Research in Transportation Economics xxx (xxxx) xxx–xxx

Contents lists available at ScienceDirect

Research in Transportation Economics

journal homepage: www.elsevier.com/locate/retrec

Research paper

Multi-agent simulation for planning and designing new shared mobility

services
Giuseppe Inturria,∗, Michela Le Pirab, Nadia Giuffridab, Matteo Ignaccolob, Alessandro Pluchinoc,
Andrea Rapisardac,d, Riccardo D'Angeloe
a
Department of Electric, Electronic and Computer Engineering, University of Catania, via Santa Sofia 64, 95125, Catania, Italy
b
Department of Civil Engineering and Architecture, University of Catania, via Santa Sofia 64, 95125, Catania, Italy
c
Department of Physics and Astronomy, University of Catania, and Infn sezione di Catania, via Santa Sofia 64, 95125, Catania, Italy
d
Complexity Science Hub Vienna, Josefstädter Str. 39, 1080, Wien, Austria
e
Mvmant srl, Via Ignazio Silone 21, 95040, Mirabella Imbaccari, Catania, Italy

ARTICLE INFO ABSTRACT

Keywords: Limiting private cars' use while promoting sustainable modes of transport is one of the main challenges of urban
Sustainable mobility transport planning. In this context, characterized by scarce resources and increasing demand for mobility,
Flexible transit Demand Responsive Shared Transport (DRST) services can bridge the gap between shared low-quality public
Demand responsive transport transport and unsustainable individual private transport. Taking advantage of Information and Communication
On-demand mobility
Technologies (ICT), they can supply transport solutions ranging from flexible transit to ride sharing services,
Mobility as a service
providing real-time “on demand” mobility through fleets of vehicles shared by different passengers. The optimal
Agent-based model
design of a DRST service requires a trade-off among efficiency (from the operators' point of view), service quality
JEL classification:
(from the users' point of view) and sustainability (from the community's point of view). In this paper, an agent-
R41
based model (ABM) fed with GIS data is used to explore different system configurations of a specific type of
R42
C63 DRST service, i.e. flexible transit, and to estimate the transport demand and supply variables that make the
service feasible and convenient. The model reproduces a mixed fixed/flexible route transit service with different
fleet size and vehicle capacity in the city of Ragusa (Italy) with the aim to: (i) make a first test of the ABM model
with GIS-based demand and road network models; (ii) explore different vehicle dispatching strategies; (iii) find
appropriate indicators to monitor the service quality and efficiency. Simulation results show the impact of fleet
composition and route choice strategy on the system performance. In particular, they show an optimal range of
operating vehicles that minimizes a total unit cost indicator, accounting both for passenger travel time and
vehicle operation cost. By reproducing the microinteraction between demand and supply agents (i.e. passengers
and vehicles), it is possible to monitor the macroscopic behaviour of the system, and derive useful suggestions for
the correct planning, management and optimization of DRST services.

1. Introduction
Mobility is a fundamental component of urban life. The increase in
city population and climate change challenges raise the need for a
more sustainable and efficient transport system (EC-European
Commission, 2011; United Nations, 2016). Policies for sustainable
mobility generally act on three main factors: (1) the amount of
transport activity, i.e. how many kilometres people travel; (2) the
amount of vehicles needed to transport people, which depends on the
capacity of the vehicle and its load factor as well; (3) the energy to
move the vehicle for one km. These factors can be interpreted re-
spectively as land-use, transport and energy intensity (Ignaccolo,
Inturri, Le Pira, Caprì, & Mancuso, 2016).
Public transport has the potential to reach the lowest transport
intensity, as it traditionally uses high-capacity vehicles, well suited to
serve high-density and high-demand mobility corridors. Nevertheless,
fixed route and scheduled transit services often result in low quality of
travel experience and load factors in low density urban areas with
weak and dispersed transport demand. In general, the only alternative
is a massive use of private cars, with very low occupancy rates,

Corresponding author.
∗

E-mail addresses: ginturri@dica.unict.it (G. Inturri), mlepira@dica.unict.it (M. Le Pira), nadia.giuffrida@dica.unict.it (N. Giuffrida),
matteo.ignaccolo@unict.it (M. Ignaccolo), alessandro.pluchino@ct.infn.it (A. Pluchino), andrea.rapisarda@ct.infn.it (A. Rapisarda),
riccardo.dangelo@edisonweb.com (R. D'Angelo).

https://doi.org/10.1016/j.retrec.2018.11.009
Received 30 November 2017; Received in revised form 31 October 2018; Accepted 15 November 2018
0739-8859/ © 2019 Elsevier Ltd. All rights reserved.

