Available via license: CC BY-SA 4.0
Content may be subject to copyright.
Integrating Parcel Deliveries into a Ride-Pooling Service -
An Agent-Based Simulation Study
Fabian Fehn∗, Roman Engelhardt, Florian Dandl, Klaus Bogenberger and Fritz Busch
Chair of Traffic Engineering and Control, Technical University of Munich,
Arcisstr. 21, 80333 Munich, Germany
∗corresponding author: fabian.fehn@tum.de
Abstract
This paper examines the integration of freight delivery into the passenger transport of an on-demand ride-
pooling service. The goal of this research is to use existing passenger trips for logistics services and thus reduce
additional vehicle kilometers for freight delivery and the total number of vehicles on the road network. This is
achieved by merging the need for two separate fleets into a single one by combining the services. To evaluate
the potential of such a mobility-on-demand service, this paper uses an agent-based simulation framework and
integrates three heuristic parcel assignment strategies into a ride-pooling fleet control algorithm. Two integra-
tion scenarios (moderate and full) are set up. While in both scenarios passengers and parcels share rides in one
vehicle, in the moderate scenario no stops for parcel pick-up and delivery are allowed during a passenger ride to
decrease customer inconvenience. Using real-world demand data for a case study of Munich, Germany, the two
integration scenarios together with the three assignment strategies are compared to the status quo, which uses
two separate vehicle fleets for passenger and logistics transport. The results indicate that the integration of logis-
tics services into a ride-pooling service is possible and can exploit unused system capacities without deteriorating
passenger transport. Depending on the assignment strategies nearly all parcels can be served until a parcel to
passenger demand ratio of 1:10 while the overall fleet kilometers can be deceased compared to the status quo.
Keywords: integration of passengers and freight, mobility-on-demand, ride-pooling, fleet control, parcel deliv-
ery, agent-based simulation
1 Introduction
The development of urban transportation is subject to constant change. In recent years, passenger transporta-
tion has been subject to the influences of new mobility services such as car-sharing, ride-hailing, ride-pooling,
or on-demand public transportation services, including autonomous shuttle buses. In addition to passenger
transportation, urban freight transportation is also subject to disruptive developments. Many providers are
relying on fast delivery options, such as same-day or next-day delivery options, which are increasingly provided
by subcontractors. Furthermore, there has been a trend towards more environmentally friendly delivery forms,
such as the use of bicycle couriers or CO2-compensated delivery options in recent years. The term ’crowd logis-
tics’ is currently also heavily discussed and describes a (typically fast) shipping service outsourcing the delivery
to many individuals, often private persons. The European Commission [1] compared the influences of passenger
and freight traffic in terms of total CO2emissions in the European Union and found that passenger and freight
transportation account for approximately 60% and 40%, respectively. More specifically, urban passenger traffic
accounts for 17% and urban delivery traffic for 6% of the total amount of transportation-related CO2emissions
(Figure 1).
Along with that, the attitude of city residents is also evolving. While it was inconceivable a while ago
that people would completely do without their privately owned cars, many citizens now rely on a combination
of public transportation, sharing, and on-demand services. Generally, one can observe a more ’eco-friendly’
lifestyle among consumers [2]. The so-called ’sharing economy’ describes a social trend of our time, with people
foregoing owning things, preferring instead to rent them or share them with others. However, there has been an
increase in passenger kilometers traveled per day from 2,717 million km in 2002 over 3,080 million km in 2008
to 3,214 million km for the year 2017 in Germany [3]. Moreover, online shopping increased significantly, and
1
arXiv:2205.04718v1 [cs.MA] 10 May 2022
Submitted 30. April 2022 2 STATE OF THE ART ANALYSIS AND LITERATURE REVIEW
Figure 1: Freight and passenger shares of European CO2emissions from transportation1
accordingly, a drastic increase in home delivery traffic emerged. The annual relative growth in parcel shipments
in Germany was 10.9% in 2020 compared to 2019, with a total of 4.05 billion shipments. This is mainly due
to private customer shipments, which increased by 18.6%, whereas business customer shipments even decreased
by 5.2% [4].
Along with the mentioned developments in transportation services and social trends, urban transportation
problems have also evolved. In addition to the more transport engineering related problems, such as increased
travel times or noise and pollutant emissions, transport planning problems, such as the lack of space or the
division of urban districts, become also present in many European cities. Therefore, it is time to find and
promote environmentally friendly and space-saving forms of transportation. In particular, the sustainable
utilization of existing infrastructure plays a decisive role, as this is not directly linked to the creation of new
infrastructure with the associated emissions and space consumption. With the introduction of automation in
urban transportation, a time-optimized usage of the existing infrastructure could be realized, by shifting certain
transportation tasks to off-peak times. Furthermore, new cost structures could establish, since the largest cost
factor, the driver would be eliminated. This development could accelerate the trend towards better utilization
of existing infrastructure and has therefore high disruptive potential.
One possible solution to the above-mentioned urban transportation problems, making use of current social
trends and ecological and technological developments, is the integrated transport of passengers and freight in
an urban context. Especially, in the field of mobility on-demand (MoD), due to its high temporal and spatial
flexibility and centralized decision-making, there seem to be promising opportunities to serve passengers and
freight with one fleet of vehicles. For this reason, this paper deals with strategies that enable the integrated
transportation of passengers and parcels in a MoD fleet. This study evaluates the potential of such a service
and assesses its impact on different stakeholders and the overall transportation system. Furthermore, this paper
examines the real-world applicability and explores system boundaries with the help of a case study.
The outline of this paper is as follows: first, it describes the state of the art in research and real-world
applications. Thereby, the different areas of application of integrated transport of passengers and freight, as well
as existing models and solutions and case studies are discussed. The next chapter describes the methodology
for the simulation and case study in Munich, Germany. Subsequently, it presents the obtained results and
investigates Key Performance Indicators (KPIs) for the customers and the service operator. Last but not least,
this paper discusses the findings and future research directions.
2 State of the Art Analysis and Literature Review
2.1 Integrated Passenger and Freight Transportation and Real-World Applica-
tions
The idea of transporting freight simultaneously with passengers is by no means new. As early as 1610, the first
documented stagecoach traveled between Edinburgh and Leith, carrying passengers as well as smaller parcels on
its journey [5]. Nowadays, the joint transport of passengers and parcels is also not uncommon. Today’s passenger
aircraft typically also handle freight. The combined transport of people and goods is also quite common in
1According to PRIMES and TREMOVE Models of the European Commission [1]
2
Submitted 30. April 2022 2 STATE OF THE ART ANALYSIS AND LITERATURE REVIEW
maritime shipping and ferry service. However, much fewer services exist for combined transportation in an
urban context. Yet, intelligent solutions are needed here in particular, as urban transport phenomena, such as
increased travel times [6], space shortages, noise emissions, and poor air quality have worsened considerably in
recent years, according to the German Ministry of Environment [7]. For this reason, a couple of smart approaches
have already emerged that could lead to an improvement in the aforementioned combined transportation of
passengers and freight in an urban context. To classify the existing urban concepts and to investigate existing
simulation studies, this paper analyses the existing concepts of integrated passenger and freight transportation.
A classification between the different forms can be made by distinguishing between public transportation (rail-
and road-based) and individual transportation.
The integration of rail-based public transportation systems and freight transportation can have many faces,
thereby one of the most researched forms is the railroad transportation sector [8, 9, 10]. The combined trans-
portation in urban subway systems, especially in off-peak times, is also a very promising approach and has
therefore already been examined in several research papers [11, 12, 13]. Combinations of Personal Rapid Tran-
sit and Freight Rapid Transit are also subject of research and combine the two transportation streams on rail [14,
15]. Another interesting approach is to use urban tramways [16]. The city of Zurich launched an operational
service called Cargo-Tram and E-Tram in 2003 [17], which acts as a recycling center on rails and transports
large and heavy waste and electronic devices in off-peak times, however, without passengers on board. The
Karlsruhe Institue of Technology and the University Frankfurt am Main of Applied Sciences research integrated
passenger and freight transportation approaches in the ’LogIKTram’ and ’LastMileTram’ research projects [18,
19]. Another solution could be the combined transport of passengers and freight in urban cable cars [20]. Road-
based public transportation systems are also a matter of discussion when it comes to integrating passenger
and freight transport. Especially, bus systems are subject of current research [21, 22, 23, 24]. An interesting
concept study in this category is the ’Freight*Bus’, which according to the persons in charge could change the
economic and environmental costs of passenger and freight transportation in modern cities [25]. One of the
probably best-known examples of a road-based, multi-purpose vehicle approach is the Toyota e-Palette concept,
which was presented in the context of the Olympic Games 2020 in Tokyo [26]. The e-Palette concept aims at
combining multiple purposes, such as passenger and freight transportation or mobile shopping facilities, on one
autonomous vehicle platform.
Apart from public transportation infrastructure, there is also the possibility to integrate passenger and
freight flows into private transportation. In the case of crowdsourced delivery, the introduction of new vehicles
or infrastructure becomes superfluous, because these approaches rely on already existing vehicles or passenger
trips. The concept is a kind of ride-sharing service for parcels, where the parcels are picked up and driven to their
destination by registered private individuals for a small fee [27, 28] or bonus points [29]. Privately operated
transportation services, such as taxi services, should also be mentioned in this context and are themselves
already subject of research [30, 31]. Especially, centrally controlled MoD vehicle fleets reveal a high potential
for the combined transportation of passengers and freight, as the decision-making (i.e. vehicle assignment and
routing) is bundled. In autonomous mobility on-demand (AMoD) systems, the driver is no longer a cost- and
time-limiting factor, thus enabling new application scenarios for combined transportation [32]. Apart from the
car as a transport vehicle, the combined transportation of passengers and freight on two-wheeled vehicles (i.e.
bicycles and motorcycles) has also already been studied [33].
The combination of MoD for passengers and parcels is the core research field of this paper and this research
refers to it as ride-parcel-pooling (RPP), in reference to the widely used term ride-pooling, which describes the
joint and simultaneous transportation of passengers with similar origin-destination relationships in one vehicle.
2.2 Integrated Transport in Mobility-on-Demand Research
Several studies dealt with the control and the efficiency of mobility on-demand services in recent years. Ride-
hailing services can reduce the overall needed fleet size compared to private vehicles [34] or car-sharing ser-
vices [35] because of higher temporal utilization, even without shared rides. To also reduce vehicle kilometers
in a mobility system, rides have to be shared using ride-pooling services to overcome empty pick-up trips [36,
37]. Nevertheless, it has been shown that the efficiency of pooling, i.e. the probability of finding shareable
trips, heavily depends on fleet size and especially the overall demand [38, 39]. While the utilization of fleet
vehicles can be increased by pro-actively distributing idle vehicles (i.e. re-balancing) according to expected
demand [40, 41], vehicles remain idle when no passenger demand is present. The integration of parcel demand
can fill these idle items, however, new fleet control algorithms have to be developed. Most state-of-the-art ride-
pooling assignment algorithms heavily utilize explicit time constraints on customer pick-ups and in-vehicle travel
times enabling graph-based approaches [42, 43, 44]. Nevertheless, these approaches become computationally
intractable if these time constraints are relaxed, which is the case for most parcel assignment problems.
The integration of on-demand passenger and freight transportation is a heavily discussed topic in the expert
community. The underlying optimization problem for the routing of vehicle fleets is a variant of the Vehicle
Routing Problem (VRP). The VRP is one of the most studied optimization problems in transportation research
3
Submitted 30. April 2022 2 STATE OF THE ART ANALYSIS AND LITERATURE REVIEW
and has already been addressed in numerous publications. A very good overview of the different approaches can
be found in the literature reviews by Eksioglu et al. [45] for the general VRP and Kim et al. [46] for the city VRP.
In terms of joint transport of passengers and freight, Cavallero and Nocera [47] conducted a concept-centric
literature review to classify the different integration concepts.
The present paper focuses on the integration of MoD of passengers and same-day delivery of freight and
classifies the existing literature on the subject according to the categories: overall integration concept, passen-
ger mobility, and logistics service characteristics, optimization approach, and, last but not least, the real-world
examination of the results in a case study (Table 1).
This chapter collects the different existing approaches, case studies, and mathematical models and algorithms
dealing with the modeling of integrated passenger and freight transport in MoD ride-pooling and ride-hailing
services. Mourad et al. [48] summarize models and algorithms for optimizing shared mobility, which serves as
an input for the literature collection in this chapter and Table 1. In their paper the authors summarize the
recent research activities in the field, including different optimization approaches, to provide guidelines and give
promising directions for future research. In the following, this paper gives a brief overview of the methodology
and the results of existing research, and the respective contribution to the scientific discourse.
