Content uploaded by Frederik Schulte
Author content
All content in this area was uploaded by Frederik Schulte on Oct 12, 2022
Content may be subject to copyright.
The Share-A-Ride Problem with Integrated
Routing and Design Decisions: The Case of
Mixed-Purpose Shared Autonomous Vehicles
Max van der Tholen, Breno A. Beirigo, Jovana Jovanova, and Frederik Schulte
Delft University of Technology, The Netherlands
{m.p.vandertholen, b.alvesbeirigo, j.jovanova, f.schulte}@tudelft.nl
Abstract. The shared autonomous vehicle (SAV) is a new concept that
meets the upcoming trends of autonomous driving and changing demands
in urban transportation. SAVs can carry passengers and parcels simul-
taneously, making use of dedicated passenger and parcel modules on
board. A fleet of SAVs could partly take over private transport, taxi,
and last-mile delivery services. A reduced fleet size compared to con-
ventional transportation modes would lead to less traffic congestion in
urban centres. This paper presents a method to estimate the optimal ca-
pacity for the passenger and parcel compartments of SAVs. The problem
is presented as a vehicle routing problem and is named variable capacity
share-a-ride-problem (VCSARP). The model has a MILP formulation
and is solved using a commercial solver. It seeks to create the optimal
routing schedule between a randomly generated set of pick-up and drop-
off requests of passengers and parcels. The objective function aims to
minimize the total energy costs of each schedule, which is a trade-off
between travelled distance and vehicle capacity. Different scenarios are
composed by altering parameters, representing travel demand at differ-
ent times of the day. The model results show the optimized cost of each
simulation along with associated routes and vehicle capacities.
Keywords: Shared autonomous vehicles ·Capacity optimization ·Ve-
hicle routing problem.
1 Introduction
Urbanization is a phenomenon that is becoming ever more apparent across the
world. Already in 2018, over 55% of the world’s population was located in urban
areas with prospects of an increase to almost 70% mid-21st century [22]. The
ongoing demographic changes within cities give cause for new developments in
the transportation of people and goods. Other trends too will have an impact
on transport within urban centres. E-commerce is growing fast, with a massive
demand for business-to-customer movements. On top of that comes a desire for
quick deliveries, sometimes even as fast as a couple of hours. Another trend is
that of the sharing economy, in which customers and businesses share resources,
potentially reducing freight movements and fleet sizes. Finally, climate change
2 M. van der Tholen et al.
awareness and sustainability play an ever-increasing role in reducing emissions
and improve quality of life in heavily congested areas [20]. With these aspects
in mind, new approaches to vehicle design are taken.
The development of autonomous vehicles is expected to bring significant
changes to the mobility patterns of vehicle users. Connecting vehicles through
an internet of autonomous vehicles enables services such as intelligent trans-
portation and ridesharing [10]. The concept of ridesharing promises to improve
the efficiency of individual, on-demand transportation in densely populated ar-
eas. Combining the benefits of autonomous driving and ridesharing allows for the
introduction of autonomous mobility-on-demand (AMoD). This approach con-
sists of fully autonomous driving vehicles that can combine multiple traveling
requests into one journey. AMoD has the potential to reduce traffic congestion
and parking problems while offering fast, on-demand mobility, relieving passen-
gers from the task of driving, and improving safety [24].
Naturally, not only the transport of passengers in urban areas is growing.
Transportation of goods through parcel delivery is increasing and is required to
become faster and cheaper. Short trips through cities, such as last-mile delivery
services, could potentially be done by purpose-built autonomous vehicles (AV)
[4].
While AMoD can already be more efficient and sustainable than a conven-
tional approach, the results heavily depend on traveller demand. Passenger re-
quest numbers are typically much higher at day time than at night and peak
during morning and afternoon rush hours [3]. As a result, large portions of the
fleet of vehicles will be idle or inefficiently occupied during low-demand periods.
Unifying the separate vehicle fleets for passenger and parcel transport into one
fleet with mixed-purpose vehicles can be a solution to improve occupancy levels
and further reduce the number of vehicles on the roads. This share-a-ride ap-
proach was introduced by Li et al. [12], where taxis can combine single parcel
and rider requests, and later expanded by Beirigo et al. [1] in the context of the
share-a-ride with parcel lockers problem (SARPLP).