Please cite this article as: Giuseppe Inturri, et al., Research in Transportation Economics, https://doi.org/10.1016/j.retrec.2018.11.009
G. Inturri, et al.
Fig. 1. Classification of Demand Responsive Transport services (own setup).
causing road congestion and many other external costs, both at local
and global levels.
This paper focuses on the potential of innovative technological
mobility platforms able to provide transport services that might reduce
transport intensity, by covering the gap between shared low-quality
public transport and unsustainable individual private transport.
In this context, Demand Responsive Transport (DRT) services are
well suited to be both shared and flexible, using fleets of vehicles
scheduled “on demand” to pick up and drop off passengers in ac-
cordance with their needs (Ambrosino, Nelson, & Romanazzo, 2003).
Taking advantage of new ICT, they can dynamically match supply
and demand and allow travellers to request rides in real-time from po-
tential suppliers (e.g. ride sourcing services). They stand between un-
sustainable, flexible and individual transport services offered by private
vehicles (e.g. cars) and sustainable, shared, but low-flexible traditional
public transport services (metro, tram and bus), with different degrees of
sustainability/shareability/flexibility according to the service (Fig. 1). In
this respect, following the emerging concept of sharing economy, they
can be shared among different passengers. Demand Responsive Shared
Transport (DRST) services can be regarded as a tool to shift away from a
culture where consumers own assets (cars), toward the Mobility-as-a-
Service (MaaS) culture, where consumers “share access to assets”
(Ambrosino, Nelson, Boero, & Pettinelli, 2016). DRST encompasses dif-
ferent types of services, from vehicle sharing (e.g. car sharing or peer to
peer car sharing) to ride sharing. Ride sharing services can be performed
by taxi sharing or carpooling, or they can have the connotation of
“flexible” transit (e.g. microtransit or bus on demand). In particular,
well-designed flexible transit services are able to find the right compro-
mise between flexibility and fixed scheduling in order to have a reliable
service (like traditional public transport), while flexible and adaptable,
e.g. by changing (route) according to the demand.
The optimal design of a generic DRST service requires a trade-off
among efficiency (from the operators' point of view), service quality
(from the users' point of view) and sustainability (from the community's
point of view). The system stands between an expensive conventional
exclusive-ride door-to-door service (like a conventional taxi), and a
cheaper service where a dynamic sharing of trips makes users experi-
ment longer travel distances and times, while the vehicles drop off and
pick up other passengers (like a conventional bus). From the operator's
point of view, it is important to select the optimal strategy to assign
vehicles to passengers' requests, so to perform high load factor (to in-
crease revenues) and low driven distance (to reduce operation costs),
2
Research in Transportation Economics xxx (xxxx) xxx–xxx
while minimizing the additional time and distances travellers have to
experience (to assure the expected level of service). The problem could
be approached by using operations research models, able to provide a
real-time optimization to satisfy the dynamic demand. By the same
token, but with a different approach, microsimulation models can be
used to reproduce DRST services and predict the impacts of different
strategies on the service efficiency. In particular, agent-based simula-
tions are a suitable tool to understand the complexity of transport
systems and the emergent phenomena deriving from the interaction
between individual agents with different objectives and behaviours.
This paper contributes to the development DRST, in particular
flexible transit services, by presenting an ad-hoc agent-based model
(ABM), implemented to test the impact of different operation strategies
on service efficiency. The city of Ragusa (Italy) is chosen as case study,
where an innovative DRST service has already been tested. The model is
used as a realistic environment where to simulate different scenarios,
with simple rules assigned to agents’ behaviour, in order to explore the
transport demand and supply variables that make the service feasible
and convenient. Starting from the microinteraction between demand
and supply agents (i.e. passengers and vehicles), the macroscopic be-
haviour of the system is monitored, via appropriate indicators, and its
performance evaluated to establish criteria for an optimal service de-
sign and operation.
The remainder of the paper is organized as follows. Section 2 will
present a literature review on shared transport services models; in
Section 3, the methodology will be described, in particular the ABM
with its main features (3.1); Section 4 will present the case study to-
gether with a description of scenario simulations (4.1) and results (4.2);
Section 5 will resume the paper highlighting future research directions.
2. Modelling approaches to shared mobility
Shared transport services have gained increasing attention in the
last years. From the demand side, the potential acceptability of these
services has been investigated, e.g. via stated preference surveys (Frei,
Hyland, & Mahmassani, 2017; Krueger, Rashidi, & Rose, 2016; Ryley,
Stanley, Enoch, Zanni, & Quddus, 2014). From the supply side, data
analysis techniques and optimization models have been proposed to
study service functioning (Santi et al., 2013; Tachet et al., 2017). Tra-
ditionally, the problem has been formalized as dial-a-ride (Stein, 1978)
or multiple depot vehicle scheduling (Bodin & Golden, 1981). More
recently, simulation models have been developed for different
G. Inturri, et al.
purposes, e.g. to: study the interaction between demand and fleet size
(Bailey & Clark, 1987); compare usability and profitability of dial-a-ride
and fixed-route systems (Shinoda, Noda, Ohta, Kumada, & Nakashima,
2004); evaluate the efficient scheduling of dynamic DRT services, e.g.
via a call-in simulator (Diana, 2006); evaluate dynamic taxi-sharing
systems, e.g. via the “DIVERT” simulation platform (D'Orey, Fernandes,
& Ferreira, 2012; Bischoff, Maciejewsk, & Nagel, 2017); investigate the
effects of using a zoning vs. a no-zoning strategy and time-window
settings on the performance of a DRT service (Quadrifoglio, Dessouky,
& Ordonez, 2008); evaluate the performance and profitability of DRT
services providers over multiple years of operation (Cich et al., 2017;
Horn, 2002); analyse first and last miles connections, e.g. via the
“NITS” (Network-Inspired Transportation System) simulation model
(Edwards, Elangovan, & Watkins, 2012).
In the framework of simulation models, ABM are very useful to help
the design of DRST services, by testing different strategies and system
configurations, while reproducing the dynamic interaction between
passengers and vehicles (Cich et al., 2017).
Contrary to traditional simulation modelling, where one imple-
ments a model according to a centralized theory and analyses the be-
haviour of the system as a results of different input parameters and
scenarios, in ABM we do not have a pre-set model: one provides agents
with few rules of behaviour and searches for emerging patterns through
the simulation. Another difference stands in the level of aggregation:
even if traditional behavioural models can be disaggregated and de-
tailed, they always presume limited options, a rational behaviour and
maximizing goals. ABM can be thought as the natural evolution of
behaviour models, where autonomous entities (i.e. agents) can act and
interact according to different simple behavioural rules, increasing the
level of realism and allowing for more complex modelling of interac-
tions. They are suitable to reproduce the interaction among single
stakeholders involved in transport decision-making (Le Pira, Inturri,
Ignaccolo, Pluchino, & Rapisarda, 2015; Le Pira, Inturri, Ignaccolo, &
Pluchino, 2018; Marcucci et al., 2017; Čertický, Jakob, & Píbil, 2015).
They are particularly suitable to simulate passenger and freight trans-
port interactions in urban environment, since such flows co-exist and
share the same physical scarce spaces (Maggi & Vallino, 2016). They
have been proposed to test different transport services, e.g. taxi fleet
(Cheng & Nguyen, 2011), collective taxi systems (Lioris, Cohen, & de La
Fortelle, 2010), car sharing (Fagnant & Kockelman, 2014; Lopes,
Martineza, & de Almeida Correia, 2014; Martínez, de Almeida Correia,
Moura, & Lopes, 2017), shared self-driving services (Martínez & Viegas,
2017), and also to investigate DRT services, analysing the profitability
of thin flow service providers over multiple years (Cich et al., 2017) and
developing an open-source simulation testbed (Čertický, Jakob, & Píbil,
2016).
Main benefits of ABM can be summarized as follows (Bonabeau,
2002):
• possibility to capture emergent phenomena, resulting from interac-
tion among the constituent units of the system and which are dif-
ficult to predict;
• providing of a natural description of a system, since for some sys-
tems it is more convenient to describe the behaviours of the in-
dividual entities rather than to come up with the equations that
govern the system's dynamics;
• flexibility, for the easiness to add more entities to the model, to
modify behaviour, degree of rationality, ability to learn and evolve,
and rules of interactions of agents. They also allow to change levels
of description and aggregation, making possible to start the simu-
lation before the level of system's complexity is still unknown.
In Table 1, several examples of recent approaches to shared mobility
modelling are classified according to the type of shared mobility and
the techniques used.
Literature on modelling approaches to shared mobility is
3
Research in Transportation Economics xxx (xxxx) xxx–xxx
increasingly growing in the last years. However, there is still room for
improvement to increase the capability of models to reproduce real-
world systems and tailor them according to the specific context.
In this respect, Quadrifoglio, Dessouky, and Ordonez (2008) con-
ducted a study of the operations of DRT providers on a network based
on data for a service in Los Angeles County; however, it is a static
problem, as they suppose that all the requests for any given day of
service are made at least one day in advance. Carotenuto et al. (2012)
address the simulation of a DRT in a dynamic mode, where requests are
assigned to vehicles, taking into account the actual time the request has
been received and the actual vehicle position; however, they use a
discrete event simulation environment composed of a set of modules
each of which reproduces the behaviour of the physical objects, but this
is done without any reference to a real territory or a real road network.
Čertický et al. (2014) state that transport specific simulation tools
(such as MATSIM or SUMO3) have proven insufficient for simulation-
based assessment of a wider variety of DRT systems. The authors pre-
sented an agent-based simulation testbed built on top of the transport
simulation framework AgentPolis to assess the performance of on de-
mand mobility services under different conditions; however, most of
the software architecture is hidden and does not allow for fine tuning of
some variables, e.g. users' availability to wait for the service or ran-
domness in the dispatching of vehicles to routes.
In this paper, we investigate a DRST service, which is a dynamic
flexible route transit service. It is dynamic because arriving demand