B. Li et al. [49] investigate a Share-a-Ride Problem (SARP) for passengers and parcels sharing taxis. There-
fore, the authors propose a reduced SARP and use the Freight Insertion Problem (FIP) to insert parcel requests
into the vehicle routes. They evaluate their model with historical taxi data from San Francisco and performed a
numerical study of static and dynamic scenarios. L. Li et al. [31] research a multi-agent cooperative inter-modal
freight transportation planning approach for multiple inter-modal freight transport operators. In a simulation
study, they show the potential of their cooperative planning approach. Ngoc-Quang et al. [50] introduce a prac-
tical hybrid transportation model for serving passengers and parcels in the same fleet of taxis. They show the
feasibility and efficacy of their time-dependent model using two different heuristic algorithms. In the next step,
the authors use a real-world data set of the ’Tokyo-Musen’ taxi company to simulate the joint transport of pas-
sengers and freight. Therefore, they introduce a practical hybrid transportation model for Tokyo city to handle
passengers and parcels in the same vehicle. The paper adopts the model of Li et al. [31] to a real-life use case.
The simulation results are subsequently analyzed on the total benefit for the taxi provider, the accumulated fleet
distance over the day, and the number of used taxis and served requests. Chen and Pan [30] propose a taxi fleet-
based crowd-sourcing solution to last-mile delivery for an e-commerce delivery use case. Their approach relies
on a two-phase decision model, first an offline taxi trajectory mining, and second an online package for routing
and taxi scheduling. Ronald et al. [51] create a simulation of a VRP and a FIP. They explore the integrated
MoD passenger and freight transportation problem using a simulation of an on-demand transportation scheme.
They investigate three different scenarios: 1) each shop has its delivery vehicles, 2) all shops share delivery ve-
hicles, and 3) a co-modal scheme where passengers and parcels can share vehicles. The authors show, that their
co-modality approach, using household survey data from Melbourne to generate demand for passengers and
parcel customers, can provide an improved experience for both, operators and customers and that on-demand
co-modality is more resilient to uneven or unexpected demands, and provides more options for travel compared
to the other two scenarios. Soto Setzke et al. [52] suggest an algorithm optimizing the assignment of drivers
to transport requests, using transport routes and time constraints as inputs. They evaluate their algorithm
based on a mobility data set from a major German city. Chen et al. [53] try to exploit taxi trips by creating
relays of multiple trips to integrate the transport of parcels, without degrading the quality of passenger services.
They develop a city-wide parcel delivery system, leveraging a crowd of taxis in Hangzhou, China. The authors
introduce a two-phase framework, which in the first phase mines historical taxi data offline and in the second
phase uses an online adaptive taxi scheduling algorithm to iteratively find near-optimal delivery paths. Their
results show, that over 85% of parcels can be delivered within 8 hours, with around 4.2 trans-shipments/relays
on average. Kafle et al. [54] consider cyclists and pedestrians as crowdsource for last- and first-mile operations.
The distributors request deliveries by submitting bids to the logistics operator. The authors find, that the total
truck miles traveled and the total cost can be reduced compared to pure-truck delivery. Beirigo et al. [55] model
a variation of the People and Freight Integrated Transport (PFIT) Problem. They find that mixed-purpose
fleets perform on average 11% better than single-purpose fleets. Qi et al. [56] present new logistics planning
models and managerial insights. The proposed model uses open-loop car routes to assign passenger vehicles to
parcels and fulfill the last-mile delivery job. Their findings suggest that crowd-sourcing shared mobility is not as
scalable as conventional truck-based logistics systems in terms of operating cost. However, the results also show
that reducing the truck fleet size and exploiting additional operational freedom, like avoiding high-demand areas
and peak hours could be of interest. Wang et al. [57] use a crowd of connected vehicles to pick-up and deliver
parcels, by connecting them with a logistics provider. In their simulation study, they used real-world car trips
and assigned the parcel trips accordingly. The authors believe that ride-sharing will be a core service for con-
nected vehicles, which they refer to as ride-sharing as a service. Arslan et al. [58] consider a service platform for
crowdsourced delivery using existing journeys. The authors propose a rolling horizon framework and searched
for an exact solution to the matching problem. Their results suggest that ad-hoc drivers have the potential to
4
Submitted 30. April 2022 2 STATE OF THE ART ANALYSIS AND LITERATURE REVIEW
Ref. Author,
Year Mode MoD Service Logistics Service Optimization Case
Study
hailing pooling imme-
diate
sche-
duled static dyna-
mic
[31] B. Li et al.,
2014 Taxi X-X-X X X
[49] L. Li et al.,
2014 Multi - - X- - X-
[50] Ngoc-Quang et
al., 2015 Taxi X-X- - X X
[30] C. Chen et al.,
2015 Taxi X-X- - X-
[51] Ronald et al.,
2016 MoD X-X- - X X
[52] Soto Setzke et
al., 2017 MoD X- - X-X X
[53] C. Chen et al.,
2017 Taxi X-X-X X X
[54] Kafle et al.,
2017
Ped.,
Cycl. - - X- - X-
[55] Beirigo et al.,
2018 AMoD - X X - - X-
[56] Qi et al.,
2018 Car - - X-X- -
[57] Wang et al.,
2018 Car - - X- - X X
[58] Arslan et al.,
2018 Car X-X- - X-
[59] Mourad et al.,
2019 Multi - X X - - X-
[60] Najaf Abadi,
2019 Taxi X- - X-X X
[61] Y. Chen et al.,
2020 Taxi X-X- - X X
[62] Manchella et
al., 2020 Taxi - X X - - X X
[32] Schlenther et
al., 2020 MoD - X X - - X X
[63] Van der Tholen
et al., 2021 AMoD - X X - - X-
[64] Alho et al.,
2021 MoD X X -X-X X
[65] Fehn et al.,
2021 MoD - X-X X -X
[66] Meinhardt et
al., 2022 MoD - X X - - X X
[67] Zhang et al.,
2022 AMoD - X X - - X X
Table 1: Existing Models in Literature and assorted research characteristics (i.e. Transport Mode, Type of
MoD Service, Optimization Approach, and Case Study)
save up to 37% of vehicle kilometers compared to traditional delivery systems. Mourad [59] investigates in his
PhD thesis how to synchronize people and freight flows. He develops a matching algorithm using an Adaptive
Large Neighborhood Search (ALNS) heuristic and tests the approach in several experiments. Najaf Abadi [60]
develops an on-demand dynamic crowd-shipping system and tried to take advantage of the unused capacities
in taxis. In her PhD-thesis she investigates the effects of an on-demand dynamic crowd-shipping system, using
the publicly available New York City taxi data and freight demand from structural data. She studies the effects
on trip cost, vehicle miles traveled, and peoples’ travel behavior and finds that the proposed crowd-shipping
model has a substantial positive impact on the average total system-wide vehicle miles savings for all scenar-
ios, ranging from 47% to 50%. Chen et al. [61] present a new parcel delivery scheme that takes advantage of
5
Submitted 30. April 2022 2 STATE OF THE ART ANALYSIS AND LITERATURE REVIEW
multi-hop ride-sharing. They tackle the assignment problem using a two-phase solution, which first predicts
the passenger orders and then plans the parcel delivery routes using real-world data from the city of Chengdu.
Their results suggest, that the successful delivery rate may reach 95% on average daytime and is at most 46.9%
higher than those of the benchmarks. Manchella et al. [62] develop a deep reinforcement learning algorithm for
joint passengers and goods transportation using the publicly available New York City taxi data. The proposed
algorithm pools passengers and delivers goods using a multi-hop transit method. The multi-agent simulations
carried out in the paper show, that the approach achieves 30% higher fleet utilization and 35% higher fuel
efficiency in comparison to 1) model-free approaches where vehicles transport a combination of passengers and
goods without the use of multi-hop transit, and 2) model-free approaches where vehicles exclusively transport
either passengers or goods. Schlenther et al. [32] propose a methodology to simulate the behavior of a vehi-
cle fleet serving passengers and freight within an urban traffic system and evaluate its performance using the
simulation environment MATSim. Their results suggest, that the vehicle miles traveled for freight purposes
increase due to additional access and egress trips. Van der Tholen et al. [63] present a method to estimate
the optimal capacity for the passenger and parcel compartments of AMoD systems. They aim at creating an
optimal routing schedule between a randomly generated set of pick-up and drop-off requests of passengers and
parcels. Alho et al. [64] investigate the use of MoD services performing same-day parcel deliveries. Therefore,
they evaluate a cargo-hitchhiking service for e-commerce using MoD vehicles in a simulation-based approach.
Their research aims at testing MoD-based solutions using an agent- and activity-based simulation platform for
the joint transport of passengers and freight. The authors obtain e-commerce demand carrier data collected in
Singapore and investigate operational scenarios fulfilling the deliveries with MoD service vehicles on existing
passenger flows. The results of a case study in Singapore indicate that the MoD services have the potential to
fulfill a considerable amount of parcel deliveries and decrease freight vehicle traffic and total vehicle kilometers
traveled without compromising the quality of MoD for passenger travel. Fehn et al. [65] investigate the combined
transport of passengers and parcels using historical MoD trips and parcel demand. The authors set up a static
optimization problem for the combined transport of MoD passengers and intra-city parcel shipments. Therefore,
the paper uses real-world individual transportation and parcel data from the city of Munich and matches the
parcel data in a static optimization approach on pre-computed MoD passenger trips. The results show, that
depending on the simulated scenario about 80% of the distance traveled to provide the logistics services could
be saved, offering a 100% delivery rate. The study also gives insights into the positive environmental impacts
of such an integrated transportation approach. Meinhardt et al. [66] study the joint transport of passengers
and freight using the Multi-Agent transport Simulation (MATSim) software. The authors apply real-world data
from Berlin in the simulation and showed, that when assuming a relatively large vehicle fleet, passenger waiting
time statistics barely change. Furthermore, a rough cost analysis suggests a large saving potential when using
AMoD vehicles instead of a privately owned vehicle fleet. Zhang et al. [67] propose an AMoD system performing
joint ride-sharing and parcel delivery. The distributed approach, setting up a mixed-integer linear programming
problem and solving it with a Lagrangian dual decomposition method, shows near-optimal solutions in reduced
computation time. Nevertheless, due to the complexity of the problem, only 10 passenger and 5 parcel requests
could be modeled in reasonable computational time.
2.3 Research Gap and Contribution
As the literature review reveals, the range of possible combinations of passengers and freight is very wide. It
comprises different transportation modes, like rail-based public transportation services or individual crowd-
sourced approaches to the idea of central assignment of passengers and parcels to a vehicle fleet, like in the
case of MoD transportation services. In this context, the solution approaches in the MoD area emerge as very
promising due to their high flexibility in terms of spatio-temporal network coverage and relatively low capacity
utilization when no passenger demand is present.
Most recent studies focused either on temporally separated integration of passenger and freight or the
integration into ride-hailing service, where passengers do not share the rides. Furthermore, they assume time
windows for the pick-up and drop-off of parcels, which simplifies the assignment, but does not exploit the full
integration potential of existing passenger trips. This study focuses on the integration of freight transport into
the operation of an MoD ride-pooling service. Thereby, special focus is put on the simultaneous transport of
passengers and freight (i.e. passengers and parcels can be on board the same vehicle). A dynamic simulation
environment is proposed where decisions to serve customers and or parcels have to be made online to model a
realistic setting. As can be seen in Table 1, the authors are only aware of the approach of Alho et al. [64] to
meet similar contributions. Nevertheless, a different approach to integration is investigated in this study. While
recent studies focus on exploiting idle times of MoD fleets explicitly for logistic services and/or modeled the
logistics service as an as-soon-as-possible delivery service (i.e. employing strict time constraints on parcel pick-
up and delivery), the goal of this study is to actively integrate parcel pick-up and drop-off into vehicle routes.
No explicit time constraints are enforced on parcel pick-up and delivery, but the goal is the insertion into vehicle
schedules resulting from the underlying MoD service to minimize the additional driven fleet kilometers for the
6
Submitted 30. April 2022 3 METHODOLOGY
integrated logistic service. To solve the assignment problem of which parcel to assign when into the dynamically
evolving vehicle schedules, three heuristic assignment approaches are proposed. The efficiency of this integration
and the developed assignment approaches is evaluated regarding the impacts on operator, customers and traffic
effects, based on real-world logistics demand, within a case study for Munich, Germany.