The SARPLP considers autonomous vehicles with passenger compartments
and parcel lockers, such that both commodity types can be transported simulta-
neously on single journeys. The effectiveness of this concept has been proven, and
92% of simulated scenarios result in higher profit with one mixed-purpose fleet
rather than two single-purpose fleets [1]. However, the model relies on a fixed
vehicle capacity. While proving effective, this approach hardly links the logistic
challenges of shared autonomous driving to the design of shared autonomous
vehicles (SAV). Varying the capacity of SAVs is still an unexplored area.
This study seeks to create a model for a people and freight integrated trans-
portation system (PFIT) in an AMoD setting with variable vehicle capacity.
We aim to find the optimal capacity for mixed purpose SAVs whose internal
space can be outfitted to simultaneously transport passengers and parcels. This
optimal capacity can then be used to constrain the design and give an early
approximation of the dimensions and features of SAVs according to the demand
patterns of the service area. The outcome shows whether the SAV will look like
The Share-A-Ride Problem with Integrated Routing and Design Decisions 3
an ordinary passenger car, a large bus, or anything in between. From this point
on, the problem will be referred to as the variable capacity share-a-ride problem
(VCSARP).
This paper will continue by mentioning some of the relevant literature that is
related to the subject. After that, the problem definition and model formulation
are explained. Next, the experimentation section will picture the scenarios and
give the results of the simulations. The final section will summarize the findings,
conclude the current research and give recommendations for further research.
2 Related Work
The vehicle routing problem (VRP) is a classical optimization problem that
aims to determine the optimal set of routes to be taken by a fleet of vehicles to
serve a set of customers [21]. The first mathematical programming formulation
and algorithmic approach for the VRP was the truck dispatching problem by
Dantzig and Ramser [7] from 1959. Ever since, efforts have been made to extend
VRPs and make them more realistic. The VCSARP is based on a combina-
tion of previous VRPs. The Dial-a-Ride Problem (DARP) is a well-known VRP
variation that consists of designing vehicle routes for on-demand pick-up and
delivery requests. The DARP is built up from a combination of existing VRPs,
including the Pick-up and Delivery Vehicle Routing Problem and the Vehicle
Routing Problem with Time Windows [6]. With the popularization of app-based
transportation services, the DARP has been the basis for passenger ridesharing
services (see, e.g., [13], [19]). More recently, the advantages of autonomous vehi-
cles for ridesharing have been explored, for instance, considering service quality
improvements when platforms activate idle/ parked vehicles [2].
However, short-haul integration of passenger and good flows is hardly ob-
served both in practice [20] and in the literature [15], especially in a ridesharing
context. Among the few models for PFIT systems is the SARP, an extension of
the DARP introduced by Li et al. [12] that allows taxis to transport one pas-
senger and one parcel simultaneously. This problem has been further covered by
Nguyen et al. [16] and Do et al. [8], where a taxi is able to carry one passenger
and multiple parcels. A more flexible version of the SARP is the SARPLP [1],
in which the vehicles consist of passenger compartments and parcel lockers, thus
being able to serve several customers at once. This study shows that a shared,
mixed-purpose fleet proves more profitable to the transport company. More work
on passenger and parcel ridesharing was done by Ronald et al. [18]. Their model
considers passengers requesting transport between homes and activity locations
and parcels that are transported from shops to homes. Ultimately, they find
that ridesharing resources in vans and taxis results in shorter waiting and travel
times. Finally, to the best of our knowledge, the only noteworthy contribution
considering variable capacity and VRPs is by Louveaux and Salazar-Gonz´alez
[14]. Their model, however, does not consider any ridesharing.
4 M. van der Tholen et al.
3 Problem Description
The VCSARP aims to create the most cost-effective routing schedule for SAVs
across a city with a known set of transport requests. The requests consist of pick-
up and delivery points and need to be satisfied within a certain time window.
The SAVs start and end their routes at a centrally located depot. The output
of the model gives the total cost of the routing and the optimal vehicle capacity
that is needed to achieve this. The MILP model is explained in the remainder
of this section.