requests affect the way routes are assigned to vehicles; it is a flexible
transit because it is composed of fixed and optional routes. The fixed
routes assure a certain regularity of the service allowing, in principle,
users to access it without booking in advance, while other routes are
optional and demand responsive, allowing low density areas to be
served by a transit service.
A new ABM is presented to simulate this particular category of
DRST service, pertaining to flexible transit services, under different
system configurations, by simulating the interaction between vehicles
travelling along a flexible road network and users walking from their
origins to stop nodes to get a transport service to their destinations. The
main novelties of the model rely on the implementation of: (1) an ad-
hoc GIS-based demand model integrated in the simulation environment,
implying an easy transferability to other contexts, (2) different dis-
patching strategies that allow a dynamic matching between vehicles
and passengers; (3) a flexible simulation interface that allows to easily
change variables and monitor the state of the system in real-time. The
model has been first described in Inturri et al. (2018) and preliminary
results presented. Here, we extend the description, provide more details
about agent dynamics, and present results of different sets of simula-
tions, useful for service design and customization.
The model can be easily adapted so to reproduce different DRST
services and be applied to other contexts. It allows exploring the
emergence of optimal operation configurations, by identifying ad-hoc
indicators to monitor the system's performance. The ABM is able to
capture the performance of the DRST which depends on how vehicles
are allocated to passengers, routes and schedules, on the particular
topology of the road network and on the spatio-temporal pattern of
passenger demand. In the following, the main characteristics of the
ABM are presented.
3. Methodology
The aim of the ABM is to reproduce innovative DRST services and
test the impact of different vehicles dispatching strategies on the service
efficiency and quality. NetLogo has been chosen as modelling platform
(Wilensky, 1999), being totally programmable and customizable, thus
offering a suitable environment to test and monitor the performance of
new transport solutions under different system configurations. The
main characteristics of the model are:
G. Inturri, et al.
Table 1
Literature review on shared-mobility modelling.
Scheme of shared mobility Approach Authors
Taxi sharing Data analysis Santi et al., 2013; Tachet et al., 2017
ABM Cheng & Nguyen, 2011; Lioris et al., 2010
Simulation model D'Orey et al., 2012; Bischoff et al., 2017
Flexible Transit Simulation model Quadrifoglio, Dessouky, & Ordonez, 2008; Horn, 2002; Edwards et al., 2012; Carotenuto, Monacelli, Raponi, & Turco,
2012
Stated preference survey Frei et al., 2017; Ryley et al., 2014
ABM Cich et al., 2017; Čertický, Jakob, Píbil, & Moler, 2014
Shared Autonomous Vehicles Stated preference survey Krueger et al. (2016)
ABM Martínez & Viegas, 2017; Scheltes & de Almeida Correia, 2017
Car sharing ABM Lopes et al., 2014; Fagnant & Kockelman, 2014
i. Dynamic vehicle sharing,
ii. Mixed fixed and flexible routes,
iii. On-demand stops in prefixed points,
iv. Flat fare,
v. Origin/destination demand pattern characterized by trip generation
proportional to zone population and trip distribution based on a
gravity model.
The simulated service is “demand responsive” since vehicles can
adapt their routing according to the demand. It is “flexible” since they
travel along a fixed route and can switch to optional routes according to
the route choice strategy. Actually, this first version of the model in-
cludes a limited cognitive agent behaviour, in order to have a simple
tool able to reproduce a basic DRST service that could be easily mod-
ifiable and adaptable to reality. In this respect, vehicles do not have a
complete knowledge of the state of the system, i.e. they only know how
many people are waiting at the nearest stop or at the stops along the
flexible route they are allowed to drive on. There is no cooperation
among drivers; users do not know the location of the vehicles on the
network nor the waiting time for the next arrival. The system is self-
organized and the model measures its performance at a global level. It
is clear that in a real-world case the system would be based on a digital
platform, allowing users to know the position and arrival time of ve-
hicles (e.g. via an app), and vehicles to know the demand. However, the
aim of the first version of the model is to help the strategic design of
such service, implying an easy matching between demand and supply,
trying to understand how local strategies and system variables can
change the overall performance. The details of the model are presented
below.
3.1. Description of the model
A description of the model based on Inturri et al. (2018) is here
presented, according to its main features, i.e. (i) transport network, (ii)
demand model, (iii) agent (passenger and vehicle) dynamics, (iv) route
choice strategies, (v) performance indicators. In this paper, more details
about relevant features (e.g. agents’ behaviour) are provided, and a new
analytical description based on the ODD (Overview-Design Concepts-
Details) protocol (Grimm et al., 2006; 2010) is given (Table 2).1
Transport network. The network consists of a fixed route and some
optional routes, composed of network nodes and links, stop nodes and
diversion nodes (to skip from fixed to optional routes and vice versa).
The GIS extension of NetLogo is used both to build the actual road
network and map the distribution of population (residents and em-
ployees) at census zone level.
Demand model. A users’ group trip request is generated from an
origin (O) zone to a destination (D) zone with a negative exponential
distribution, according to a gravitationally distributed probability.
1
This description based on ODD protocol should be intended as preliminary
and not exhaustive, since we are presenting the first basic version of the model.
4
Research in Transportation Economics xxx (xxxx) xxx–xxx
Given a set of n zones, the trip rate TRij generated for the generic O-
D ij is calculated with equation (1), where TRi is the generation trip rate
of the zone i, proportional to population density and to an average trip
rate (ATR) (simulation variable), calculated with equation (2), and pij is
the probability that a trip with origin i has destination in the zone j,
calculated with equation (3).
TRij = TRi pij (1)
Popi
TRi = n ATR
k=1
Popk (2)
Addj dij e dij
pij = n dkj
Addj dkj e (3)
k=1
Addj is the number of employees working in zone j, dij is the distance
from i to j, α and β are the parameters of the decay function
f (d ) = d e d .
Agent (passenger and vehicle) dynamics. Agent dynamics is de-
scribed in Fig. 2.
A trip request of a passenger group (with a maximum prefixed size)
is stochastically generated, according to the demand model. For the
sake of simplicity, in this first version we used a simplified mode choice
model. In this respect, the choice is twofold, i.e. between flexible transit
and other generic modes of transport (e.g. car, taxi, local public
transport). This assumption is justified by the aim of the simulation, i.e.
to study the basic variables that can help the strategic design of a new
flexible transit service.
The mode choice depends on two variables, i.e. the walking distance
from the origin zone to the closest stop and from the destination stop to
final destination, and the waiting time. In this respect, passenger-agents
assume different “status”, as explained in the following (please see
"accessibility test" in Fig. 2).
If distance from nearest stop to origin or to destination overcomes a
given threshold, passenger group assumes the status “rejected”, since,
in a simplified way, it is assumed that the passenger might decide not to
use flexible transit due to excessive access time and will use other
modes. Otherwise, request is accepted and passenger group moves to
the nearest stop assuming the status “waiting”, while waiting for a
vehicle with an appropriate number of available seats. If a prefixed
maximum waiting time is overcome before a vehicle reaches the stop,
each passenger group gives up and assumes the status of “unsatisfied”.
In this case, it is implicit that they will choose another transport mode
to make their trip. Otherwise, each passenger boards the vehicle as-
suming the status of “satisfied”, and alights at the nearest stop to its
required destination.
For what concerns vehicles, a given number of vehicles with pre-
fixed seat capacity, generated at random stops, starts traveling along
the fixed route at constant speed. At each stop where passengers are in
“waiting” status, they load them following the First-Come-First-Served
(FCFS) queue rule, but only if the passenger group size is not greater
than vehicle available seats (please see "seat test" in Fig. 2). In general,
at each switch node a vehicle can shift to an optional route if there are
G. Inturri, et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Table 2
Model description according to the ODD protocol.
Overview Purpose The aim of the model is to test the impact of different operation strategies (i.e. route choice strategies – RCS and different fleet
size) on efficiency, quality, and effectiveness of the DRST service
Entities, State variables and There are two types of agents: vehicles and passenger groups. Vehicles are characterized by a certain capacity (fixed at the
scales beginning of the simulation) and a number of available seats that is dynamically updated. Passenger groups are characterized
by a number of persons sharing the same trip origin/destination. Time step is 1 s, and simulation time is typically set to 6 h.
The world is composed by a grid of 369 × 201 cells of 15 m each
Process overview and A users' group OD trip request is randomly generated. Users requesting a trip reach the nearest stop and wait for the vehicle to
scheduling come. Vehicles are randomly distributed at stops, and start driving on the assigned routes according to the RCS to pick up/
drop off passengers along their routes. Please refer to Fig. 2 where a block diagram is presented representing passenger and
vehicle dynamics. The simulation is time-discrete, i.e. the system state variables are updated at each time step
Design concepts Design concepts The characteristics of the system (supply + demand) emerge from the microinteraction between passengers and vehicles. In
this first version of the model, no interaction between vehicles is considered nor adaptation or learning strategies. The
objective is to maximize the number of satisfied users and minimize the total number of km travelled by empty vehicles.
Please refer to the performance indicators defined in Section 3
Details Initialization At the beginning of the simulation, N vehicles are randomly generated at different stops. They have a fixed capacity and all
the seats are available. No demand is present at t = 0
Input data The input data regarding demand generates from a gravity model based on census data about population and workers and
network distance. Please refer to Table 3 for more details about the input data used for simulations
Submodels 1. Import map (via NetLogo GIS extension); 2. Import matrix (i.e. OD data and network shortest path distances); 3. Setup (set
variables to initial values at t = 0); 4. Start simulation (“new requests” routine, “vehicle/passenger movement” routine,
“passenger loading/unloading” routine, “time update” routine, “plot” routine)