3 Methodology
The methodology of this study is described in the following three sections: 1) the different scenarios examined
in this paper to integrate the parcel delivery into the MoD service are defined, 2) the simulation framework
and the proposed algorithms to control the fleet of vehicles to serve passenger, as well as parcel demand, is
introduced and 3) the simulation inputs based on real-world logistics and travel demand data for a case study
in Munich, Germany is described.
3.1 Service Definition and Scenarios of Integration
This research considers a Ride-Parcel-Pooling (RPP) service to investigate the integrated transport of passengers
and freight within the same MoD vehicle fleet. The service consists of a MoD ride-pooling vehicle fleet and
a central operator for decision-making (i.e. vehicle assignment and routing). In the envisioned RPP service,
the transport of passengers has priority over parcel transport requests, as the service for passengers should not
deteriorate. The service assumes that the parcels are transported from urban logistics hubs to the customers. It
is assumed that a pick-up or drop-off of the parcels at the customer’s premises is possible at any time, which in
reality could be realized by small parcel lockers. Therefore, no explicit business hours are modeled for the logistics
service. Nevertheless, parcel pick-ups and deliveries are integrated into passenger trips, therefore most parcel
pick-ups and deliveries will occur during daytime when most passenger demand is present. The characteristics of
the parcels in size and weight are assumed to be boxes, which can be carried by one person and fit into the trunk
of a conventional passenger vehicle. The driving comfort of the passengers does not deteriorate while parcels
are on board the vehicle, as it is assumed that parcels are transported spatially separated, in the trunks of the
vehicles. Furthermore, the service performs a typical logistics service and does not guarantee a certain delivery
period during the day. The goal of the operator is to provide an additional service without compromising too
much on the additional vehicle kilometers covered by the fleet. It also allows the fleet operator to achieve higher
utilization and occupancy rates for the vehicle fleet, and to keep the transport service running even during
periods of low passenger demand, which in turn could lead to higher customer satisfaction. Additionally, the
RPP service could contribute to the goal of the so-called ’livable city’, as two vehicle fleets (i.e. logistics fleet
and ODM fleet) become one and travel kilometers can be saved by pooling passengers and parcels.
To be able to evaluate the RPP service, three scenarios are introduced. Scenario No. 1 displays the current
Status Quo, where passenger and parcel requests are served by two independent vehicle fleets, one specialized in
a ride-pooling MoD service, and the other on a typical urban pick-up and delivery logistics use case. In Scenario
No. 2, denoted by Moderate RPP Integration, the logistics service is integrated into the MoD service and both,
passenger and parcel requests, are served by the same MoD vehicle fleet. However, there is the requirement,
that no parcels are collected or delivered during a passenger trip so that passengers do not experience the parcel
pick-up and drop-off (PUDO) or take detours due to the additional parcel transport. Scenario No. 3, Full RPP
Integration loosens this requirement and thus allows the collection or delivery of a parcel during a passenger
trip. In this scenario, a passenger’s journey may be extended or a detour may have to be accepted due to
the additional parcel transport. Figure 2 illustrates the respective simulation scenarios and gives a schematic
Figure 2: Illustration of the RPP Scenarios with Urban Logistics Depot
7
Submitted 30. April 2022 3 METHODOLOGY
Symbol Description
GStreet network, consisting of nodes (N) and edges (E)
NNetwork nodes
ENetwork edges
rc
iRequest of customer i
oc
i,dc
i,tc
iOrigin, destination, and request time of MoD customer i
rp
iRequest of parcel i
op
i,dp
iOrigin, and destination of parcel i
vMoD vehicle
VMoD vehicle fleet
cc
v,cp
vVehicle capacity for customers, and parcels
twait
max,ttr avel
max Maximum waiting, and travel time for a customer
∆Detour factor
ttdirect
iDirect travel time for request i
ψkFeasible vehicle schedule
Rψ,PψSet of customer and parcel requests
tbTime needed for boarding and alighting
φ(ψk)Objective function for the rating of vehicle schedule ψk
d(ψk)Distance to drive to complete vehicle schedule ψk
PAssignment reward to prioritize passenger transport
Trepo Period of MoD fleet re-balancing
Tmax
pTime when remaining parcels in the vehicles are delivered
τth Parcel detour threshold parameter for assignment
Table 2: List of symbols for the methodology section
overview of potential assignments of requests to vehicles and the resulting routes. It is important to note that
a passenger request always consists of an origin-destination (OD) pair and that a parcel request always starts
at a parcel depot. Thus, for parcel transport, either origin or destination consists of a depot location and the
respective other point corresponds to the pickup or delivery location of a logistics end customer.
3.2 Agent-based Simulation and Fleet Control Strategies
To model the integration of logistics in an on-demand ride-pooling service, the agent-based simulation framework
’FleetPy’ [68] is extended for this use case. The framework consists of four main agents: 1) customers requesting
trips from the fleet operator; 2) parcels that need to be transported by the service; 3) a fleet operator offering
the service and assigning schedules for its fleet of vehicles to pick-up and drop-off customers and parcels, and 4)
vehicles traveling along these assigned schedules within a street network and fulfilling the corresponding pick-up
and drop-off tasks. All symbols used in the following can be found in Table 2.
The street network G= (N, E)consists of nodes Nand edges Econnecting these nodes. Each edge is
associated with a travel time and a distance. A customer request rc
i= (oc
i, dc
i, tc
i)is represented by its origin
location oc
i∈N, the destination location dc
i∈Nand the request time tc
i. In this study, it is assumed that the
parcel delivery request was submitted at least the day before and is therefore known in advance for the whole
simulation period. Thus, a parcel request rp
i= (op
i, dp
i)is only represented by the corresponding origin (pick-up)
location op
iand destination (drop-off) location dp
i.
The goal of the fleet operator is to assign schedules to its vehicles v∈Vto serve customer and parcel
requests. A schedule ψdescribes the sequence of pick-up and drop-off stops assigned to a vehicle. A schedule
is considered feasible, if:
•the drop-off succeeds the pick-up for each request.
•at no point during the schedule the maximum passenger capacity cc
vand maximum parcel capacity cp
vis
exceeded by on board passengers and parcels, respectively.
•for each customer request rc
i, the waiting time (time between request time tc
iand expected pick-up) does
not exceed twait
max.
•the in-vehicle time of each customer rc
idoes not exceed ttravel
max = (1 + ∆)ttdirect
i, with the direct travel
time from its origin to destination ttdirect
iand a detour factor ∆.
In case of modelling the Moderate RPP Integration, an additional constraint is added for a schedule to be
feasible:
8
Submitted 30. April 2022 3 METHODOLOGY
•while passengers are on board the vehicle, no stop is allowed where only parcels are picked up or dropped
off; the transport of parcels and a stop for other passengers, however, is possible.
A feasible vehicle schedule ψk(v;Rψ, Pψ)is defined as the k-th feasible permutation of stops of vehicle vto
serve the set of customer requests Rψand parcel requests Pψwithin the schedule. Stops are associated with
the location, boarding and alighting customers or parcels and the time needed for boarding or alighting tb. In
between stops, vehicles travel along the fastest route in the network G.
Schedules are rated by an objective function φ(ψk(v;Rψ, Pψ)). The goal of the fleet operator is to assign
schedules minimizing the aggregated objective function for all of its vehicles. In this study, the objective function
is defined as:
φ(ψk(v;Rψ, Pψ)) = d(ψk(v;Rψ, Pψ)) −P(|Rψ|+|Pψ|).(1)
d(ψk(v;Rψ, Pψ)) refers to the distance to drive to complete the schedule. |Rψ|and |Pψ|are the number of
customers and parcels to be served with the schedule, respectively. Pis a large assignment reward to prioritize
serving customers and parcels over minimizing the driven distance.
The high-level simulation flow is shown in Figure 3. The demand for the simulation is split into a passenger
request set and a parcel request set. It is assumed that the operator has access to all parcel requests for the
whole simulation period and can freely decide when to serve which parcel. Passenger requests, on the other
hand, are revealed to the fleet operator dynamically during the course of the simulation. Within each time step,
the operator first tries to accommodate new customer requests by inserting them into current vehicle schedules.
With a given frequency, namely every Trepo, re-balancing trips to distribute idle vehicles according to expected
demand are computed. Based on new vehicle assignments, the decision is made to serve specific parcels. Lastly,
vehicle movements and boarding processes are performed. Details on customer insertion, re-balancing, and the
decision process to serve parcels are provided in the following subsections. Because the applied control strategy
of ride-pooling fleets is explored in various other research, the focus of this chapter is on the description of the
methodology to integrate parcel pick-up and deliveries into the existing ride-pooling control strategy.
Figure 3: Flowchart of the simulation framework with different proposed parcel assignment strategies.
3.2.1 Passenger Assignment and Re-Balancing
In order to assign new customer requests to vehicles and corresponding schedules, a simple insertion heuristic
is applied in this study. With the currently assigned schedule ψk(v;Rψ, Pψ)of vehicle v, the pick-up and drop-
off processes for a new customer request rc
iare inserted at all possible positions within the currently existing
sequence of stops (drop-off must follow the pick-up stop). The new set of feasible schedules can be enumerated
again and results in schedules ψ˜
k(v;Rψ∪ {rc
i}, Pψ)if a feasible insertion can be found. The selected vehicle va
and schedule ψlfor serving the customer request is then determined by:
va, ψl= argmin
v,ψ˜
k
φ(ψ˜
k(v;Rψ∪ {rc
i}, Pψ)) −φ(ψk(v;Rψ, Pψ)) ∀v, ψ˜
k,(2)
i.e. the vehicle schedule is assigned that decreases the change in objective value the most when the new request
is served. Each time a new customer requests a trip, the schedules are updated iteratively. If no solution is
9
Submitted 30. April 2022 3 METHODOLOGY
found by the insertion heuristic, i.e. no vehicle can serve the customer within the given time constraints, the
customer leaves the system unserved.
While more sophisticated algorithms to solve the ride-pooling assignment can be found in the literature
(e.g. [43, 69]), using this simple insertion heuristic has the advantage that the assignment of customer and
parcel requests can be decoupled in different decision processes reducing the overall complexity and thereby
computational time.
To distribute idle vehicles dynamically according to expected demand in the network, a re-balancing algo-
rithm is applied in periodical intervals of Trepo. For each zone in the network expected demand is determined
and idle vehicles estimated. A parameter-free re-balancing strategy based on [70] is applied to distribute idle
vehicles by solving a minimum transport problem.
3.2.2 Parcel Assignment
Because no explicit time constraints for parcel pick-up and delivery are imposed in this study, different assign-
ment strategies (in comparison to the passenger assignment) were developed to assign parcels to the vehicle
schedules. While parcels could be served in times of low customer demand to increase temporal vehicle utiliza-
tion, the goal of this study is to evaluate if parcel pick-up and delivery can be performed when vehicles pass
by occasionally to minimize the need for additional driven fleet kilometers. Three different parcel assignment
strategies are developed which will be described in the following.
Combined Decoupled Parcel Assignment (CDPA) In the first assignment strategy, both the origin as
well as the destination of a parcel are assigned at once. An assignment of a parcel request rp
ito vehicle vis
only made if the detour to add the pick-up and the drop-off into the currently assigned schedule ψk(v;Rψ, Pψ)
is small compared to the distance of the direct parcel route d(op
i, dp
i). The detour is measured by comparing the
distance that has to be driven to complete the schedule, including the new parcel request ψl(v;Rψ, Pψ∪ {rp
i}),
with the distance not considering the parcel, i.e. ψk(v;Rψ, Pψ)). A possible assignment is identified if:
d(ψl(v;Rψ, Pψ∪ {rp
i})) −d((ψk(v;Rψ, Pψ))) <(1 −τth)d(op
i, dp
i),(3)
with a threshold parameter τth ∈ {0,1}indicating the amount of detour relative to a direct route to be accepted
to serve the parcel rp
i. If τth approaches 1no detour is accepted to serve the parcel.