3.1 Model Formulation
The virtual city in which the SAVs operate is expressed as a 20 by 20 grid struc-
ture with intervals of 100 metres between each node. Within this grid structure,
a total of ntransport requests is generated, split up into nhpassenger and np
parcel requests. Vrepresents the complete set of nodes, including requests and
start/end depot, while Arepresents the arcs connecting all those nodes. The
distances and travel times between all nodes are captured by di,j and ti,j respec-
tively.
The requests are characterised by their quantity qr
i, where resource ris 0 for
a passenger request or 1 for a parcel request. Pick-up quantities qr
iare generated
as positive amounts and drop-off quantities qr
i+nare of the same magnitude but
negative, representing a loss of load. Each request has a pickup time window
[ei,li], which is Trwide, and a maximum travel time delay δr. Both must be
satisfied for a feasible solution. All requests are generated within a time horizon
H. Each pick-up and delivery stop has a service time delay s.
The set of vehicles is K. Each vehicle has a capacity Qr, which is variable
but constrained by an upper and lower bound [Qr
min,Qr
max] for realism and
computational speed. The velocity of the vehicles is assumed to be constant and
the same across each arc in A.
The model seeks to minimize the total costs of a routing schedule. The total
costs are calculated as the product of travelled distance and cost per unit of
distance that varies with vehicle type). The vehicle capacity impacts the travel
cost per kilometer because larger-capacity vehicles are heavier and bulkier, thus
consuming more energy. This relationship between vehicle capacity and operat-
ing costs is visualized in Figure 1. A complete overview of variables, parameters,
sets, and indices needed for the formulation of the VCSARP can be seen in Table
1.
Figure 2 shows some examples of how the VCSARP model works. Each sce-
nario has two passenger and two parcel requests. When requests are positioned
close together and very few detours have to be made, the model will most likely
choose a larger capacity vehicle to serve multiple requests simultaneously. Fig-
ure 2a shows this case. Assuming that each request quantity is equal to 1, the
minimum vehicle capacity for the vehicle in this scenario must be 2 for passen-
gers and 2 for parcels. Figure 2b shows a case where the requests are not all
located favourably. Here one large vehicle is not able to fulfil all the requests
The Share-A-Ride Problem with Integrated Routing and Design Decisions 5
within their time windows. Using smaller vehicles covering slightly larger dis-
tances might even be more efficient. The vehicle capacity in this scenario is 1 for
passengers and 2 for parcels.
Fig. 1: Linear relation between vehicle capacity and operating cost per unit of
distance. Constants αand βrare explained in section 4.
a) b)
Fig. 2: Two examples of the VCSARP. Blue nodes are pick-ups and green nodes
are drop-offs. Example a) shows a scenario in which only one vehicle (solid path)
is deployed from the depot. Example b) shows a situation where one vehicle is
not sufficient to meet all the constraints, such that two vehicles are deployed
(solid and dashed paths). Vehicle loads along each path are displayed for each
resource, where qr
iis the amount of resources that must be loaded at node iand
wk
iis the load of resources on vehicle kafter node i.
3.2 Model formulation
The MILP formulation of the model is as follows:
Minimize:
6 M. van der Tholen et al.
Table 1: Sets, indices, parameters and variables of the VCSARP.