Fig. 2. Passenger (left) and vehicle (right) dynamics.

passengers in “waiting” status or destination stops for on-board pas-
sengers. The available vehicle seats are updated at each passenger
loading/unloading.
Route choice strategies. All vehicles drive on the fixed route. At
diversion nodes, a vehicle may drive on a flexible route according to the
Route Choice Strategy (RCS) it is assigned to.
In this first version of the model, there are three RCSs:
• “Fully Random” (FR);
• “All Vehicles drive on All flexible Routes” (AVAR); a prefixed per-
centage of vehicles can be assigned at random;
• “Each Vehicle is Assigned to a flexible Route” (EVAR); a prefixed
percentage of vehicles can be assigned at random.
If RCS is FR, each vehicle at a diversion node randomly chooses to
go on the flexible route; approximately, half of the vehicles will drive
on the flexible route, and the other half will keep on driving on the
fixed route, until the next diversion node is reached. If RCS is AVAR,
each vehicle is allowed to drive on a flexible route, but it chooses to do
so only when a boarded passenger has to leave or he/she is waiting at a
stop or according to a random component. If RCS is EVAR, each vehicle
drives only on its prefixed assigned flexible route, except when it is
randomly extracted to keep on driving on the fixed route. The ran-
domness component has been considered to add some “noise” to the
system, since it has been shown that random strategies can have a
beneficial role in increasing the efficiency of social and economic
complex systems (Biondo, Pluchino, Rapisarda, & Helbing, 2013;

