthetic population of trips for all scenarios. We
will start by analysing the obtained indicators for the reference scenario and compare
directly the estimates for each alternative scenario.
The analysis of the Baseline modes shares throughout the day are presented in Figure 2.
The results show the current dominance of the private car, especially in off-peak hours. The
bus and the subway present a consistent mode share around 20% that is replaced by the taxi
during their non-operating hours (1 to 5 am). Alternatively, Figure 3 shows modal shares
for the 3M scenario. The results illustrate that Shared Taxi is the main mode throughout the
day, with Taxi-Buses increasing their efficiency during peak period and small Taxi-Buses
replacing the role of subway and rail during dawn.
The current mobility leads to a significant number of vehicle kilometres, and CO2
emissions registered within the city of Lisbon, where the car presents an average occupancy
rate of 1.2. Moreover, the occupancy levels of the public transport system are low, which
leads to low frequencies and consequently long travel times, especially when compared
with the private transportation options (i.e. car, motorcycle (M2W) and taxi).
18
Figure 2: Modal shares (pkm/h) across the day in the Baseline scenario.
Figure 3: Modal shares (pkm/h) across the day in the 3M scenario.
The introduction of the shared alternatives as discussed above in two different alternative
scenarios may lead to different transport modes shares. The shared alternatives present a
transport mode share around 70% along the day, complemented by subway and walking.
These changes generate significant impacts on the city mobility and its externalities. This
analysis was produced in equivalent motorised vehicles (pcu), as regular buses that
currently operate and replaced in some of the tested scenarios by smaller and less CO2
intensive vehicles (minivans or minibuses). The equivalence factor to regular cars was set
as 3 for regular buses, 1.3 for the minivans (8 passenger vehicles) and 1.7 for minibuses (16
passenger vehicles).
The summary of the main mobility outputs across the day are presented in Table 4. The
results demonstrate the significant load (vkm and emissions) reductions in the alternative
scenarios compared to the Baseline, especially during off peak periods, where modal share
is currently more favourable to private motorised modes.
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
70.0%
80.0%
90.0%
0 5 10 15 20
% of total Pax.km in the hour
Hours of the day
Baseline scenario, 2010
Car M2W Taxi Bus Subway Walk Rail
0.0%
10.0%
20.0%
30.0%
40.0%
50.0%
60.0%
0 5 10 15 20
% of total pax.km in the hour
Hours of the day
Shared Taxi Taxi-bus 8 Taxi-bus 16 Subway Walk+Acces s Rail
19
Table 4: Vehicle-kilometres and CO2 emissions across the day for the tested scenarios (Baseline scenario used
as base 100).
Total VKM in pcu units
Total CO2 Emissions (ton/h), all road modes
Hour
Baseline
2M
3M
Baseline
2M
3M
0
52 118
71.25%
70.86%
12.43
63.76%
61.04%
1
52 310
62.88%
60.38%
12.39
56.62%
52.45%
2
46
093
66.41%
64.83%
10.86
60.13%
56.13%
3
42 489
67.13%
66.61%
10.02
60.77%
57.37%
4
36 734
64.66%
63.19%
8.66
58.54%
54.85%
5
37 262
64.92%
63.24%
8.78
58.77%
54.73%
6
58 456
63.13%
59.52%
13.81
57.03%
51.98%
7
205 674
69.55%
60.31%
48.60
62.81%
52.59%
8
286 009
84.04%
74.38%
67.60
75.87%
64.60%
9
185 688
88.20%
84.45%
43.91
79.59%
74.48%
10
112 312
86.30%
80.00%
26.57
77.85%
70.72%
11
67 890
80.63%
75.79%
16.09
72.61%
66.57%
12
93 499
73.67%
68.85%
22.12
66.45%
60.00%
13
117 981
77.11%
75.40%
27.99
69.36%
65.35%
14
120 606
72.03%
69.37%
28.71
64.56%
60.49%
15
112 159
74.67%
71.26%
26.60
67.18%
62.32%
16
139 848
69.77%
66.35%
33.13
62.84%
57.86%
17
181 446
71.04%
68.01%
42.92
64.08%
59.19%
18
187 272
77.96%
75.24%
44.22
70.45%
66.01%
19
134 466
84.34%
80.72%
31.78
76.16%
71.24%
20
94 170
82.77%
78.12%
22.27
74.70%
68.81%
21
85 997
70.10%
67.56%
20.45
62.92%
58.87%
22
82 990
59.47%
57.30%
19.88
52.97%
49.32%
23
84 350
59.31%
57.04%
20.23
52.76%
48.79%
All
day
2 617 817
1 966 822
1 853 053
620.0
420
383
2010%
100%
75.13%
70.79%
100%
67.69%
61.78%
The main operational indicators of the shared alternatives show high occupancy levels (see
Table 5). Comparing to the current daily averages for motorised road modes in Lisbon, as
the private car with 1.2 and the bus with 13 passengers in 80 passenger vehicles, we can
conclude that this improvement is the key to the reduction in vkms and CO2 emissions.