Sets and indices
Ph={1, ..., nh}. Human pickup nodes
Pp={nh+ 1, ..., n}. Parcel pickup nodes
P=Ph∪Pp. All pickup nodes
Dh={n+ 1, ..., n +nh}. Human drop-off nodes
Dp={n+nh+ 1, ..., 2n}. Parcel drop-off nodes
D=Dh∪Dp. All drop-off nodes
V={0} ∪ P∪D∪ {2n+ 1}. All nodes including start/end depots
A={i, j :i∈V, j ∈V , i 6=j}. Arcs connecting all nodes
K={1, ..., nk}. Vehicles
R={0,1}. Resources (0 = human, 1 = parcel)
Parameters
nhNumber of passenger requests
npNumber of parcel requests
n=nh+np. Total number of requests
nkNumber of vehicles
vavg Average vehicle velocity
di,j Shortest distance between nodes iand j
ti,j Shortest travel time between nodes iand j
qr
iAmount of resource rthat must be loaded at node i
{ei, li}Time window for request i
TrPickup time window width for each resource r
HTime horizon
δrMaximum ride time delay for each resource r
sService time
Qr
min Lower bound of vehicle capacity for each resource r
Qr
max Upper bound of vehicle capacity for each resource r
αConstant of the objective function
βrSlope of the objective function for each resource r
pphPassenger pickup probability [residential, industrial, campus]
dphPassenger delivery probability [residential, industrial, campus]
pppParcel pickup probability [residential, industrial, campus]
dppParcel delivery probability [residential, industrial, campus]
Variables
xk
i,j Traveled arcs (i, j) of each vehicle k
τk
iArrival time of vehicle kat node i
wr,k
iLoad of each resource ron vehicle kafter node i
trk
iRide time of pickup ion vehicle k
QrVehicle capacity for each resource r
The Share-A-Ride Problem with Integrated Routing and Design Decisions 7
C=X
i,j∈A
X
r∈R
X
k∈K
(α+βrQr)di,j xk
i,j (1)
Subject to:
X
j∈V
j6=i
X
k∈K
xk
i,j = 1 ∀i∈P(2)
X
j∈V
j6=0
xk
0,j =X
i∈V
i6=2n+1
xk
i,2n+1 = 1 ∀k∈K(3)
X
j∈V
j6=i
xk
i,j =X
j∈V
j6=i+n
xk
i+n,j ∀i∈P, ∀k∈K(4)
X
j∈V
j6=i
xk
j,i =X
j∈V
j6=i
xk
i,j ∀i∈N, ∀k∈K(5)
τk
j= (τk
i+ti,j +s)xk
i,j ∀i, j ∈A, ∀k∈K(6)
τk
i+n≥τk
i∀i∈P, ∀k∈K(7)
ei≤τk
i≤li∀i∈P, ∀k∈K(8)
wr,k
j≥(wr,k
i+qr
j)xk
i,j ∀i, j ∈A, ∀r∈R, ∀k∈K(9)
max(0, qr
i)≤wr,k
i≤Qr∀i∈V, ∀r∈R, ∀k∈K(10)
Qr
min ≤Qr≤Qr
max ∀r∈R(11)
trk
i=τk
i+n−τk
i∀i∈P, ∀k∈K(12)
ti,i+n≤trk
i≤ti,i+n+δr∀i∈P, ∀r∈R, ∀k∈K(13)
xk
i,j ∈ {0,1} ∀i, j ∈A, ∀k∈K(14)
τk
i∈N∀i∈V, ∀k∈K(15)
wr,k
i∈Z∀i∈V, ∀r∈R, ∀k∈K(16)
trk
i∈N∀i∈V, ∀k∈K(17)
Qr∈Z∀r∈R(18)
The objective function (1) aims to minimize the total cost, which is calculated
as (cost/km)*(travelled distance). Based on [11] and [23], we consider the en-
ergy consumption and running costs (denoted by cost/km) increase linearly with
vehicle capacity. Parameters αand βrare the intercept and slope of the linear
function and are quantified in Section 4. Equation (2) guarantees that each re-
quest is served once. Each vehicle must start and end its route at the depot,
which is controlled by (3), while (4) ensures that pick-up and delivery of one re-
quest are done by the same vehicle. The final routing constraint (5) guarantees
conservation of flow, meaning that a vehicle entering a node must also leave that
node again. The definition of the arrival time of SAVs at nodes is given by (6).
Vehicles must first complete the pick-up of a request before the drop-off, which
is guaranteed by (7). Equation (8) ensures that arrival at the pick-up nodes is on
8 M. van der Tholen et al.
time and within the required time window. The vehicle load or weight after each
node is defined by equation (9). This load must never become negative, be al-
ways larger than the previous request quantity, and never exceed the maximum
loading capacity, of which (10) makes sure. The vehicle capacity has a lower
and upper bound, which are imposed by (11). Each request has a total time
spent on the vehicle. The minimum ride time is given by (12). The actual ride
time cannot exceed this by more than the maximum ride time delay, which is
guaranteed by (13). The model’s five decision variables are traveled arcs, arrival
times, compartment loads, ride times, and vehicle capacities. These are defined
respectively by (14), (15), (16), (17), and (18).