Algorithm 1 sketches the procedure of assigning parcels with the CDPA strategy. Let Pube the set of currently
unassigned parcels and Vca the set of vehicles with schedules, that have been updated in the current simulation
time step. The method insert (ψk(v;Rψ, Pψ), r p
i)returns the best feasible insertion of rp
iin ψk(v;Rψ, Pψ)with
respect to the objective function φ. For each unassigned parcel, an insertion is checked for each vehicle with an
updated schedule (an insertion in other vehicles would have already been checked in previous time steps). If
a new schedule fulfills the constraint of Equation 3, a candidate insertion is found. In the end, the candidate
schedule with the minimum objective value is picked to be assigned. If no candidate is found, the parcel is tried
to be assigned again at a later time step.
Algorithm 1 CDPA Insertion
for all rp
i∈Pudo
ψbest =None
vbest =None
for all v∈Vca do
ψ˜
k(v;Rψ, Pψ∪ {rp
i}) = insert(ψk(v;Rψ, Pψ), rp
i)
if d(ψ˜
k(v;Rψ, Pψ∪ {rp
i})) −d(ψk(v;Rψ, Pψ)) <(1 −τth)d(op
i, dp
i)then
if φ(ψ˜
k(v;Rψ, Pψ∪ {rp
i})< φ(ψbest)then
ψbest ←ψ˜
k(v;Rψ, Pψ∪ {rp
i})
vbest ←v
end if
end if
end for
if vbest 6=None then
assignSchedule(vbest, ψbest)
Pu←Pu\ {rp
i}
end if
end for
10
Submitted 30. April 2022 3 METHODOLOGY
Algorithm 2 S-PA Origin Insertion
for all rp
i∈Pudo
ψbest =None
vbest =None
for all v∈Vca do
ψ˜
k(v;Rψ, Pψ∪ {op
i}) = insertOrigin(ψk(v;Rψ, Pψ), rp
i)
if d(ψ˜
k(v;Rψ, Pψ∪ {op
i})) −d(ψk(v;Rψ, Pψ)) <(1 −τth)d(op
i, dp
i)/2then
if φ(ψ˜
k(v;Rψ, Pψ∪ {op
i})< φ(ψbest)then
ψbest ←ψ˜
k(v;Rψ, Pψ∪ {rp
i})
vbest ←v
end if
end if
end for
if vbest 6=None then
assignSchedule(vbest, ψbest)
Pu←Pu\ {rp
i}
Pa
vbest ←Pa
vbest ∪ {rp
i}
end if
end for
Subsequent Decoupled Parcel Assignment (SDPA) The idea of the second assignment strategy is that
– because no time constraints are imposed on the parcel drop-off – the decision on when to drop-off the parcel
does not have to be made when the decision of the parcel pick-up is made. This assumption can be made
under the condition that the loaded parcels do not significantly affect the energy consumption of the vehicles.
Therefore, the decision to pick-up a parcel is separated from the decision to drop-off a parcel. The decision
to pick-up a parcel is taken similar to the CDPA strategy and summarized in Algorithm 2. The differences to
Algorithm 1 can be summarized as follows: firstly, only the insertion of the origin op
iis tested for parcel request
rp
iby the method insertOrigin, secondly, the possible assignment is identified if:
d(ψl(v;Rψ, Pψ∪ {op
i})) −d((ψk(v;Rψ, Pψ))) <(1 −τth)d(op
i, dp
i)/2.(4)
Note that the threshold is divided by 2(compared to CDPA) to account for an even split of the overall detour
for parcel pick-up and drop-off. If a feasible insertion of the origin of rp
iis assigned to vehicle v,rp
iis added to
the set Pa
vto keep track of assigned parcel pick-ups for each vehicle.
A similar approach is chosen to assign the parcel drop-off in any later simulation time step and is sketched in
Algorithm 3. For all vehicles with scheduled pick-ups or on-board parcels, the insertions of drop-offs for all
parcels rp
i∈Pa
vis checked. A possible assignment is considered if:
d(ψl(v;Rψ, Pψ∪ {rp
i})) −d((ψk(v;Rψ, Pψ∪ {op
i}))) <(1 −τth)d(op
i, dp
i)/2.(5)
Algorithm 3 SDPA Destination Insertion
for all v∈Vca do
ψbest =None
rbest =None
for all rp
i∈Pa
vdo
ψ˜
k(v;Rψ, Pψ∪ {rp
i}) = insertDestination(ψk(v;Rψ, Pψ∪ {op
i}), rp
i)
if d(ψ˜
k(v;Rψ, Pψ∪ {rp
i})) −d(ψk(v;Rψ, Pψ∪ {op
i})) <(1 −τth)d(op
i, dp
i)/2then
if φ(ψ˜
k(v;Rψ, Pψ∪ {rp
i})) < φ(ψbest)then
ψbest ←ψ˜
k(v;Rψ, Pψ∪ {rp
i})
rbest ←rp
i
end if
end if
end for
if rbest 6=None then
assignSchedule(v, ψbest)
Pa
v←Pa
v\rp
i
end if
end for
11
Submitted 30. April 2022 3 METHODOLOGY
Thereby, ψl(v;Rψ, Pψ∪ {rp
i})refers to the schedule including origin op
iand destination dp
iof parcel rp
i.
In the simulation, at first possible insertions of parcel drop-offs are tested for vehicles with an updated sched-
ule, then possible pick-up insertions are created for parcel pick-ups analogously to the comprehensive parcel
assignment strategy. This strategy has the downside that it is not guaranteed to find a drop-off for each parcel
until the end of the simulation. However, parcels should not remain on board the vehicles until the end. Hence,
when a certain simulation time Tmax
pis exceeded, all remaining on-board parcels are scheduled to be dropped
off by iteratively inserting them into the current vehicle schedule.
Subsequent Coupled Parcel Assignment (SCPA) In the last assignment strategy, the decision of assign-
ing the drop-off of a parcel is again made independently from the decision of assigning the pick-up. While the
decision of assigning the pick-up remains the same compared to the subsequent independent parcel assignment
strategy (Algorithm 2), the decision of assigning the drop-off is coupled to the passenger assignment. The idea
is to assign passengers and thereby new vehicle schedules passing by the drop-off location of on-board parcels.
By that, the possible solution space for parcel drop-off assignments increases. To this end, the formulation of
the passenger assignment is revisited.
Let Rnew
tbe the set of new customer requests in time step t. In the first step, the best possible solution for just
serving a new customer request rc
i∈Rnew
tis calculated using the same insertion heuristic as for the passenger
assignment. The resulting schedule ψl(va) = ψl(v;Rψ∪ {rc
i}, Pψ)is used as a benchmark for the decision to
assign a parcel drop-off. In a second step, instead of inserting a parcel drop-off into the overall best solution
ψl(va), an insertion of each on-board parcel requests rp
iis tried for feasible schedules in combination with the
new request rc
i, resulting in the schedules ψ˜
l(v;Rψ∪ {rc
i}, Pψ∪ {rp
i})∀v∈Vrc
i. A possible assignment of the
parcel is found if:
d(ψ˜
l(v;Rψ∪ {rc
i}, Pψ∪ {rp
i})) −d(ψl(va;Rψ∪ {rc
i}, Pψ)) <(1 −τth)d(op
i, dp
i)/2,(6)
Algorithm 4 SCPA Destination Insertion
for all rc
i∈Rnew
tdo
ψbest,u =None
vbest,u =None
for all v∈Vdo
ψ˜
k(v;Rψ∪ {rc
i}, Pψ) = insert(ψk(v;Rψ, Pψ), rc
i)
if φ(ψ˜
k(v;Rψ∪ {rc
i}, Pψ)) < φ(vbest,u)then
ψbest,u ←ψ˜
k(v;Rψ∪ {rc
i}, Pψ)
vbest,u ←v
end if
end for
ψbest =ψbest,u
vbest =vbest,u
rbest =None
for all v∈Vdo
ψ˜
k(v;Rψ∪ {rc
i}, Pψ) = insert(ψk(v;Rψ, Pψ), rc
i)
for all rp
i∈Pa
vdo
ψl(v;Rψ∪ {rc
i}, Pψ∪ {rp
i}) = insertDestination(ψ˜
k(v;Rψ∪ {rc
i}, Pψ), rp
i)
if d(ψl(v;Rψ∪ {rc
i}, Pψ∪ {rp
i})) −d(ψbest,u)<(1 −τth )d(op
i, dp
i)/2then
if φ(ψl(v;Rψ∪ {rc
i}, Pψ∪ {rp
i})) < φ(ψbest)then
ψbest ←ψl(v;Rψ, Pψ∪ {rp
i})
rbest ←rp
i
vbest ←v
end if
end if
end for
end for
if vbest 6=None then
assignSchedule(v, ψbest)
if rbest 6=None then
Pa
vbest ←Pa
vbest \ {rbest}
end if
end if
end for
12
Submitted 30. April 2022 3 METHODOLOGY
i.e. the driven distance of the best possible solution without parcel delivery is only increased by at most a
threshold factor compared to the direct distance of the inserted parcels. If multiple of these options exist, the
vehicle schedule minimizing the objective φis assigned. If none of these options exist, only the best schedule for
serving the customer is assigned. The corresponding logic is sketched in Algorithm 4. For reasons of clarity and
comprehensibility, not the computationally most efficient version of the algorithm is depicted in Algorithm 4. For
example, all solutions from the first customer insertion can be stored in a list, which makes the re-computation
of the insertion in the second loop over vehicles redundant.
3.3 Case Study for Munich, Germany
The proposed framework is applied and evaluated for a case study in Munich, Germany. The considered service
area of the MoD operator extends almost to the freeway ring, which surrounds the city center. The street
network G= (N, E)with edge travel times for each hour of a usual working day is extracted from a calibrated
microscopic traffic simulation described in [71]. The street network and operating area are shown in Figure 4.
Passenger demand for the MoD service is created by sampling from private vehicle trip OD-matrices, ex-
tracted from the same microscopic traffic simulation. 24 matrices for each hour of the day are available, which
contain around one million trips starting and ending within the service area. Poisson processes are used to sam-
ple requests for the MoD service with Poisson rates defined by the corresponding OD-entries times a penetration
factor. In this study, a penetration of 5% is applied, which can be interpreted as a MoD service replacing 5%
of the private vehicle trips within the operating area resulting in approximately 50,000 trips per day. Origin
and destination nodes for the sampled requests are matched randomly onto intersection nodes within associated
zones defined by the OD-matrices. Using different random seeds for the sampling process, three different sets
of requests are created and used for the simulations.
One month of real-world parcel shipment data of a local logistics provider was available to create parcel
demand for the RPP service. The data includes the date of delivery, the local depot the parcel has been
delivered from, and a destination address. In the first step, the destination address is converted into coordinates
using the open-source geo-coding API ’Nominatim’ relying on publicly available OpenStreetMap data. All
deliveries outside of the operating area of the MoD service are removed resulting in 56,000 parcels within the
operating area. Using the coordinates, parcel delivery destinations are matched onto the nearest intersection
node in the network. Maximally cp
vparcels with same origin-destination relation are aggregated into the same
parcel request. Because no information about the size of the parcels is given in the data, the size of a parcel
request is set by the number of aggregated parcels. All parcels are shipped from two depots, both positioned
outside of the MoD operating area (in the north and east of Munich). It is assumed that if a RPP service as
Figure 4: Street network, operating area and depots for the modelled RPP service in the case study of Munich,
Germany.
13
Submitted 30. April 2022 4 RESULTS
represented in this study is introduced, corresponding depots have to be implemented within the operating area.
Therefore, two new depots in the northern and eastern part of the city are defined and shown in Figure 4. Parcels
that are shipped from the original northern (eastern) depot are assumed that they have to be delivered from
the introduced inner-city northern (eastern) depot. One goal of the study is to observe the system boundaries,
i.e. which amount of parcel demand can be served with given passenger demand. Therefore, the parcel demand
data for the whole month is used as input for the simulations. To be able to vary the ratio between passenger
and parcel demand is sub-sampled to shares of 1% to 50% of the overall parcel demand. In the base scenario,
a 10% sub-sample of the parcel demand is used resulting in a ratio of approximately 1 parcel to 10 passenger
requests to represent a service with a priority on serving passenger demand. Analogously to the passenger
demand, three different sets of parcel requests are created using different random seeds within the sub-sampling
process.
Contrary to the simulations modeling the RPP service, the Status Quo is modeled with two independent
vehicle fleets serving as a baseline to evaluate the efficiency of integrating parcel delivery into the MoD service.
The first vehicle fleet corresponds to the MoD service without delivering any parcels. Hereby, the described
simulations are conducted without any parcel demand. The second fleet corresponds to the pure logistic service.