Furthermore, the required fleet to produce the same mobility shows an impressive
reduction, with the number of Shared Taxis required being only some 3% of the current
Lisbon fleet of private cars, and 568% of the number of vans or minibuses (3048 minibuses
and 1018 vans) than the current bus fleet, but just representing 79 percent of the current
seated capacity (see Table 5). Savings are even greater in terms of parking space
requirements (as the Shared Taxis are in motion a much larger share of the time), allowing
a total release of the on-street parking space and a significant reduction of the off-street
parking facilities that could be converted to other uses (e.g. logistic centres).
The reduction in car fleet was found significantly larger than the reduction of vkms due to
more intensive use of cars than currently. In average, each vehicle in the 2M scenario is
active 12.87 hours per day (compared to the current average of 50 minutes a day)) and does
about 158 km (around 10 times more than currently).
20
Table 5 Operational indicators of the tested alternative shared mobility scenarios.
Indicator Two modes Three modes
Average pax on board (Shared
Taxis)
2.3
(peak 3.0)
2.0
(peak 2.6)
Average pax on board (Taxi-
Bus) NA 4.2 (8 seats) / 11.4 (16 seats)
Peak: 5.0 (8 seats) / 14.6 (16 seats)
Fleet size as share of current
city vehicle fleet
(Shared Taxis + Taxi-Buses)
4.8% 2.8% (cars)
Bus*: 568% veh. / 79 % (seats)
* - but these will be minivans and minibuses with capacities 8 and 16, not standard urban buses, with
capacity 80
This is an important feature of the 3M scenario, as it allows using a much more flexible
instrument (the Shared Taxi) to better serve specific mobility requests of clients which do
not find enough “partners” with mobility wishes at the same time to justify the deployment
of a Taxi-Bus.
Figure4 indicating the incidence of the service upgrade offers interesting perspectives:
• In relative terms, this scheme has a very high incidence growth from 2am to about
4:30 am, reaching almost 70% of all passengers asking for a Taxi-bus service, and
rather low incidence between 8:30 am and 9:30, coming even a bit lower than 10%
of those passengers. Then during most of the day period it is quite stable, around
20%, then moving up and down between 20% and 40% until 2am.
• In absolute terms, we have a different perspective: the lowest numbers are in the
night hours when the percentage is highest, which is also associated with the fact
that the Heavy Public Transport modes are not operating. And the highest peaks are
at moments of low percentage, but associated with the general mobility peaks in the
morning and afternoon periods.
• The averages across the whole day are at 23.5% (average weighted by the numbers
in each period).
Figure 4 Passengers Upgraded from Taxi-bus to Shared Taxi - in % of Taxi-bus clients
(left) and in Absolute numbers (right)
21
The impacts on non-private transport modes accessibility were also assessed. For
measuring accessibility to jobs an S-shaped curve to represent attraction decay was used to
compute equivalent job opportunities reached, meaning the number of jobs that we should
expect to be considered as accessible by people considering their willingness to move at
least the distance to each job or group of jobs. This curve was liberated with data collected
for Lisbon specifically [45]. The results presented in Figure 4 suggest a very large
improvement in access to jobs for public transport users. This change would improve
access equity largely and provide similar levels of access to everyone like private car
nowadays.