4 Experimental study
Once traffic flows and transportation demand fluctuate throughout the day, we
carry out an experimental study to obtain insights into the ideal capacity of
an SAV considering different demand scenarios. Ideally, an SAV system should
function efficiently at any time, consistently featuring high occupancy rates and
low idleness. First, to simulate the various times of day in which an SAV operates,
the following scenarios have been considered:
– Morning rush hour: During morning hours, there is a high density of
passenger requests from homes to workplaces. This causes traffic to flow
from residential areas to industrial and commercial areas. The amount of
parcel movements is significantly smaller.
– Afternoon: A well-mixed blend of parcel and passenger movements. Parcels
tend to move from industrial areas towards residential areas, while passenger
travel patterns are more evenly distributed across the city.
– Late-afternoon rush hour: Similar to the morning hours, passenger de-
mand is higher than parcel demand. However, passenger traffic flow is re-
versed from industrial/commercial areas to residential areas.
– Evening: The time of day at which most people are at home creates great
opportunities for parcel deliveries. This scenario is dominated by parcel
transport requests from industrial to residential areas, but some scattered
passenger transport occurs too.
Considering that previously the city map was formulated as a rather abstract
grid structure without any information on the function of each node, neighbour-
hoods are added to the grid. These are described as four rectangular sections
on the map. Two sections are representing residential areas, one section is an
industrial area, and the final section is a campus with a university and offices.
These sections or neighbourhoods can now be used to create scenarios. Request
locations are generated using a probability distribution that can shape the traffic
flow patterns of the scenario. For example, in the morning scenario, there will be
a high probability that passenger pick-ups will occur on any of the nodes within
the residential areas and a much lower probability of them occurring at indus-
trial or campus nodes. Likewise, passenger destinations will more likely occur in
The Share-A-Ride Problem with Integrated Routing and Design Decisions 9
the industrial or campus neighbourhood, rather than in one of the residential
areas. A visual representation of the map with the different neighbourhoods is
shown in figure 3.
Fig. 3: City grid divided into neighbourhoods with different functions. The depot
is located in the middle.
The general parameters of the VCSARP are constant across all scenarios. The
total number of requests and available vehicles are set to 8 and 4, respectively.
These relatively low numbers are needed to limit computation times, which can
become very large due to the complexity of the model. To make up for the
low number of requests, request quantities are chosen randomly from a U(4,8)
distribution. The case of a few large requests (quantities between 4 and 8) can be
considered analogous to a larger number of smaller requests with similar origin
and destination, essentially creating a scenario that serves much more customers.
The parcel and passenger capacity lower and upper bounds are both set to 4 and
24, respectively. Assuming that one passenger seat takes up the space of about
4 large parcels, that results is a total “passenger size” capacity between 4 and
30. Passenger time windows and delivery delays are set quite tight once most
people would not want to experience much delay during their trip. Conversely,
parcel time constraints are much less strict, allowing for a maximum delay of
1 hour at delivery. The average speed is set at 20 km/h, which is realistic in
cities with short stopping intervals [17]. A short 10-minute total time horizon is
chosen because of the small number of requests. To simulate the (un)loading of
resources, a 1-minute service time at each node is added. The objective function
10 M. van der Tholen et al.
parameters αand βrare retrieved from real-life electricity costs and consumption
data of electric vehicles (see [9] and [5]).
Table 2: General parameter values.
Parameter Value
n8
nk4
q0
iU(4,8)
q1
iU(4,8)
[Q0
min, Q0
max] [4,24]
[Q1
min, Q1
max] [4,24]
T03 minutes
T130 minutes
δ010 minutes
δ010 minutes
vavg 20 km/h
H10 minutes
s1 minute
α0.022 euro/km
β00.00308 euro/(km x passenger)
β10.00077 euro/(km x parcel)
Table 3: Scenario specific parameter values.