Hereby, vehicles are placed initially at each of the two depots. Parcels for the corresponding demand scenario
are inserted iteratively into their schedules, minimizing the driven distance including their return to the depot
at the end of the route. The aggregate driven distance within these schedules is used to approximate the fleet
vehicle kilometer by the logistic service.
Within the simulation, the service parameters describing customer maximum waiting time twait
max is set to
10 min while the maximum detour factor ∆is set to 40%. The fleet is operated with a capacity of cc
v= 4
passengers and cp
v= 8 parcels. In the Status Quo scenario, it is assumed that the logistics provider operates
trucks with a capacity of 100 parcels [72]. Re-balancing trips are performed every Trepo = 15 min and based on
forecasts according to the zones and values of the corresponding entries in the OD-matrices times the penetration
factor of 5%. The fleet size of the RPP operator for the following simulations is determined by first performing
multiple simulations without parcel demand and varying fleet size. Finally, a fleet size of 600 vehicles is chosen
that allows a service rate of approximately 90% served customers.
First simulations are performed for all parcel assignment strategies (CDPA, SCPA, SDPA) as well as for
Full RPP Integration and Moderate RPP Integration. Within these simulations, the influence of the assignment
threshold parameter τth on various service Key-Performance-Indicators (KPIs) is evaluated. In a second step,
simulations with varying parcel demand penetration ranging from 0% to 50% of the overall parcel demand data
set are performed while keeping τth constant to evaluate the number of parcels that can be accommodated by
the RPP service keeping the passenger demand fixed. Within the SDPA and SCPA strategy remaining on board
parcels are actively delivered after Tmax
p= 10pm.
4 Results
Results of the simulation from the presented case study for Munich, Germany are presented in the following.
In the upcoming two subsections influences on customer KPIs and operator KPIs are evaluated based on the
different proposed RPP service integration and parcel assignment strategies with a varying threshold parameter
τth. In the third subsection, effects resulting from different parcel demand levels are evaluated and compared
to the Status Quo.
4.1 Influence on Customer KPIs
A crucial question for the success of the proposed RPP service will be, whether the ride-pooling fleet can still
ensure sufficient service quality for passengers despite the additional transport of parcels. The attitude in this
paper is that logistics services are subordinated to passenger requests. In this context, for the scenario Moderate
RPP Integration the pick-up or delivery of parcels is only allowed while no passengers are on board, while this
constraint is lifted for the scenario Full RPP Integration. Nevertheless, time-constraints regarding passenger
pick-up and maximum detour, described in the beginning of this chapter, have to be fulfilled in either scenario
ensuring a certain quality of service.
The quality of the mobility service from the customer’s perspective is evaluated by average customer waiting
and travel times. The simulation results show, that waiting and travel times of the customers are rarely
influenced by the integration of freight transport into the MoD service. Figure 5a shows, that the customer’s
waiting times for all simulation scenarios and the different assignment thresholds (τth) are maximally increased
by 2%. Looking closer at the blue lines, which indicate the CDPA strategy for the full (solid line) and moderate
(dashed line) scenario, one can observe that increasing values for τth tend to result in lower waiting times, as
the allowed detour for pick-up and drop-off of parcels becomes smaller. In the cases of the SCPA and SDPA
strategies, the trend seems to be less compliant. The influences on customer’s travel times are shown in Figure 5b.
14
Submitted 30. April 2022 4 RESULTS
(a) Waiting Time (b) Travel Time
Figure 5: Impact of threshold parameter on waiting time and travel time.
It becomes apparent, that similar to the waiting times, the travel times are also only marginally influenced by the
additional transport of freight in the MoD system. Looking at the blue lines for the CDPA strategy the increase
in travel time is constantly between 0.1% and 0.5% for the investigated assignment thresholds (τth). Taking
into account the SCPA and SDPA strategies it becomes clear that especially for small assignment thresholds a
considerable decrease in travel times can be observed.
Figure 6a and Figure 6b show the absolute waiting and travel time distribution of the customers for the
different RPP scenarios and the assignment strategies. Generally, the customer assignment tends to insert new
passengers with waiting times close to the maximum allowed waiting time of 10 min. It once again becomes
clear, that the difference in waiting and travel times is relatively small comparing the Status Quo to the RPP
scenarios. This means that RPP has only little negative impact on the service quality of the represented MoD
provider. Looking at Figure 6c and Figure 6d the relative changes compared to the base scenario (Status Quo)
of the MoD ride-pooling service in terms of waiting and travel time become easier to identify. Especially, the
ratio of passengers experiencing low waiting times (up to 3 minutes), resulting from preceding parcel pick-ups
or drop-offs, tends to decrease. This is especially the case for the moderate integration because parcel insertion
can mainly be done within passenger approaches of the vehicles, while no customer is on board yet. The
decrease of passengers experiencing waiting times close to the maximum of 10 min might be an effect of longer
assigned schedules by the additional parcel service, decreasing the probability of finding a feasible insertion. For
the travel time distribution, a similar trend can be observed. Short and long travel times tend to be slightly
increased, whereas mid-range travel times (around 15 min) rather experience a small decrease. This indicates
the trend that parcels are mainly inserted in short to medium-long trips.
Looking closer at the difference between the Full RPP Integration and Moderate RPP Integration scenarios
one can observe that from the customers’ perspective, i.e. waiting and travel times, no big advantage is gained
by limiting the parcel pick-up and drop-off to times, when no customer is on board. The detours for parcel
pick-ups and drop-offs, which have to be accepted by the passengers, seem to get compensated by the additional
freedom within the creation of the vehicle schedules.
4.2 Influence on Operator KPIs
In addition to the attractiveness of a mobility service for the customer, the operator’s perspective plays a decisive
role for success. It is important for the provider of a MoD ride-pooling service, that the vehicle fleet can be
operated efficiently even after the logistics service has been integrated. Figure 7a shows the number of served
parcels given assignment thresholds (τth) between 0.6 and 1.0. One can observe that generally, the number of
served parcels declines with an increasing assignment threshold, resulting from more stringent pick-up or drop-
off constraints for a parcel in a given schedule. However, the CDPA reacts more sensitive to higher thresholds,
than the SDPA and SCPA strategies, which are both able to serve 100% of all parcels for all investigated τth
15
Submitted 30. April 2022 4 RESULTS
(a) Absolute Waiting Time
(b) Absolute Travel Time
(c) Waiting Time - Change to Status Quo
(d) Travel Time - Change to Status Quo
Figure 6: Waiting and travel time distribution of served customers within different RPP assignment strategies.
τth = 0.80 is considered in all scenarios shown. (fleet size = 600)
16
Submitted 30. April 2022 4 RESULTS
(a) Served Parcels (b) Served Customers
(c) Fleet KM (d) Fleet Utilization
Figure 7: Impact of threshold parameter on the number of served parcels, served persons, fleet kilometers
traveled and fleet utilization.
for the Full RPP Integration. Looking at Figure 7b one can see the number of served customers for varying
assignment thresholds. It shows that for all strategies, a notable decline in service rate for passengers cannot
be observed. The SCPA and SDPA strategies are even able to serve additional passengers compared to the
Status Quo, which can presumably be traced back to increased vehicle availability within the network due to
parcel pick-up and drop-off trips which can provide better coverage of the service area. Figure 7c displays the
overall fleet kilometers traveled throughout the whole day. One can observe that the comprehensive (CDPA)
strategy even produces a lower driven distance than the Status Quo without the integrated parcel delivery.
Compared to the fleet KM to serve only the parcels in the Status Quo (2,614 km on average), this equals
a reduction share of 48% in traveled distance for τth = 0.8. This means, that only 1,252 km compared to
the Status Quo were needed in the integrated transport approach. However, the slightly lower number of served
customers and parcels compared to the Status Quo have to be considered. The SCPA strategy and especially the
SDPA approach, however, lead to considerably higher driven distances compared to the Status Quo. Another
17
Submitted 30. April 2022 4 RESULTS
interesting aspect in Figure 7a is that the subsequent strategies (SDPA and SCPA) can accommodate nearly all
parcels with all thresholds for the Full RPP Integration. This means that all parcels are picked-up because even
with a high threshold the depots are very often located on the route of the vehicle fleet. Nevertheless, at the
end of the day the collected but not yet dropped-off parcels, are then delivered, which also contributes to the
significantly increased vehicle kilometers of the fleet using these strategies. Both, SDPA as well as SCPA, do
not consider the destination of parcels when picking them up, which can be a big disadvantage when delivering
them finally because parcel destinations might be distributed throughout the whole network. For the Moderate
RPP Integration the SDPA and SCPA approaches produces less fleet distance with higher τth values, which
corresponds to the falling numbers of served parcels. Figure 7d shows the fleet utilization, which of course leads
to similar results as Figure 7c. One can observe that depending on the chosen scenario and assignment strategy
the fleet utilization over the whole day varies between 54% and 61%.
The evaluations have shown that the Moderate RPP Integration performs consistently worse than the Full
RPP Integration: Significantly fewer parcels can be served while the experience for passengers does not deterio-
rate much applying the Full RPP Integration. Therefore, only full integration is considered for further analysis.
Additionally, the threshold parameter will be fixed to τth = 0.8, the value where a first notable decline in served
parcels can be observed for the CDPA strategy.
Figure 8 shows the temporal distribution of pick-ups and drop-offs for passengers and parcels for the different
assignment strategies. Most passenger demand starts in the morning at around 6am and lasts until the late
evening at around 8pm. It can be observed that parcel delivery only takes place, while there is passenger
demand to exploit the already existing passenger trips for parcel delivery. Strategies using off-peak deliveries
from other studies could easily be added on top, which would increase the capacity for parcel transport even
further.
For the comprehensive (CDPA) strategy the temporal distribution for pick-ups and drop-offs of parcels is
similar indicating a rather fast delivery after pick-up, resulting from the simultaneous assignment of pick-up
and drop-off. Therefore, the vehicle is actively routed toward drop-off locations. The subsequent (SDPA and
SCPA) strategies tend to collect the parcels right at the beginning of the day at around 7am. The peak in the
morning indicates, that it seems to be easy to find routes including the logistics depots. The SCPA strategy
shows a higher success rate in delivering parcels during the day compared to the SDPA strategy, as nearly all
parcels could be served and the number of drop-offs strongly declines after a peak around 11am. The SDPA
strategy, however, shows a strong peak in the number of drop-offs at 10pm, when the parcels which are still on
board the vehicles and could not be delivered so far are driven to their destinations.
Figure 8: Time-dependent pick-ups and drop-offs of parcels for different parcel assignment strategies and thresh-
old parameters τth = 0.8. Full integration is considered in all scenarios shown.
Figure 9 gives a detailed view of the temporal parcel occupancy states of the vehicles throughout the day.
If no parcels or passengers are on board, vehicles are represented by a black color. White color is used for idle
vehicles. It can be observed that for the CDPA strategy only around 50 of 600 vehicles have parcels on board
during the day, indicating a rather fast delivery once a parcel is picked up, as stated previously. For the SDPA
and SCPA strategies, most of the vehicles are filled with parcels in the morning, when they pass by the logistics
depots and carry them around during the day. In the case of SCPA most of the parcels can be delivered during
the day (Figure 8), which results in low occupation states at the end of the day. Looking at the SDPA strategy,
18
Submitted 30. April 2022 4 RESULTS
(a) CDPA (b) SDPA
(c) SCPA (d) Legend
Figure 9: Parcel occupancy states of moving vehicles for different parcel assignment strategies. τth = 0.80 and
full integration is considered in all scenarios shown. (fleet size = 600)
the occupation states at the end of the day are still high, which results in a delivery peak at around 10pm to
drop-off the remaining parcels. In general, one can observe that the SCPA and SDPA strategies result in higher
overall vehicle utilization, indicated by the area below the grey shape, compared to the CDPA strategy. This
reflects the fact of higher fleet kilometers and the higher number of served passengers and parcels shown in
Figure 7.
4.3 Variation of Logistics Demand
To investigate the system boundaries of the introduced RPP service, the logistics demand charged on the MoD
fleet was varied. The overall data set of 56,000 parcel shipments is sub-sampled to shares of 1 to 50%.