5
G. Inturri, et al.
Pluchino, Rapisarda, & Garofalo, 2010; Pluchino, Garofalo, Rapisarda,
Spagano, & Caserta, 2011).
Performance indicators. The local strategies determining interaction
between passengers and vehicles give raise to global patterns that can
be monitored via appropriate performance indicators. They are chosen
to capture the different objectives and points of views of the system
actors, i.e.: (1) a user is interested in reducing the trip cost (length,
travel time, fare); (2) a private company providing the service is in-
terested in maximizing the profit, by increasing the number of pas-
sengers within a prefixed travelled distance or, conversely, in reducing
the amount of travelled distance to serve a prefixed demand; (3) the
community is interested in reducing transport externalities, e.g. air
pollution and congestion.
The main indicators chosen for the analysis are:
• total number of passengers transported NP (pax),
• total driven distance TDD (km),
• average passenger travelled distance APTD (km),
• average vehicle load factor ALF,
• passenger travel time, in terms of average waiting time AWT (min),
average on-board time AoBT (min), and average total travel time
APTT (min),
• average vehicle speed AVS (km/h),
• cost index CI (km/pax), as ratio of TDD and NP,
• total passenger travel time TPTT (h) (including a penalty for each
unsatisfied user),
• operation cost OC (€),
• total unit cost TUC (€/pax), as described below.
The total unit cost per passenger takes into account the total pas-
senger travel time TPTT (h), the value of time VOT (€/h) for passengers,
and the vehicle operation cost OC (€), evaluated according to equation
(4):
TPTT VOT + OC
TUC =
NP (4)
It considers the cost of users and the cost of the operator per each
transported passenger, so it can be seen as a unit cost for the transport
system (demand and supply) as a whole.
Next section will illustrate how the model was tested in a real world
case study.
4. Case study
To test the model, we decided to simulate the flexible transit service
in a real context. The case study is Ragusa, a small-medium city with
approximately 70,000 inhabitants located in the south-eastern part of
Sicily (Italy), where an innovative DRST service, called MVMANT, was
experimented in 2016. MVMANT is an urban mobility platform which
enables the deployment of a fleet of vehicles circulating on prefixed
routes.2 It has been tested as a viable solution to solve the problem of
last mile connections, especially in a city that experiments a poor urban
public transport supply. Ragusa is characterized by two distinct areas,
the upper town and the lower and older town of Ragusa Ibla, with a
high touristic vocation. Several park-and-ride facilities have been built
in the upper town, but they were not properly integrated in a smart
mobility strategy. MVMANT has connected those facilities with the
main destinations in Ragusa Ibla, which is scarcely connected to the
centre, offering a continuous service with midsize passenger vans. The
dynamic forecast of the mobility flow, coupled with the requests in real
time generated by customers, has significantly increased the service
quality and made touristic attractions more accessible. More than 600
citizens were enrolled before starting the service by participating in a
2 3
http://www.mvmant.com.
6
Research in Transportation Economics xxx (xxxx) xxx–xxx
survey where the majority of them stated that they typically used pri-
vate cars, but that they would be willing to change habits and reduce
the number of family cars if the city had offered an efficient transport
system.3 Furthermore, the planning of the same service in Ragusa for
University students is currently under way. Even if the pilot gained a
considerable success, it was conducted for a very short period of time
(i.e. three weeks), thus preventing the possibility to collect statistically
significant data. Nevertheless, it confirmed the effectiveness of well-
designed DRST services, stimulating its replication in other contexts
and further development. Based on this premise, Ragusa's case study
was reproduced via the ABM to: (i) make a first test of the model with
real GIS data; (ii) explore different vehicle dispatching strategies to
optimize the system; (iii) find appropriate indicators to monitor the
service quality and efficiency.
The network used for MVMANT was reproduced in the ABM, and it
is reported in Fig. 3: transport network has fixed (blue) and flexible
(orange) routes; census zones are coloured according to population
(from light to dark green); network and stops are the real ones from the
MVMANT pilot. The service demand is based on a GIS model, which
takes into account population and employment data, while agents’
behaviour is regulated by the simplified assumptions described in
Section 3.1.
The main input variables of the system are:
• service variables, i.e. total simulation time (h), number of vehicles,
vehicle maximum capacity (seats), vehicle average speed (km/h);
• demand variables, i.e. demand rate ATR (request/hour), maximum
number of passengers per request, maximum waiting time (min);
• route choice strategy, i.e. FR, EVAR, AVAR, with a variable per-
centage of randomness.
4.1. Scenario simulations
Two sets of scenario simulations were performed. The first set
considers system operation with different numbers of vehicles, ranging
from 1 to 30, with different seat capacities, so that the total seat ca-
pacity is constant and equal to 30. This was done to test different ser-
vices that, at the two extremes (i.e. 1 vehicle with 30 seats and 30
vehicles with 1 seat) could be assimilated respectively to a minibus
service and a taxi service.
The second set of simulations consider different route choice stra-
tegies (i.e. FR, EVAR, AVAR) with increasing levels of randomness, so to
test the overall system performance. The main input data are sum-
marized in Table 3.
5. Results
Main simulation results are described below. Fig. 4 shows how, with
a given route choice strategy (in this case EVAR with 30% randomness),
waiting time is reduced by higher number of running vehicles, while
on-board time is reduced because more vehicles (of lower set capacity)
reduce shared trips and travelled distance by each passenger. Similarly,
Fig. 5 (left) shows the average vehicle speed (blue line) that increases
with the number of operating vehicles and the “cost index” (green line),
calculated as the ratio between the total driven distance by all vehicles
and the total transported passengers. It can be considered a “transport
intensity index” or a proxy of pollutant emissions per passenger. It
shows a significant double slope that indicates the need of higher fares
to cover the costs as the service gets closer to a taxi service. Fig. 5
(right) shows that the average distance travelled (orange line) by each
passenger is reduced by the number of vehicles in operation, while at
same time the average load factor is reduced (light blue line).
Fig. 6 on the left shows the total hours spent by all passengers (in
https://www.mvmant.com/pilot-in-ragusa/.
G. Inturri, et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Fig. 3. Virtual map (left) and satellite map (right) (Inturri et al., 2018).