Scenario
Parameter Morning Afternoon Late-afternoon Evening
Number of requests
nh6 4 6 2
np2 4 2 6
Pickup and delivery probabilities [res, ind, cam]
pph[1, 0, 0] [1
/3,1
/3,1
/3] [0, 1
/2,1
/2] [1
/3,1
/3,1
/3]
dph[0, 1
/2,1
/2] [1
/3,1
/3,1
/3] [1, 0, 0] [1
/3,1
/3,1
/3]
ppp[0, 1, 0] [1
/5,4
/5, 0] [0, 1, 0] [1
/5,4
/5, 0]
dpp[1, 0, 0] [4
/5,1
/5, 0] [1, 0, 0] [4
/5,1
/5, 0]
5 Results
The computations were performed by an Intel Core i7 @ 2.20GHz processor,
16GB RAM computer. The programming was done in Python, and the MILP
model was solved using Gurobi Optimizer 9.0.2.
The Share-A-Ride Problem with Integrated Routing and Design Decisions 11
Table 4 shows for each scenario the average optimal vehicle capacity for
both resources and the total vehicle capacity (i.e., the combination of passenger
and parcel capacities). One passenger space is considered the same size as 4
parcel spaces. A vehicle with a passenger capacity of 4 and a parcel capacity
of 8 would thus have a total capacity of 6. This value determines the overall
interior volume of the vehicle, which can be of use for the design of the vehicle,
and, ultimately, the number of vehicles used. Simulations that did not converge
to a 0% optimality gap, thus not reaching the most optimal solution, within
a 30-minute time limit were discarded. In total, at least 30 simulations with
optimal solutions were generated for each of the four scenarios. Table 5 shows
the average total costs, distance travelled, and cost/km across scenarios and
Figure 4 illustrates the outcome routes of a single simulation for each scenario.
Table 4: The results of each scenario that are related to vehicle capacity and
design. These are average values from 30 instances on each scenario.
Scenario Passenger cap. Parcel cap. Total cap. # of vehicles
Morning 12.8 8.23 14.9 3.60
Afternoon 8.50 11.9 11.5 3.31
Late-afternoon 12.9 8.93 15.2 3.50
Evening 6.75 17.6 11.1 2.05
Table 5: Average costs, distance traveelled, and cost/km for each scenario across
30 different instances.
Scenario Costs [e] Distance travelled [km] Cost/km [e/km]
Morning 1.84 20.6 0.0898
Afternoon 1.46 18.3 0.0793
Late-afternoon 1.67 18.5 0.0907
Evening 1.14 15.1 0.0783
Table 4 shows that the optimal vehicle composition and fleet size differ
markedly for each scenario. As input parameters heavily influence the model’s
outcome, engineers can take multiple approaches to design suitable SAVs that
ultimately match the operational requirements. The most straightforward ap-
proach would be designing vehicles using the maximum capacity for each resource
across all scenarios, resulting in a vehicle with around 15 passenger seats and
18 parcel spaces. Naturally, these vehicles would end up under-occupied most of
the time, and costs would be higher than calculated, but most requests could be
satisfied easily. Another approach consists of repeating the simulations for the
afternoon and evening scenarios with adjusted capacity constraints. Considering
12 M. van der Tholen et al.
that vehicles tend to be idler in these scenarios, using a smaller parcel capacity
and deploying more vehicles would also satisfy the conditions. This results in
higher overall costs but more efficient vehicle occupation.
Fig. 4: The results of 4 simulations, where a) is a morning scenario, b) is an
afternoon scenario, c) is a late-afternoon scenario, and d) is an evening scenario.
The red node is the depot, the blue nodes are human pick-ups, the cyan nodes
are human drop-offs, the green nodes are parcel pick-ups and the yellow nodes
are parcel drop-offs. Different colour dotted lines represent different vehicles in
operation. The neighbourhoods are also displayed, as rectangles (see figure 3).
6 Conclusions
This research paper presented a MILP formulation for the variable capacity
share-a-ride problem. The SARP was reformulated to set vehicle capacity as a
decision variable and allow for shared autonomous vehicles. The objective was
to find the optimal capacity of SAVs to give insights into the design of such
vehicles given several operational scenarios, featuring various parcel and people
demand patterns.