Figure 10a shows the total number of served parcels depending on the applied share of parcel demand. It
becomes apparent that all strategies reach a certain limit of parcel transport. The CDPA and SDPA strategies
match the Status Quo until a parcel demand of approximately 10%. The SCPA strategy is even able to serve a
parcel demand of up to 20% ( 11200 parcels per day). As already observed, the SDPA and the SCPA approaches
tend to pick up as many parcels as possible in the morning. Nevertheless, contrary to the SCPA strategy the
SDPA strategy is not able to deliver the majority of parcels throughout the day, leading to a fleet state of close
to full load, indicated by a horizontal plateau in Figure 10a. CDPA and SCPA on the other hand can also
deliver most of the parcels during the service freeing up additional capacity to serve further parcels.
19
Submitted 30. April 2022 4 RESULTS
(a) Served Parcels (b) Served Customers
(c) Fleet KM (d) Fleet KM Per Served Customer and Parcel
Figure 10: Impact of varying parcel demand on the number of served parcels, served persons and fleet kilometers
traveled. In all simulations τth = 0.8and full integration is considered.
Figure 10b displays the relative number of served customers. Similar effects that have already been evaluated
for the sensitivity of parameter τth in Figure 7b can be observed: For any strategy, no notable drop in served
customers can be observed, even for high parcel demand penetrations showing the potential of free capacity
for a parcel service during the MoD operation. The higher fleet utilization seems to lead to a higher vehicle
availability, resulting in an even slightly higher share of served customers (at least when using the assignment
and re-balancing control strategies applied in this study).
When it comes to the traveled distance of the vehicle fleet in Figure 10c, it can be seen that only the CDPA
strategy results in similar traveled distances compared to the Status Quo. The CDPA strategy actually even
stays below the Status Quo, saving driving distance through the integrated transport of passenger and freight
streams. The SCPA and SDPA tend to increase fleet kilometers heavily and produce around 15,000 km and
30,000 km more total distance when considering high parcel demand penetration rates. Until the number of
served parcels stabilizes at around 20% penetration, Especially, SDPA shows a strong increase in the driven
20
Submitted 30. April 2022 5 CONCLUSION
distance indicating a bad scaling behavior for finding effective routes for parcel pick-up and delivery. These
findings are also in line with the insights and discussion presented in Figure 7c. However, similar scaling of
fleet-driven distance with parcel penetration can be observed for the CDPA strategy. In this regime, a very
low driven distance can be observed, especially for high parcel penetration rates (even lower than the Status
Quo). The down point of this strategy is that significantly fewer parcels are served in this regime. Looking
at Figure 10d one can observe the fleet kilometers per served customer and parcel in relation to the parcel
demand penetration. This quantity also takes the number of served customers and parcels into account when
comparing fleet kilometers. Thereby, ’better’ distance to request ratios can be achieved by either transporting
more passengers or parcels or both, while keeping the driven distance low. All strategies show lower traveled
distance per served request with rising parcel demand, indicating a better integration. However, with high
parcel penetration efficient routes for parcel delivery can be found while all parcels are served in the Status
Quo, leading to the lowest values for fleet kilometers per served customer and parcel in this scenario. These
results indicate that the modeled integration of parcel delivery into the MoD service is only reasonable for parcel
penetration rates of around 10%.
5 Conclusion
5.1 Summary
This paper investigates the integrated transport of passengers and freight, assuming a mobility-on-demand ride-
pooling service, called Ride-Parcel-Pooling (RPP). Thereby, the operator of the vehicle fleet combines parcel
delivery into the passenger routes of the MoD service. This study uses two different integration approaches,
which are compared to the Status Quo consisting of a separate logistics fleet and the MoD service serving
passengers only. The Moderate RPP Integration on the one hand, where parcels are only picked-up or dropped-
off when no passengers are on board, is not very beneficial, as it results in lower numbers of served parcels
and does not show better results in customer service quality compared to the full integration. The Full RPP
Integration scenario on the other hand, shows similar results in passenger waiting and detour times as the
previous scenario and performs significantly better in the number of served parcels. The assignment of parcels
is investigated using three different heuristic parcel assignment strategies (CDPA, SCPA and SDPA). Each of
these strategies aims at inserting parcels into the schedules serving passengers with small detour for parcel
pick-up and delivery. Because no explicit time constraints on parcel pick-up and delivery are employed, the
assignment of parcel deliveries is separated from the parcel pick-ups in the SCPA and SDPA strategy contrary to
the CDPA strategy. The results of a simulation case study for Munich, Germany suggest that the MoD service
is not deteriorated by the integration of logistics services and is able to serve nearly all parcels until a parcel-
to-passenger ratio of at least 1:10. The SCPA and SDPA strategies are able to transport more parcels than
CDPA and can even achieve an increase of served passengers compared to the Status Quo, however they result
in significantly higher driven distance. The CDPA approach, however can decrease driven distances compared
to the Status Quo, where two vehicle fleets serve passenger and logistics demand independently.
5.2 Discussion
The presented RPP service offers a good opportunity to increase the utilization of MoD ride-pooling vehicle
fleets. It could, assuming the existence of a MoD service, already today complement existing logistics services and
extend the best practices in the sector. Thereby, it especially offers a solution to use cases, where conventional
logistics approaches can not create any bundling effects, i.e. combining multiple parcels in one logistics vehicle.
Furthermore, RPP has the potential to reduce the number of vehicles on the urban street network, by integrating
passenger and logistics flows in one fleet of vehicles.
This paper examines a forward logistics use case, meaning that the parcels are transported from a depot
to the recipient. In reality, this is the most relevant, but not the only logistics form. Reverse logistics from
the sender to a consolidation center and courier services between two individuals could be interesting for future
analysis. In the case of integrating courier services into the MoD ride-pooling service, a relatively cost-effective
local distribution system could be built up. Especially, considering developments in autonomous driving, the
integration into autonomous mobility-on-demand (AMoD) services could further decrease operational costs, by
eliminating the need to pay a driver. This decrease in cost might further lead to decreased fares and therefore
amplified demand, enhancing scaling properties of ride-pooling services that will also translate to more efficient
options for parcel delivery integration. In the future, one could even imagine multi-purpose autonomous vehicles,
which are able to convert passenger space into logistics space and vice versa. Such a vehicle concept could further
promote the applicability of an integrated passenger and freight MoD service. A crucial aspect here is the time
taken to load and unload the parcels. This study assumed that both loading and unloading are linked to a finely
distributed parcel locker station system or that the senders and recipients take over or hand over the parcels
at the vehicle, resulting in a relatively small boarding time of one minute. In reality, significantly higher times
21
Submitted 30. April 2022 5 CONCLUSION
could arise, especially at the current stage without a high number of parcel lockers, which could deteriorate
passenger satisfaction in particular. Overall, RPP might be a chance for local stores to compete with online
shops, by offering fast delivery and a sustainable way of shopping. If the service can succeed in reducing the
overall kilometers traveled in a city, by pooling passengers and parcels, this could have substantial benefits for
the city and its residents. Not only it could induce a more livable city environment, by reducing the number of
cars and traffic load in cities, but lead to a more sustainable way of city transportation without compromising
on the customer experience, which MoD services offer.
5.3 Future Work
Generally, one can observe that the integration of a secondary parcel demand into the existing schemes of a MoD
ride-pooling system is possible. The CDPA approach already offers promising results, although this heuristic
approach is not of great complexity. The SCPA and SDPA strategies do still offer lots of room for improvement.
In theory, at least a similar fleet performance should be achievable for a separated parcel pick-up and delivery
assignment within the SCPA and SDPA strategies compared to the CDPA strategy. Several improvements can
thereby be tested in the future for these strategies: Firstly, the decision of an assignment in the SCPA and
SDPA strategies is made based on a threshold detour controlled by the parameter τth which is the same for the
pick-up and the drop-off. Results showed that it seemed to be more probable to assign the pick-up compared
to the delivery of a parcel. An asymmetric assignment threshold for pick-up and delivery could thus amplify
efficiency. Secondly, the assignment of a parcel origin and therefore the delivery of a parcel by a certain vehicle
does not consider the destination of parcels that are already assigned. It could be beneficial to assign parcels
with similar destinations to the same vehicle. And thirdly, the delivery of parcels could be actively integrated
into the re-balancing algorithm, i.e. if a lack of vehicle supply is recognized in a certain zone, vehicles with parcel
deliveries in the corresponding zone could be prioritized for the corresponding re-balancing trip. Furthermore,
future research could investigate the influences of a variation in loading and unloading times for parcels for
finding the influence on customer waiting and travel times and by that customer satisfaction. Nevertheless, the
overall goal should be the development of an integrated control algorithm that optimizes vehicle schedules and
decides on parcel and passenger assignment in a unified optimization framework.
Apart from that, the presented RPP service could be extended by integrating vehicle idle times into the
availability for logistics services. This could increase the system capacity for logistics services strongly and
already developed efficient algorithms for planning routes for pick-up and delivery could be exploited. Never-
theless, care has to be taken on when and how many vehicles perform these pure logistics services because these
vehicles will no longer be available for passenger transport. Additionally, a further increase in vehicle kilometers
traveled is expected.
Looking at the RPP service definition, one could think of multiple extensions. The current service inten-
tionally did not impose pickup and delivery time windows for parcels. However, this representation of parcels
excludes certain types of logistics services that could also be use cases for RPP, for example, food and grocery,
drugs and medicine, or high priority parcels. Furthermore, the effects of different logistics service forms (e.g.
same-day vs. next-day delivery or forward vs. reverse vs. courier logistics) on the RPP performance could
be investigated. It would be particularly interesting to look at edge cases that are currently financially not
interesting for logistics companies due to low bundling effects and to investigate whether RPP could remedy
this situation. Last but not least, constraints for the design of such a multi-purpose vehicle could be examined.
Depending on the use case the optimal parcel capacity for a RPP vehicle could be determined in future studies.
Additionally, a dynamic allocation of capacity based on the current demand to convert seats into parcel storage
and vice-versa could be implemented.
Conflict of Interest Statement
The authors declare that they have no known competing financial interests, or personal relationships that could
have appeared to influence the work reported in this paper.
Author Contributions
Study conception and design: FF, RE, FD, KB, FB; data collection: FF, RE; analysis and interpretation of
results: FF, RE, FD; simulation model: RE, FD, FF; draft manuscript preparation: FF, RE, FD, KB, FB. All
authors reviewed the results and approved the final version of the manuscript.
22
Submitted 30. April 2022 REFERENCES
Funding
The German Federal Ministry of Transport and Digital Infrastructure provides funding through the project
“TEMPUS” with grant number 01MM20008K. The authors remain responsible for all findings and opinions
presented in the paper.
References
[1] European Commission. COMMISSION STAFF WORKING DOCUMENT - A call to action on urban
logistics. Dec. 2013. url:https://op.europa.eu/en/publication-detail/-/publication/ffac1877-
67ca-11e3-a7e4-01aa75ed71a1.
[2] Pricewaterhouse Coopers. A time for hope: Consumers’ outlook brightens despite headwinds - December
2021 Global Consumer Insights Pulse Survey. 2021. url:https: / /www. pwc . com /consumerinsights
(visited on 04/22/2022).
[3] Claudia Nobis, Tobias Kuhnimhof, Robert Follmer, and Marcus Bäumer. Mobilität in Deutschland -
Zeitreihenbericht 2002-2008-2017. Dec. 2019. url:https://www.bmvi.de/SharedDocs/DE/Anlage/G/
mid-zeitreihenbericht-2002-2008-2017.pdf?__blob=publicationFile.
[4] Bundesverband Paket und Expresslogistik e. V. (BIEK). KEP-Studie 2021 – Analyse des Marktes in
Deutschland. June 2021. url:https://www. biek . de / files / biek / downloads / papiere / BIEK _KEP-
Studie_2021.pdf.
[5] Lay Maxwell G. Ways of the world : a history of the world’s roads and of the vehicles that used them. New
Brunswick, N.J.: Rutgers University Press, 1992, pp. 125–126. isbn: 0813526914. url:https: //trid.
trb.org/view/368373.
[6] Bob Pishue. 2021 INRIX Global Traffic Scorecard. Dec. 2021. url:https://inrix.com/scorecard/.
[7] Federal Ministry for the Environment, Nature Conservation, Nuclear Safety and Consumer Protection
(BMUV). Umweltbelastungen durch Verkehr. 2022. url:https:// www . umweltbundesamt . de /daten/
verkehr/umweltbelastungen-durch-verkehr#undefined (visited on 04/22/2022).
[8] Zhujun Li, Amer Shalaby, Matthew J. Roorda, and Baohua Mao. “Urban rail service design for collabo-
rative passenger and freight transport”. In: Transportation Research Part E: Logistics and Transportation
Review 147 (2021), p. 102205. issn: 1366-5545. doi:https://doi.org/10.1016/j.tre.2020.102205.
url:https://www.sciencedirect.com/science/article/pii/S1366554520308474.