Current (2010) 3M scenario
Figure 4: Accessibility comparison of public transport access to jobs between current and in the 3M scenario.
In terms of the costs to provide the equivalent mobility to nowadays, we found that Shared
Taxi solutions could cost approximately 55 percent less per kilometres than current taxi
services. 35 percent of these saving come from the average fare split inside the service
while additional 20% would result from removing the driver costs from taxi fare structure
(around 30 percent, although that just considered 20% due to the additional technical
requirements to remotely manage a self-driving system).
If we compare the car ownership costs, even if parking costs are disregarded, a car owner
would just have cheaper mobility than using a SAV when driving more than 63 kilometres
a day. This value was estimated under the assumption of an acquisition cost of 30 000 euros
for a private vehicle used for 5 years and with a residual value of 30% of the original cost.
This value indicates that for most of the daily commuters, self-driving ridesharing services
could justify financially replacing their vehicle by this service.
Regarding the Taxi-Bus services, their average cost per kilometre can be 45 percent lower
than current public transport fare (computed from a public transport pass used 50 times a
month and with the average trip length of public transport trips in the city – 3.1 kilometres).
These saving could suggest a radical change in terms of affordability to high levels of
access, which could affect potentially the location of residents and activities in the future.
22
5. Conclusions
This paper presents an agent-based model developed to assess the potential impact of
shared mobility services on daily urban mobility. This simulation was calibrated for the city
of Lisbon and an asymptotic scenario were all private mobility plus current bus services
would be replaced by two new mobility solutions: a self-driving Shared Taxi, which
presents similar features to a current taxi but that requires app-based technology for the
service request and clusters passengers in a same vehicle along a real-time computed route;
or a self-driving Taxi-Bus that revisits the conventional bus concept by setting dynamic
routing in smaller vehicles with no transfer for users and travel time close to the current
private car.
The simulation model encompasses several optimisation models to perform the central
dispatcher task of matching clients and vehicles dynamically along the day.
With full adoption of the described paradigm we observed that carbon emissions would be
reduced drastically (almost 40 percent) in the most positive scenario, without change of
vehicle technologies. Likewise, congestion would strongly decrease leading to greater
traffic fluidity in vehicle mileage (30 percent reduction).
Additionally, to the savings due to higher vehicle occupancy, under all tested scenarios,
vehicles are used much more intensely than before – rising from approximately 50 minutes
to 12 hours per day and daily travel will increase from approximately 30 kilometres to
nearly 250 kilometres. This will reduce operating life-cycles and with it allow for a quick
renewal of fleets and thus a younger and environmentally cleaner fleet in average.
Moreover, accessibility may improve greatly with the tested shared services and transport
costs may also reduce drastically with SAV technology.
All result indicate very positive potential impact of SAV services, especially if introducing
a market segmentation and the deployment of on-demand self-driving bus services (Taxi-
Bus). Nonetheless, we should acknowledge the limitations of these estimations. First, the
scenarios tested do not try to emulate a real deployment scenario or how current system
would transition to the tested asymptotic scenario, as well as considering mobility options
of users regarding trip production, trip timing and destination choice unaltered. The tested
model tries to measure maximum potential impacts of such mobility options in an urban
context. As discussed in previous ITF studies [44], it is important to know potential impacts
of some mobility and technological solutions to generate transport policies to align the
urban transport system with the current global transport sustainability targets. The
deployment of SAV systems, if no ensured simultaneously sharing of vehicles and larger
vehicle load factors may lead a large increase of mobility and congestion in urban areas
derived from the self-driving vehicles relocation activities. Previous studies estimated the
impact on mobility (vkms) of self-driving carsharing system (sequential sharing of the
vehicles) in additional 40 percent vkms [44]. This fact could result in a problem to urban
mobility if not ensured that the use of SAV will lead also to significantly higher load
factors than current private car (1.2-1.5 in develop countries). Other studies arrive to
significantly lower values. Fagnant and Kockelman [45] arrive to values of extra vkms
between 5% to 9% and Zhang et al. [46] obtain a value of 25.5%.