Table 3
Input data used for scenario simulations.
Type of input variable Input variables
service variables total simulation time (h)
number of vehicles
vehicle maximum capacity
(seats)
vehicle average speed (km/h)
demand variable demand rate (request/hour),
maximum number of passengers
per demand
maximum waiting time (min)
route choice strategy FR
EVAR
AVAR
the system (10 €/h), adding the operation cost of vehicles (in the range
of 0.5–1.0 €/km according with the vehicle size) and drivers’ cost (20
Input data
€/h), and dividing the sum by the number of passengers. It can be
considered a measure of the total cost sustained by the transport system
6 (demand and supply) to move one person. In this case, an optimal range
1–30 of shared services can be identified within a range between 5 and 10
30–1
vehicles.
30 km/h By considering different route choice strategies, it can be seen that
100 EVAR with a 30% of randomness is the best dispatching strategy, in
1–4 terms of TUC, with 6 vehicles (Fig. 7).
Comparing TUC for EVAR and AVAR strategies with variable ran-
10
100% randomness domness, best results are found with EVAR and no randomness, while
variable randomness AVAR is the worst. This is because the assignment of vehicles to specific
variable randomness routes (EVAR) reduces the empty driven distances. By increasing ran-

Fig. 4. Average waiting time, average on-board time and total travel time (route choice strategy EVAR with 30% randomness; maximum group size = 3; total seat
capacity = 30).

yellow), while waiting (in light blue), on board (in orange) plus a
penalty of 60 min for each users that waited more than a prefixed
threshold at the stop (in grey), and the total number of passengers
transported (in dark blue). This last parameter decreases with 15 and
30 vehicles because group requests with 3 passengers cannot be sa-
tisfied by vehicles with low capacity. On the right, Fig. 6 shows the total
unit cost TUC (€/pax) by the number of vehicles. It is calculated by
attributing a monetary value to each hour spent by the passengers in
domness, it can be seen that the two strategies get closer in terms of
TUC. With randomness more than 40%, the two strategies are the same
with intermediate and almost constant performance. It can be con-
cluded that a certain rate of randomness is beneficial for the AVAR
strategy, since it implies that not all vehicles will simultaneously ex-
plore all the routes if demand is present (thus reducing the empty
driven distance). Vice versa, EVAR strategy works better without ran-
domness. Fig. 8 reports this result in the case of 5 vehicles with 8 seats.

7
G. Inturri, et al. Research in Transportation Economics xxx (xxxx) xxx–xxx

Fig. 5. Cost index and average vehicle speed (on the left), average load factor and average passengers' travel distance (on the right) (route choice strategy EVAR with
30% randomness; maximum group size = 3; total seat capacity = 30).

Fig. 6. Passenger travel time and number of passengers (on the left) and total unit cost TUC (on the right) (route choice strategy EVAR with 30% randomness;
maximum group size = 3; total seat capacity = 30) (Inturri et al., 2018).

Fig. 7. Total unit cost TUC for different number of vehicles (RCS: FR, EVAR with 30% randomness, AVAR with 30% randomness).