[9] Emilio Larrodé and Victoria Muerza. “Chapter 4 - Improving cost efficiency and environmental impact
through the integration of light freight and passenger railway transport and last-mile distribution analysis”.
In: Urban Freight Transportation Systems. Ed. by Ralf Elbert, Christian Friedrich, Manfred Boltze, and
Hans-Christian Pfohl. Elsevier, 2020, pp. 55–78. isbn: 978-0-12-817362-6. doi:https: / / doi.org/10 .
1016/B978-0- 12-817362-6.00004- 5.url:https://www.sciencedirect.com/science/article/pii/
B9780128173626000045.
[10] Walid Behiri, Sana Belmokhtar-Berraf, and Chengbin Chu. “Urban freight transport using passenger rail
network: Scientific issues and quantitative analysis”. In: Transportation Research Part E: Logistics and
Transportation Review 115 (2018), pp. 227–245. issn: 1366-5545. doi:https://doi.org/10.1016/j.tre.
2018.05.002.url:https://www.sciencedirect.com/science/article/pii/S136655451730162X.
[11] Yiming He, Qiang Wang, Hailong Chen, and Chuanqi Gao. “Metro Passenger and Freight Transport: A
Framework for Underground Logistics System”. In: 2019 IEEE Intl. Conf. on Dependable, Autonomic and
Secure Computing, Intl. Con.f on Pervasive Intelligence and Computing, Intl. Conf. on Cloud and Big Data
Computing, Intl. Conf. on Cyber Science and Technology Congress (DASC/PiCom/CBDCom/CyberSciTech).
2019, pp. 880–882. doi:10.1109/DASC/PiCom/CBDCom/CyberSciTech.2019.00160.
[12] Fangting Zhou and Jin Zhang. “Freight Transport Mode Based on Public Transport: Taking Parcel Deliv-
ery by Subway as an Example”. In: ICTE 2019, pp. 745–754. doi:10.1061/9780784482742.083. eprint:
https://ascelibrary.org/doi/pdf/10.1061/9780784482742.08.url:https://ascelibrary.org/
doi/abs/10.1061/9780784482742.083.
[13] Alexandra G. Dampier and Marin Varbanov Marinov. “A Study of the Feasibility and Potential Imple-
mentation of Metro-Based Freight Transportation in Newcastle upon Tyne”. In: Urban Rail Transit 1
(2015), pp. 164–182.
[14] Keith Cochrane, Shoshanna Saxe, Matthew Roorda, and Amer Shalaby. “Moving Freight on Public Tran-
sit: Best Practices, Challenges, and Opportunities”. In: International Journal of Sustainable Transportation
11 (June 2016), pp. 00–00. doi:10.1080/15568318.2016.1197349.
23
Submitted 30. April 2022 REFERENCES
[15] Ezzeddine Fatnassi, Chaouachi Jouhaina, and Klibi Walid. “Towards Creating Sustainable Urban Areas:
A Shared Passengers and Goods On-Demand Rapid Transit Proposal”. In: Proceedings of 1st International
Physical Internet Conference. Québec, Canada, 2014. url:https://www .cirrelt.ca/ipic2014/pdf/
1024a.pdf.
[16] Katrien De Langhe, Hilde Meersman, Christa Sys, Eddy Voorde, and Thierry Vanelslander. “How to
make urban freight transport by tram successful?” In: Journal of Shipping and Trade 4 (Dec. 2019). doi:
10.1186/s41072-019-0055-4.
[17] City of Zurich. Cargo-Tram and E-Tram. 2022. url:https://www.stadt-zuerich.ch/ted/de/index/
entsorgung_recycling/ entsorgen/wo- wann- entsorgen/cargo- tram_und_e- tram .html (visited on
02/22/2022).
[18] Martin Kagerbauer, Lukas Barthelmes, Anna Reiffer, and Emre Görgülü. LogIKTram - Logistics Concept
and ICT Platform for Urban Rail-based Freight Transport. 2022. url:https: / / www . ifv . kit . edu /
english/26_1542.php (visited on 02/16/2022).
[19] Kai-Oliver Schocke, Petra K. Schäfer, Silke Höhl, Andreas Gilbert, Felix Kwast, Simon Lacoste, Simon
Marhan, and Klaus-Peter Wenz. “Bericht zum Forschungsvorhaben LastMileTram - Empirische Forschung
zum Einsatz einer Güterstraßenbahn am Beispiel Frankfurt am Main”. In: Research Lab for Urban Trans-
port, Frankfurt University of Applied Sciences (2020). url:https://www.frankfurt- university.de/
fileadmin/ standard /Hochschule/Fachbereich_ 1 /FFin/Neue_ Mobilitaet /Veroeffentlichungen/
2020/Abschlussbericht_LastMileTram.pdf.
[20] Martin Pernkopf and Manfred Gronalt. “An aerial ropeway transportation system for combined freight
and passenger transport – a simulation study”. In: Transportation Planning and Technology 44.1 (2021),
pp. 45–62. doi:10.1080/03081060.2020.1851450. eprint: https://doi.org/10.1080/03081060.2020.
1851450.url:https://doi.org/10.1080/03081060.2020.1851450.
[21] Francesco Bruzzone, Federico Cavallaro, and Silvio Nocera. “The integration of passenger and freight
transport for first-last mile operations”. In: Transport Policy 100.C (2021), pp. 31–48. doi:10.1016/j.
tranpol.2020.10.
[22] Veaceslav Ghilas, Emrah Demir, and Tom Van Woensel. “The pickup and delivery problem with time
windows and scheduled lines”. In: INFOR: Information Systems and Operational Research 54.2 (2016),
pp. 147–167. doi:10.1080/03155986.2016.1166793. eprint: https://doi.org/10. 1080/03155986 .
2016.1166793.url:https://doi.org/10.1080/03155986.2016.1166793.
[23] Niklas Arvidsson, Moshe Givoni, and Johan Woxenius. “Exploring Last Mile Synergies in Passenger and
Freight Transport”. In: Built Environment 42 (Dec. 2016), pp. 523–538. doi:10.2148/benv.42.4.523.
[24] Anna Trentini, Renaud Masson, Fabien Lehuédé, Nicolas Malhéné, Olivier Péton, and Houda Tlahig.
“A shared passengers & goods city logistics system”. In: 4th International Conference on Information
Systems, Logistics and Supply Chain. Quebec, Canada, Aug. 2012, 10p. url:https:// hal.archives-
ouvertes.fr/hal-00861728.
[25] Hugh Frost. Freight*Bus, The bus that Delivers! 2008. url:http://www.onroutebus.co.uk/ (visited on
02/22/2022).
[26] Toyota. ePalette Concept. 2018. url:https://www.toyota-global.com/pages/contents/innovation/
intelligent_transport_systems /world_ congress/2018copenhagen /pdf/ e- Palette_CONCEPT .pdf
(visited on 02/16/2022).
[27] Aliaa Alnaggar, Fatma Gzara, and James H. Bookbinder. “Crowdsourced Delivery: A Review of Platforms
and Academic Literature”. In: Omega, Elsevier 98.C (2021). doi:10.1016/j.omega.2019.102139.url:
https://ideas.repec.org/a/eee/jomega/v98y2021ics030504831930578x.html.
[28] Afonso Sampaio, Martin Savelsbergh, Lucas Veelenturf, and Tom van Woensel. “Chapter 15 - Crowd-
Based City Logistics”. In: Sustainable Transportation and Smart Logistics. Ed. by Javier Faulin, Scott E.
Grasman, Angel A. Juan, and Patrick Hirsch. Elsevier, 2019, pp. 381–400. isbn: 978-0-12-814242-4. doi:
https://doi.org/10.1016/B978- 0- 12- 814242- 4.00015-6.url:https://www.sciencedirect.com/
science/article/pii/B9780128142424000156.
[29] Chengxi Liu, Qian Wang, and Yusak O. Susilo. “Assessing the impacts of collection-delivery points to
individual’s activity-travel patterns: A greener last mile alternative?” In: Transportation Research Part
E: Logistics and Transportation Review 121 (2019). Special Issue on Supply chain elements in freight
transport modelling; Guest Edited by Lóránt Tavasszy and Hanno Friedrich, pp. 84–99. issn: 1366-5545.
doi:https: / / doi . org /10 . 1016 / j . tre. 2017 . 08 . 007.url:https :/ / www . sciencedirect . com/
science/article/pii/S1366554517301886.
24
Submitted 30. April 2022 REFERENCES
[30] Chao Chen and Shenle Pan. “Using the Crowd of Taxis to Last Mile Delivery in E- commerce: a Method-
ological Research”. In: Proceedings of SOHOMA. Nov. 2015.
[31] Baoxiang Li, Dmitry Krushinsky, Hajo A. Reijers, and Tom Van Woensel. “The Share-a-Ride Problem:
People and parcels sharing taxis”. In: European Journal of Operational Research 238.1 (2014), pp. 31–
40. issn: 0377-2217. doi:https: / / doi . org / 10 . 1016/ j . ejor . 2014 . 03 . 003.url:https: / / www .
sciencedirect.com/science/article/pii/S0377221714002173.
[32] Tilmann Schlenther, Kai Martins-Turner, Joschka Felix Bischoff, and Kai Nagel. “Potential of Private
Autonomous Vehicles for Parcel Delivery”. In: Transportation Research Record 2674.11 (2020), pp. 520–
531. doi:10. 1177 / 0361198120949878. eprint: https:/ / doi.org/10 . 1177 /0361198120949878.url:
https://doi.org/10.1177/0361198120949878.
[33] John Howe. “’Filling the middle’: Uganda’s appropriate transport services”. In: Transport Reviews 23 (Apr.
2003), pp. 161–176. doi:10.1080/01441640309890.
[34] Daniel J. Fagnant, Kara. Kockelman, and Prateek Bansal. “Operations of Shared Autonomous Vehicle
Fleet for Austin, Texas Market”. In: Transportation Research Record: Journal of the Transportation Re-
search Board 2536 (2015), pp. 98–106. issn: 0361-1981. doi:10.3141/2536-12.
[35] Florian Dandl and Klaus Bogenberger. “Comparing Future Autonomous Electric Taxis with an Existing
Free-Floating Carsharing System”. In: IEEE Transactions on Intelligent Transportation Systems 20.6
(2019), pp. 2037–2047. issn: 1524-9050. doi:10.1109/TITS.2018.2857208.
[36] Roman Engelhardt, Florian Dandl, Aledia Bilali, and Klaus Bogenberger. “Quantifying the Benefits of Au-
tonomous On-Demand Ride-Pooling: A Simulation Study for Munich, Germany”. In: 2019 IEEE Intelligent
Transportation Systems Conference (ITSC). 2019, pp. 2992–2997. doi:10.1109/ITSC.2019.8916955.
[37] Claudio Ruch, ChengQi Lu, Lukas Sieber, and Emilio Frazzoli. “Quantifying the Efficiency of Ride Shar-
ing”. In: IEEE Transactions on Intelligent Transportation Systems (2020), pp. 1–6. issn: 1524-9050. doi:
10.1109/TITS.2020.2990202.
[38] Remi Tachet, Oleguer Sagarra, Paolo Santi, G. Resta, Michael Szell, Steven H. Strogatz, and Carlo Ratti.
“Scaling Law of Urban Ride Sharing”. In: Scientific reports 7 (2017), p. 42868. doi:10.1038/srep42868.
[39] Aledia Bilali, Roman Engelhardt, Florian Dandl, Ulrich Fastenrath, and Klaus Bogenberger. “Analytical
and Agent-Based Model to Evaluate Ride-Pooling Impact Factors”. In: Transportation Research Record:
Journal of the Transportation Research Board (2020), p. 036119812091766. issn: 0361-1981. doi:10 .
1177/0361198120917666.
[40] Florian Dandl, Michael Hyland, Klaus Bogenberger, and Hani S. Mahmassani. “Evaluating the impact
of spatio-temporal demand forecast aggregation on the operational performance of shared autonomous
mobility fleets”. In: Transportation 46.6 (2019), pp. 1975–1996. issn: 0049-4488. doi:10.1007/s11116-
019-10007-9.