Transport policies can influence the type and size of the fleet, the mix between public
transport and shared vehicles and, ultimately, the amount of car travel, congestion and
emissions in the city. Small and medium-sized cities, which have not started yet the
development of public transport systems with dedicated Right of Way (RoW), may find
23
that a shared fleet of small size vehicles could completely obviate the need for traditional
public transport.
Shared vehicle fleets free up a significant amount of space in the city. However, prior
experience indicates that this space must be pro-actively managed in order to lock in
benefits. Management strategies could include reallocating this space to wider sidewalks,
bicycle paths or delivery bays, or in some cases even to new construction of public facilities
missing in the neighbourhood. For example, freed-up space in off-street parking could be
used for logistics distribution centres.
Deployment of shared fleets in an urban context will directly compete with the way in
which taxi and public transport services are currently organised. The higher quality of
service to the user and higher efficiency in the use of public space argue in favour of
this becoming the new paradigm of public transport. Public Governance of transport
services, as well as current operators of bus and taxis must adapt, and others will
enter the market.
6. Acknowledgements
This paper presents the modelling framework developed by the International Transport
Forum (ITF). It also builds on work previously developed in the University of Lisbon, and
more recently upgraded for the Corporate Partnership Board of the ITF.
7. References
[1] Banister, D. The sustainable mobility paradigm. Transport Policy, Vol. 15, No. 2,
2008, pp. 73-80.
[2] Glaeser, E. and Kohlhase, J. Cities, regions and the decline of transport costs. Papers
in Regional Science, Vol. 83, No. 1, 2003, pp. 197-228.
[3] Zegras, P. C. The Built Environment and Motor Vehicle Ownership and Use:
Evidence from Santiago de Chile. Urban Studies, Vol. 47, No. 8, 2010, pp. 1793-
1817.
[4] Kutzbach, M. J. Motorization in developing countries: Causes, consequences, and
effectiveness of policy options. Journal of Urban Economics, Vol. 65, No. 2, 2009,
pp. 154-166.
[5] International Energy Agency. Indicators of Energy Use and Efficiency -
Understanding the link between energy and human activity, 1997
[6] Camagni, R., Gibelli, M. C. and Rigamonti, P. Urban mobility and urban form: the
social and environmental costs of different patterns of urban expansion. Ecological
Economics, Vol. 40, No. 2, 2002, pp. 199-216.
[7] Cervero, R. Transport Infrastructure and Global Competitiveness: Balancing Mobility
and Livability. Annals of the American Academy of Political and Social Science, Vol.
626, No., 2009, pp. 210-225.
[8] Hiscock, R., Macintyre, S., Kearns, A. and Ellaway, A. Means of transport and
ontological security: Do cars provide psycho-social benefits to their users?
Transportation Research Part D: Transport and Environment, Vol. 7, No. 2, 2002,
pp. 119-135.
24
[9] Hess, S., Rose, J. M. and Polak, J. Non-trading, lexicographic and inconsistent
behaviour in stated choice data. Transportation Research Part D: Transport and
Environment, Vol. 15, No. 7, 2010, pp. 405-417.
[10] Buehler, R. and Pucher, J. Demand for Public Transport in Germany and the USA:
An Analysis of Rider Characteristics. Transport Reviews, Vol. 32, No. 5, 2012, pp.
541-567.
[11] Litman, T., Rail Transit In America: Comprehensive Evaluation of Benefits, Victoria
Transport Policy Institute, 2004.
[12] Rodriguez, J. and Murtha, T., Travel Demand Management - Strategy Paper, 2009.
[13] Pucher, J. and Buehler, R. Cycling for Everyone: Lessons from Europe.
Transportation Research Record, No. 2074, 2008, pp. 58-65.
[14] Higgins, T. J. Parking Taxes - Effectiveness, Legality and Implementation, Some
General-Considerations. Transportation, Vol. 19, No. 3, 1992, pp. 221-230.