8
G. Inturri, et al.
Fig. 8. Total unit cost vs. randomness (number of vehicles = 5; seat capacity = 8; maximum group size = 1) (Inturri et al., 2018).
Fig. 9. Flexible transit vs taxi service (RCS = FR; total seat capacity = 30; maximum group size = 1).
Fig. 9 shows the percentage variation of some of the relevant pre-
vious performance indicators, as a function of the number of vehicles
with variable seat capacity, so that the global capacity is constant and
equal to 30 seats. The value of 100% is assigned to indicators when 30
vehicles with one seat are running; then a percentage change is re-
corded when a lower numbers of vehicles of bigger size is in operation.
An increasing number of vehicles with lower seat capacity (from 1
vehicle of 30 seats to 30 vehicles of 1 seat) represents the shift from a
flexible transit to a taxi-like service. It can be seen that 20% of the
maximum fleet size (6 vehicles) satisfies 78% of the maximum potential
demand, and improves the efficiency, lowering to 26% the amount of
travelled distances by the vehicles due to higher load factors (and so the
impacts on the environment), at the expenses of lower levels of service
for the passengers. In fact, the average travel time increases up to
166%, and the average travelled distance per passenger increases up to
154%. However, the reduction of the cost index could lead to a fare
reduction of over 74%, thus counterbalancing the worsening of the
9
Research in Transportation Economics xxx (xxxx) xxx–xxx
level of service. To sum up, the model could help to find the right trade-
off between level of service (passengers’ point of view) and transport
externalities (due to higher load factor and less travelled kilometres).
6. Conclusion
This paper presented an agent-based model able to simulate flexible
demand responsive shared transport (DRST) services, namely flexible
transit services. This is done to support their strategic planning and
design, by simulating their functioning in a real context. Different
strategies can be tested by changing the service variables and assigning
simple rules to the agents so to explore, starting from the micro-
interaction between passengers and vehicles, the macroscopic beha-
viour of the system. For a first test, the city of Ragusa (Italy) was chosen
as case study, where an innovative DRST has already been experi-
mented. Different scenarios have been reproduced by changing fleet
composition and vehicle dispatching strategy.
G. Inturri, et al.
Simulation results show that the service quality and performance
considerably vary in relation to the number and capacity of vehicles. In
particular, given a fixed supply (in terms of total seat capacity), many
vehicles with low capacity decrease the passenger travel time (and cost)
and increase the operator costs, while few high-capacity vehicles per-
form better from an operator's point of view. An optimal range can be
found by considering a total unit cost accounting both for passenger
travel time and vehicle operation cost.
Besides, different vehicle dispatching strategies have been simulated
to test different system configurations (from more flexible to more fixed
route strategies). Results show that assigning vehicles to specific routes
(i.e. avoiding that all vehicles drive on all routes) reduces travelled
distances by empty vehicles and improves the overall system perfor-
mance. If all vehicles are allowed to drive on all the routes, then a
certain level of randomness in agent's choice is found to be beneficial
for the system performance.
First results show that the model is able to reproduce different
system configurations and to monitor, via appropriate indicators, its
performance.
Further research will focus on: (a) comparing DRST performance
with pure taxi and pure bus services; (b) testing other strategies to
optimize the service (i.e. increase load factor, reduce vehicle-km), e.g.
re-balancing/idle strategies; (c) testing reactive/adaptive agent beha-
viours for route choice strategies based on system states to make the
service more demand-responsive; (d) testing pricing strategies and
public subsidies to increase the service effectiveness (in terms of sa-
tisfied demand); (e) testing the performance of the system with au-
tonomous vehicles; (f) including elasticity of demand to price; (g) in-
cluding the estimation of pollutant emissions and other transport
externalities; (h) improving the demand model (e.g. including socio-
demographics characteristics, data from surveys); (i) test the service
operation taking into account real-time traffic data; (j) validate the
model with real-world data. In this respect, it is planned to use data by
reproducing other case studies of an analogous DRST systems based on
the same platform, e.g. the case of Dubai.4
The final aim is to have a reliable decision-support tool for planning,
management and optimization of DRST services, which can help to
reduce the burden of transport in our cities and contribute to sustain-
able mobility.
Acknowledgments
This work has been partially financed by the University of Catania
within the project “Piano della Ricerca Dipartimentale 2016-2018” of
the Department of Civil Engineering and Architecture and the project
“Piano per la Ricerca 2016-2018 - Linea di intervento 2” of the
Department of Electric, Electronic and Computer Engineering.
References
Ambrosino, G., Nelson, J. D., Boero, M., & Pettinelli, I. (2016). Enabling intermodal urban
transport through complementary services: From flexible mobility services to the
shared use mobility agency: Workshop 4. Developing inter-modal transport systems.
Research in Transportation Economics, 59, 179–184.
Ambrosino, G., Nelson, J. D., & Romanazzo, M. (2003). Demand responsive transport ser-
vices: Towards the flexible mobility agency. ENEA. Italian National Agency for New
Technologies, Energy and the Environment88-8286-043-4.
Bailey, W. A., Jr., & Clark, T. D., Jr. (1987). A simulation analysis of demand and fleet size
effects on taxicab service rates. Proceedings of the 19th conference on Winter simulation
(pp. 838–844). ACM.
Biondo, A. E., Pluchino, A., Rapisarda, A., & Helbing, D. (2013). Reducing financial
avalanches by random investments. Physical Review A, 88(6), 062814.
Bischoff, J., Maciejewsk, M., & Nagel, K. (2017). October). City-wide shared taxis: A si-
mulation study in Berlin. Intelligent transportation systems (ITSC), 2017 IEEE 20th in-
ternational conference on (pp. 275–280). IEEE.
Bodin, L., & Golden, B. (1981). Classication in vehicle routing and scheduling. Networks,
11(2), 97–108.
4
https://www.rta.ae/links/promotions/en/mvmant.html.
10
Research in Transportation Economics xxx (xxxx) xxx–xxx
Bonabeau, E. (2002). Agent-based modeling: Methods and techniques for simulating
human systems. PNAS May: 99, (pp. 7280–7287). Suppl 3.
Carotenuto, P., Monacelli, D., Raponi, G., & Turco, M. (2012). A dynamic simulation
model of a flexible transport services for people in congested area. Procedia-Social and
Behavioral Sciences, 54, 357–364.
Čertický, M., Jakob, M., & Píbil, R. (2015). Fully agent-based simulation model of multimodal
mobility in european cities. Models and Technologies for intelligent transportation systems
(MT-ITS).
Čertický, M., Jakob, M., & Píbil, R. (2016). Simulation testbed for autonomic demand-