[41] Arslan Ali Syed, Florian Dandl, Bernd Kaltenhäuser, and Klaus Bogenberger. “Density Based Distribution
Model for Repositioning Strategies of Ride Hailing Services”. In: Frontiers in Future Transportation 2
(2021), p. 295. doi:10.3389/ffutr.2021.681451.
[42] Paolo Santi, Giovanni Resta, Michael Szell, Stanislav Sobolevsky, Steven H. Strogatz, and Carlo Ratti.
“Quantifying the benefits of vehicle pooling with shareability networks”. In: Proceedings of the National
Academy of Sciences of the United States of America 111.37 (2014), pp. 13290–13294. issn: 0027-8424.
doi:10.1073/pnas.1403657111.
[43] Javier Alonso-Mora, Samitha Samaranayake, Alex Wallar, Emilio Frazzoli, and Daniela Rus. “On-demand
high-capacity ride-sharing via dynamic trip-vehicle assignment”. In: Proceedings of the National Academy
of Sciences 114.3 (2017), pp. 462–467. issn: 0027-8424. doi:10.1073/pnas.1611675114. eprint: https:
//www.pnas.org/content/114/3/462.full.pdf.url:https://www.pnas.org/content/114/3/462.
[44] Andrea Simonetto, Julien Monteil, and Claudio Gambella. “Real-time city-scale ridesharing via linear
assignment problems”. In: Transportation Research Part C: Emerging Technologies 101 (2019), pp. 208–
232. issn: 0968090X. doi:10.1016/j.trc.2019.01.019.
[45] Burak Eksioglu, Arif Volkan Vural, and Arnold Reisman. “The vehicle routing problem: A taxonomic
review”. In: Computers & Industrial Engineering 57.4 (2009), pp. 1472–1483. issn: 0360-8352. doi:https:
//doi.org/10.1016/j.cie.2009.05.009.url:https://www.sciencedirect.com/science/article/
pii/S0360835209001405.
[46] Gitae Kim, Yew-Soon Ong, Chen Kim Heng, Puay Siew Tan, and Nengsheng Allan Zhang. “City Vehicle
Routing Problem (City VRP): A Review”. In: IEEE Transactions on Intelligent Transportation Systems
16.4 (2015), pp. 1654–1666. doi:10.1109/TITS.2015.2395536.
25
Submitted 30. April 2022 REFERENCES
[47] Federico Cavallaro and Silvio Nocera. “Integration of passenger and freight transport: A concept-centric
literature review”. In: Research in Transportation Business I& Management (2021), p. 100718. issn: 2210-
5395. doi:https://doi.org/10.1016/j.rtbm.2021.100718.url:https://www.sciencedirect.com/
science/article/pii/S2210539521001012.
[48] Abood Mourad, Jakob Puchinger, and Chengbin Chu. “A survey of models and algorithms for optimizing
shared mobility”. In: Transportation Research Part B: Methodological 123 (2019), pp. 323–346. issn: 0191-
2615. doi:https://doi.org/10.1016/j.trb.2019.02.003.url:https://www.sciencedirect.com/
science/article/pii/S0191261518304776.
[49] Le Li, Rudy R. Negenborn, and Bart De Schutter. “Multi-agent cooperative transport planning of inter-
modal freight transport”. In: 17th International IEEE Conference on Intelligent Transportation Systems
(ITSC). 2014, pp. 2465–2471. doi:10.1109/ITSC.2014.6958085.
[50] Ngoc-Quang Nguyen, Nguyen-Viet-Dung Nghiem, Phan-Thuan DO, Khac-Tuan LE, Minh-Son Nguyen,
and Naoto MUKAI. “People and Parcels Sharing a Taxi for Tokyo City”. In: Proceedings of the Sixth
International Symposium on Information and Communication Technology. SoICT 2015. Hue City, Viet
Nam: Association for Computing Machinery, 2015, pp. 90–97. isbn: 9781450338431. doi:10 . 1145 /
2833258.2833309.url:https://doi.org/10.1145/2833258.2833309.
[51] Nicole Ronald, Jie Yang, and Russell G. Thompson. “Exploring Co-Modality Using On-Demand Trans-
port Systems”. In: Transportation Research Procedia 12 (2016). Tenth International Conference on City
Logistics 17-19 June 2015, Tenerife, Spain, pp. 203–212. issn: 2352-1465. doi:https:/ / doi .org/10.
1016 / j . trpro . 2016 . 02 . 059.url:https : / / www . sciencedirect . com / science / article / pii /
S2352146516000600.
[52] David Soto Setzke, Christoph Pflügler, Maximilian Schreieck, Sven Fröhlich, Manuel Wiesche, and Hel-
mut Krcmar. “Matching Drivers and Transportation Requests in Crowdsourced Delivery Systems”. In:
Proceedings of the Twenty-third Americas Conference on Information Systems. Oct. 2017.
[53] Chao Chen, Daqing Zhang, Xiaojuan Ma, Bin Guo, Leye Wang, Yasha Wang, and Edwin Sha. “crowdde-
liver: Planning City-Wide Package Delivery Paths Leveraging the Crowd of Taxis”. In: IEEE Transactions
on Intelligent Transportation Systems 18.6 (2017), pp. 1478–1496. doi:10.1109/TITS.2016.2607458.
[54] Nabin Kafle, Bo Zou, and Jane Lin. “Design and modeling of a crowdsource-enabled system for urban
parcel relay and delivery”. In: Transportation Research Part B: Methodological 99 (2017), pp. 62–82. issn:
0191-2615. doi:https://doi.org/10.1016/j.trb.2016.12.022.url:https://www.sciencedirect.
com/science/article/pii/S019126151630265X.
[55] Breno A. Beirigo, Frederik Schulte, and Rudy R. Negenborn. “Integrating People and Freight Transporta-
tion Using Shared Autonomous Vehicles with Compartments”. In: IFAC-PapersOnLine 51.9 (2018). 15th
IFAC Symposium on Control in Transportation Systems CTS 2018, pp. 392–397. issn: 2405-8963. doi:
https : / / doi . org / 10 . 1016 / j . ifacol . 2018 . 07 . 064.url:https : / / www . sciencedirect . com /
science/article/pii/S2405896318307870.
[56] Wei Qi, Lefei Li, Sheng Liu, and Zuo-Jun Max Shen. “Shared Mobility for Last-Mile Delivery: Design,
Operational Prescriptions, and Environmental Impact”. In: Manufacturing & Service Operations Manage-
ment 20.4 (2018), pp. 737–751. doi:10. 1287/msom. 2017 .0683. eprint: https://doi . org/10. 1287/
msom.2017.0683.url:https://doi.org/10.1287/msom.2017.0683.
[57] Fangxin Wang, Yifei Zhu, Feng Wang, and Jiangchuan Liu. “Ridesharing as a Service: Exploring Crowd-
sourced Connected Vehicle Information for Intelligent Package Delivery”. In: 2018 IEEE/ACM 26th Inter-
national Symposium on Quality of Service (IWQoS). 2018, pp. 1–10. doi:10.1109/IWQoS.2018.8624152.
[58] Alp M. Arslan, Niels Agatz, Leo Kroon, and Rob Zuidwijk. “Crowdsourced Delivery—A Dynamic Pickup
and Delivery Problem with Ad Hoc Drivers”. In: Transportation Science 53.1 (2018), pp. 222–235. doi:
10 . 1287 / trsc . 2017 . 0803. eprint: https : / / doi . org / 10 . 1287 / trsc . 2017 . 0803.url:https :
//doi.org/10.1287/trsc.2017.0803.
[59] Abood Mourad. “The synchronization of shared mobility flows in urban environments”. Theses. Université
Paris Saclay (COmUE), June 2019. url:https://tel.archives-ouvertes.fr/tel-02286031.
[60] Shirin Najaf Abadi. “ Designing an On-Demand Dynamic Crowdshipping Model and Evaluating its Ability
to Serve Local Retail Delivery in New York City”. PhD thesis. CUNY Academic Works, 2019. url:https:
//academicworks.cuny.edu/cc_etds_theses/809.
[61] Yueyue Chen, Deke Guo, Ming Xu, Guoming Tang, Tongqing Zhou, and Bangbang Ren. “PPtaxi: Non-
Stop Package Delivery via Multi-Hop Ridesharing”. In: IEEE Transactions on Mobile Computing 19.11
(2020), pp. 2684–2698. doi:10.1109/TMC.2019.2928808.
26
Submitted 30. April 2022 REFERENCES
[62] Kaushik Manchella, Abhishek Umrawal, and Vaneet Aggarwal. “FlexPool: A Distributed Model-Free Deep
Reinforcement Learning Algorithm for Joint Passengers I& Goods Transportation”. In: IEEE Transactions
on Intelligent Transportation Systems (July 2020). doi:https://doi.org/10.48550/arXiv.2007.13699.
[63] Max van der Tholen, Breno A. Beirigo, Jovana Jovanova, and Frederik Schulte. “The Share-A-Ride Prob-
lem with Integrated Routing and Design Decisions: The Case of Mixed-Purpose Shared Autonomous
Vehicles”. In: Computational Logistics. Ed. by Martijn Mes, Eduardo Lalla-Ruiz, and Stefan Voß. Springer
International Publishing, 2021, pp. 347–361.
[64] André Romano Alho, Takanori Sakai, Simon Oh, Cheng Cheng, Ravi Seshadri, Wen Han Chong, Yusuke
Hara, Julia Caravias, Lynette Cheah, and Moshe Ben-Akiva. “A simulation-based evaluation of a Cargo-
Hitching service for E-commerce using mobility-on-demand vehicles”. In: Future Transportation 1.3 (2021),
pp. 639–656.
[65] Fabian Fehn, Roman Engelhardt, and Klaus Bogenberger. “Ride-Parcel-Pooling - Assessment of the Poten-
tial in Combining On-Demand Mobility and City Logistics”. In: 2021 IEEE International Intelligent Trans-
portation Systems Conference (ITSC). 2021, pp. 3366–3372. doi:10.1109/ITSC48978.2021.9564630.
[66] Simon Meinhardt, Tilmann Schlenther, Kai Martins-Turner, and Michal Maciejewski. “Simulation of On-
Demand Vehicles that Serve both Person and Freight Transport”. In: Procedia Computer Science 201
(2022). The 13th International Conference on Ambient Systems, Networks and Technologies (ANT) /
The 5th International Conference on Emerging Data and Industry 4.0 (EDI40), pp. 398–405. issn: 1877-
0509. doi:https: //doi .org/10 .1016/ j.procs.2022.03 .053.url:https: //www. sciencedirect.
com/science/article/pii/S1877050922004665.
[67] Shiyao Zhang, Christos Markos, and James J.Q. Yu. “Autonomous Vehicle Intelligent System: Joint Ride-
Sharing and Parcel Delivery Strategy”. In: IEEE Transactions on Intelligent Transportation Systems (Jan.
2022), pp. 1–12. doi:10.1109/TITS.2022.3162609.
[68] TUM - Chair of Traffic Engineering and Control. GitHub Repository for TUM-VT FleetPy Simulation
Framework. 2022. url:https://github.com/TUM-VT/FleetPy (visited on 02/22/2022).
[69] Roman Engelhardt, Florian Dandl, and Klaus Bogenberger. “Speed-up Heuristic for an On-Demand Ride-
Pooling Algorithm”. http://arxiv.org/pdf/2007.14877v1, July 29, 2020. doi:10 . 48550 / arXiv . 2007 .
14877.url:http://arxiv.org/pdf/2007.14877v1.
[70] Marco Pavone, Stephen L. Smith, Emilio Frazzoli, and Daniela Rus. “Robotic load balancing for mobility-
on-demand systems”. In: The International Journal of Robotics Research 31.7 (2012), pp. 839–854. issn:
0278-3649. doi:10.1177/0278364912444766.url:http://asl.stanford.edu/wp-content/papercite-
data/pdf/Pavone.Smith.ea.IJRR12.pdf.
[71] Florian Dandl, Benedikt Bracher, and Klaus Bogenberger. “Microsimulation of an autonomous taxi-system
in Munich”. In: 2017 5th IEEE International Conference on Models and Technologies for Intelligent Trans-
portation Systems (MT-ITS). 2017, pp. 833–838. doi:10.1109/MTITS.2017.8005628.
[72] DHL Global. DHL and Streetscooter develop new electric drive vehicle with hydrogen technology. 2019.
url:https://www.dhl.com/global- en/home/press/press-archive/2019/dhl-and- streetscooter-
develop-new-electric-drive-vehicle-with-hydrogen-technology.html (visited on 04/21/2022).
27