[15] Marsden, G. The evidence base for parking policies - a review. Transport Policy,
Vol. 13, No. 6, 2006, pp. 447-457.
[16] Tsamboulas, D. A. Parking fare thresholds: a policy tool. Transport Policy, Vol. 8,
No. 2, 2001, pp. 115.
[17] Viegas, J. M., Organisation and Financing of Public Transport. In: Proceedings of the
European Conference of Ministers of Transport (ECMT) 2005 - Sustainable Urban
Travel, 2005.
[18] Willson, R. W. and Shoup, D. C. Parking Subsidies and Travel Choices - Assessing
the Evidence. Transportation, Vol. 17, No. 2, 1990, pp. 141-157.
[19] Land Transport Authority - Singapore Government. "Electronic Road Pricing."
Retrieved June 20, 2011, from
http://www.lta.gov.sg/motoring_matters/index_motoring_erp.htm.
[20] Glaister, S. and Graham, D. J. An evaluation of national road user charging in
England. Transportation Research Part A: Policy and Practice, Vol. 39, No. 7-9,
2005, pp. 632-650.
[21] Albalate, D. and Bel, G. What Local Policy Makers Should Know about Urban Road
Charging: Lessons from Worldwide Experience. Public Administration Review, Vol.
69, No. 5, 2009, pp. 962-974.
[22] Köhler, J., Whitmarsh, L., Nykvist, B., Schilperoord, M., Bergman, N. and Haxeltine,
A. A transitions model for sustainable mobility. Ecological Economics, Vol. 68, No.
12, 2009, pp. 2985-2995.
[23] Enoch, M. Sustainable transport, mobility management and travel plans. Farnham ;
Burlington, VT, Ashgate, 2011
[24] Jorge, D. and Correia, G. Carsharing systems demand estimation and defined
operations: a literature review. European Journal of Transport Infrastructure
Research, Vol. 13, No. 3, 2013, pp. 201-220.
[25] Martinez, L. M., Correia, G. H. A. and Viegas, J. M. An agent-based simulation
model to assess the impacts of introducing a shared-taxi system: an application to
Lisbon (Portugal). Journal of Advanced Transportation, Vol. 49, No. 3, 2015, pp.
475-495.
[26] Fagnant, D. J. and Kockelman, K. M. The travel and environmental implications of
shared autonomous vehicles, using agent-based model scenarios. Transportation
Research Part C-Emerging Technologies, Vol. 40, No., 2014, pp. 1-13.
25
[27] Zachariah, J., Gao, J., Kornhauser, A. and Mufti, T., Uncongested Mobility for All: A
Proposal for an Area Wide Autonomous Taxi System in New Jersey. In: Proceedings
of the Transportation Research Board 93rd Annual Meeting, 2014.
[28] Spieser, K., Treleaven, K., Zhang, R., Frazzoli, E., Morton, D. and Pavone, M.,
Toward a systematic approach to the design and evaluation of automated mobility-on-
demand systems: A case study in Singapore, in Road Vehicle Automation(ed.).
Springer, 2014, pp. 229-245.
[29] Greenblatt, J. B. and Saxena, S. Autonomous taxis could greatly reduce greenhouse-
gas emissions of US light-duty vehicles. Nature Climate Change, Vol. 5, No. 9, 2015,
pp. 860-+.
[30] Martinez, L. M., Correia, G. H. D., Moura, F. and Lopes, M. M. Insights into
carsharing demand dynamics: Outputs of an agent-based model application to Lisbon,
Portugal. International Journal of Sustainable Transportation, Vol. 11, No. 2, 2017,
pp. 148-159.
[31] Arentze, T. A., Ettema, D. and Timmermans, H. J. P. Incorporating time and income
constraints in dynamic agent-based models of activity generation and time use:
Approach and illustration. Transportation Research Part C: Emerging Technologies,
Vol. 18, No. 1, 2010, pp. 71-83.
[32] Davidsson, P., Henesey, L., Ramstedt, L., Törnquist, J. and Wernstedt, F. An analysis
of agent-based approaches to transport logistics. Transportation Research Part C:
Emerging Technologies, Vol. 13, No. 4, 2005, pp. 255-271.