responsive mobility systems. Autonomic Road Transport Support Systems, 147–164.
Čertický, M., Jakob, M., Píbil, R., & Moler, Z. (2014). Agent-based simulation testbed for
on-demand mobility services. Procedia Computer Science, 32, 808–815.
Cheng, S. F., & Nguyen, T. D. (2011). Taxisim: A multiagent simulation platform for
evaluating taxi fleet operations. Proceedings of the 2011 IEEE/WIC/ACM international
conferences on web intelligence and intelligent agent technology: Vol. 02, (pp. 14–21).
Cich, G., Knapena, L., Maciejewskib, M., Yasara, A., Bellemansa, T., & Janssensa, D.
(2017). Modeling demand responsive transport using SARL and MATSim. Procedia
Computer Science, 109C.
D'Orey, P. M., Fernandes, R., & Ferreira, M. (2012). Empirical evaluation of a dynamic
and distributed taxi-sharing system. Proceedings of ITSC (pp. 140–146). IEEE 2012.
Diana, M. (2006). The importance of information flows temporal attributes for the effi-
cient scheduling of dynamic demand responsive transport services. Journal of
Advanced Transportation, 40(1), 23–46.
EC-European Commission (2011). Roadmap to a Single European Transport Area-
Towards a competitive and resource efficient transport system. White Paper,
Communication, 144.
Edwards, D., Elangovan, A., & Watkins, K. (2012). Reaching low-density urban areas with
the network-inspired transportation system. Intelligent transportation systems (ITSC),
2012 15th international IEEE conference on (pp. 826–831). IEEE.
Fagnant, D. J., & Kockelman, K. M. (2014). The travel and environmental implications of
shared autonomous vehicles, using agent-based model scenarios. Transportation
Research Part C, 40.
Frei, C., Hyland, M., & Mahmassani, H. S. (2017). Flexing service schedules: Assessing the
potential for demandadaptive hybrid transit via a stated preference approach.
Transportation Research Part C, 76, 71–89.
Grimm, V., Berger, U., Bastiansen, F., Eliassen, S., Ginot, V., Giske, J., ... Huth, A. (2006).
A standard protocol for describing individual-based and agent-based models.
Ecological Modelling, 198(1–2), 115–126.
Grimm, V., Berger, U., DeAngelis, D. L., Polhill, J. G., Giske, J., & Railsback, S. F. (2010).
The ODD protocol: A review and first update. Ecological Modelling, 221(23),
2760–2768.
Horn, M. (2002). Multi-modal and demand-responsive passenger transport systems: A
modelling framework with embedded control systems. Transportation Research Part A,
36(2), 167–188.
Ignaccolo, M., Inturri, G., Le Pira, M., Caprì, S., & Mancuso, V. (2016). Evaluating the role
of land use and transport policies in reducing the transport energy dependence of a
city. Research in Transportation Economics, 55, 60–66.
Inturri, G., Giuffrida, N., Ignaccolo, M., Le Pira, M., Pluchino, A., & Rapisarda, A. (2018).
Testing demand responsive shared transport services via agent-based simulations. In
P. Daniele, & L. Scrimali (Eds.). “New trends in emerging complex real life problems.
ODS, taormina, Italy, september 10–13, 2018”. AIRO springer seriesSpringer
International Publishing978-3-030-00472-9.
Krueger, R., Rashidi, T. H., & Rose, J. M. (2016). Preferences for shared utonomous ve-
hicles. Transportation Research Part C, 69.
Le Pira, M., Inturri, G., Ignaccolo, M., & Pluchino, A. (2018). Dealing with the complexity
of stakeholder interaction in participatory transport planning. In J. Zak, Y. Hadas, &
R. Rossi (Eds.). “Advanced concepts, methodologies and Technologies for transportation
and logistics”. Advances in intelligent systems and computing 572Springer International
Publishinghttps://doi.org/10.1007/978-3-319-57105-8_3.
Le Pira, M., Inturri, G., Ignaccolo, M., Pluchino, A., & Rapisarda, A. (2015). Simulating
opinion dynamics on stakeholders' networks through agent-based modeling for col-
lective transport decisions. Procedia Computer Science, 52, 884–889.
Lioris, E., Cohen, G., & de La Fortelle, A. (2010). Overview of a dynamic evaluation of
collective taxi systems providing an optimal performance. Intelligent vehicles sympo-
sium (IV) (pp. 1110–1115). IEEE IEEE, 2010.
Lopes, M. M., Martineza, L. M., & de Almeida Correia, G. H. (2014). Simulating carsharing
operations through agent-based modelling: An application to the city of Lisbon,
Portugal. Transportation Research Procedia, 3.
Maggi, E., & Vallino, E. (2016). Understanding urban mobility and the impact of public
policies: The role of the agent-based models. Research in Transportation Economics, 55,
50–59.
Marcucci, E., Le Pira, M., Gatta, V., Ignaccolo, M., Inturri, G., & Pluchino, A. (2017).
Simulating participatory urban freight transport policy-making: Accounting for het-
erogeneous stakeholders' preferences and interaction effects. Transportation Research
Part E, 103, 69–86.
Martínez, L. M., de Almeida Correia, G. H., Moura, F., & Lopes, M. M. (2017). Insights into
carsharing demand dynamics: Outputs of an agentbased model application to Lisbon,
Portugal. International Journal of Sustainable Transportation, 11(2), 148–159. https://
doi.org/10.1080/15568318.2016.1226997.
Martínez, L. M., & Viegas, J.,M. (June 2017). 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,
6(Issue 1), 13–27.
Pluchino, A., Garofalo, C., Rapisarda, A., Spagano, S., & Caserta, M. (2011). Accidental
politicians: How randomly selected legislators can improve parliament efficiency.
Physica A, 390(21), 3944–3954.
G. Inturri, et al.
Pluchino, A., Rapisarda, A., & Garofalo, C. (2010). The Peter principle revisited: A
computational study. Physica A, 389(3), 467–472.
Quadrifoglio, L., Dessouky, M., & Ordonez, F. (2008). A simulation study of demand
responsive transit system design. Transportation Research Part A, 42, 718–737.
Ryley, T. J., Stanley, P. A., Enoch, M. P., Zanni, A. M., & Quddus, M. A. (2014).
Investigating the contribution of Demand Responsive Transport to a sustainable local
public transport system. Research in Transportation Economics, 48.
Santi, P., Resta, G., Szell, M., Sobolevsky, S. H., Strogatz, S., & Ratti, C. (2013). Taxi
pooling in New York city: A network-based approach to social sharing problems.
Scheltes, A., & de Almeida Correia, G. H. (2017). Exploring the use of automated vehicles
as last mile connection of train trips through an agentbased simulation model: An
application to Delft, Netherlands. International Journal of Transportation Science and
Technology, 6(1), 28–41.
11
Research in Transportation Economics xxx (xxxx) xxx–xxx
Shinoda, K., Noda, I., Ohta, M., Kumada, Y., & Nakashima, H. (2004). Is dial-a-ride bus
reasonable in large scale towns? Evaluation of usability of dial-a-ride systems by
simulation. Multi-agent for mass user support (pp. 105–119). Springer.
Stein, D. M. (1978). Scheduling dial-a-ride transportation systems. Transportation Science,
12(3), 232–249.
Tachet, R., Sagarra, O., Santi, P., Resta, G., Szell, M., Strogatz, S. H., et al. (2017). Scaling
Law of urban ride sharing. Scientific Reports, 7, 42868. https://doi.org/10.1038/
srep42868.
United Nations (2016). Department of economic and social affairs, population division. The
world's cities in 2016 – data booklet. ST/ESA/SER.A/392.
Wilensky, U. (1999). NetLogo. Center for connected learning and computer based modeling.
Evanston, IL: Northwestern University. http://ccl.northwestern.edu/netlogo/.