[33] Roorda, M. J., Cavalcante, R., McCabe, S. and Kwan, H. A conceptual framework for
agent-based modelling of logistics services. Transportation Research Part E:
Logistics and Transportation Review, Vol. 46, No. 1, 2010, pp. 18-31.
[34] Tang, J., Alam, S., Lokan, C. and Abbass, H. A. A multi-objective approach for
Dynamic Airspace Sectorization using agent based and geometric models.
Transportation Research Part C: Emerging Technologies, Vol. 21, No. 1, 2012, pp.
89-121.
[35] Vliet, O. v., Vries, B. d., Faaij, A., Turkenburg, W. and Jager, W. Multi-agent
simulation of adoption of alternative fuels. Transportation Research Part D:
Transport and Environment, Vol. 15, No. 6, 2010, pp. 326-342.
[36] Ayed, H., Khadraoui, D. and Aggoune, R. Using MATSim to simulate carpooling
and car-sharing trips. 2015 World Congress on Information Technology and
Computer Applications (WCITCA), No., 2015, pp.
[37] Marczuk, K. A., Hong, H. S. S., Azevedo, C. M. L., Adnan, M., Pendleton, S. D.,
Frazzoli, E. and Lee, D. H. Autonomous Mobility on Demand in SimMobility: Case
Study of the Central Business District in Singapore. Proceedings of the 2015 7th Ieee
International Conference on Cybernetics and Intelligent Systems (Cis) and Robotics,
Automation and Mechatronics (Ram), No., 2015, pp. 167-172.
[38] Câmara Municipal de Lisboa. Lisboa: O desafio da mobilidade, Câmara Municipal de
Lisboa - Licenciamento Urbanístico e Planeamento Urbano, 2005.
[39] Shaheen, S. A., Cohen, A. P., Zohdy, I., Berman, W., United States. Federal Highway
Administration. and Booz Allen & Hamilton. Shared mobility : current practices and
guiding principles.
[40] Viegas, J. M. and Martinez, L. M., Generating the universe of urban trips from a
mobility survey sample with minimum recourse to behavioural assumptions. In:
Proceedings of the 12th World Conference on Transport Research, Lisbon, 2010.
26
[41] Santos, G. D., Martínez, L. M., Viegas, J. M. and Alves, D., Design and deployment
of an innovative mobility survey for the lisbon metropolitan area oriented to assess
the market potential and obtain the best configuration of new alternative transport
modes. In: Proceedings of the European Transport Conference, Glasgow, 2011..
[42] Moura, F., Duarte, A. and Viegas, J., CAReFUL – Car fleet renewal as a key role for
atmospheric emission reduction, CESUR, Instituto Superior Técnico, Lisbon, 2007.
[43] Martínez, L., Viegas, J. and Silva, E. A traffic analysis zone definition: a new
methodology and algorithm. Transportation, Vol. 36, No. 5, 2009, pp. 581-599.
[44] Eiró, T. and Martínez, L., Development of a discrete choice model to evaluate the
impact of new transport alternatives. In: Proceedings of the Transportation Research
Board 93rd Annual Meeting, Washington D.C., 2014.
[45] Martínez, L. M. and Viegas, J. M. A new approach to modelling distance-decay
functions for accessibility assessment in transport studies. Journal of Transport
Geography, Vol. 26, No. 0, 2013, pp. 87-96.
[46] International Transport Forum, Urban Mobility System Upgrade: How shared self-
driving cars could change city traffic, OECD Publishing, 2015.
[47] Fagnant, D. J. and Kockelman, K. M. Dynamic ride-sharing and fleet sizing for a
system of shared autonomous vehicles in Austin, Texas. Transportation, No., 2016,
pp. 1-16.
[48] Zhang, W., Guhathakurta, S., Fang, J. and Zhang, G., The Performance and Benefits
of a Shared Autonomous Vehicles Based Dynamic Ridesharing System: An Agent-
Based Simulation Approach. In: Proceedings of the Transportation Research Board
94th Annual Meeting, 2015.