Overall, the model provides a basis for estimating the optimal capacity of
SAVs in a static scenario. The parameter inputs are flexible and allow for a wide
The Share-A-Ride Problem with Integrated Routing and Design Decisions 13
variety of scenarios. A point of critique is the limited number of requests that
the model is able to solve within a reasonable time. This was countered by using
larger quantity requests, each representing multiple single requests that have
similar origins and destinations.
The results show that the optimal capacity is highly dependent on the sce-
nario parameters. Scenarios with high passenger transport demand call for ve-
hicles with large passenger capacity and smaller parcel capacity. Scenarios with
more parcel movements have an opposite effect on both capacities. While this is
to be expected, it makes it hard to find an optimum capacity that will satisfy
all scenarios and be cost-efficient at the same time. The concept of SAV in this
paper uses separate compartments for passenger and parcel transport. Empty
passenger spaces cannot be used for parcels and vice versa. One could rethink
this concept and go for a more flexible utilization of interior space. Possible
solutions include foldable seats for additional parcel storage, under-seat stor-
age of parcels, or a simple flat floor with standing/leaning space for passengers
and freely usable space for parcels. These concepts could easily be implemented
into the current model with simple reformulations. Alternatively, a more flexible
solution could be considered, where AVs are dynamically outfitted at service
points, having their people and parcel compartment capacity adjusted to match
the demand changes. Future work will focus on modeling such a flexible setting
considering additional operational characteristics such as different revenues for
passengers and parcels, penalties for extended ride time, premiums for private
travel requests, vehicle and compartment purchasing costs.
References
1. Beirigo, B.A., Schulte, F., Negenborn, R.R.: Integrating People and Freight
Transportation Using Shared Autonomous Vehicles with Compartments. IFAC-
PapersOnLine 51(9), 392–397 (2018). https://doi.org/10.1016/j.ifacol.2018.07.064,
https://doi.org/10.1016/j.ifacol.2018.07.064
2. Beirigo, B.A., Schulte, F., Negenborn, R.R.: A Learning-Based Optimization Ap-
proach for Autonomous Ridesharing Platforms with Service Level Contracts and
On-Demand Hiring of Idle Vehicles. Transportation Science (2021)
3. Boesch, P.M., Ciari, F., Axhausen, K.W.: Autonomous vehicle fleet sizes required
to serve different levels of demand. Transportation Research Record 2542, 111–119
(2016). https://doi.org/10.3141/2542-13
4. Buchegger, A., Lassnig, K., Loigge, S., M¨uhlbacher, C., Steinbauer, G.: An
Autonomous Vehicle for Parcel Delivery in Urban Areas. In: Proc. of the
IEEE Conference on Intelligent Transportation Systems. pp. 2961–2967 (2018).
https://doi.org/10.1109/ITSC.2018.8569339
5. ˇ
Cern´y, J.: Testing of Five Different Types of Electric Buses. In: Proc. of the CIV-
ITAS Forum Conference 2015. Ljubljana, Slovenia (2015)
6. Cordeau, J.F., Laporte, G.: The Dial-a-Ride Problem (DARP): Variants, modeling
issues and algorithms. 4or 1(2), 89–101 (2003). https://doi.org/10.1007/s10288-
002-0009-8
7. Dantzig, G.B., Ramser, J.H.: The truck dispatching problem. Management science
6(1), 80–91 (1959)
14 M. van der Tholen et al.
8. Do, P.T., Nghiem, N.V.D., Nguyen, N.Q., Nguyen, D.N.: A practical dynamic
share-a-ride problem with speed windows for Tokyo city. In: Proc. of the 8th In-
ternational Conference on Knowledge and Systems Engineering. pp. 55–60. IEEE
(2016). https://doi.org/10.1109/KSE.2016.7758029
9. Electric Vehicle Database: Energy consumption of full electric vehicles cheatsheet
- EV Database. https://ev-database.uk/cheatsheet/energy-consumption-electric-
car (2020), https://ev-database.uk/cheatsheet/energy-consumption-electric-car,
accessed: 2021-01-15
10. Jameel, F., Chang, Z., Huang, J., Ristaniemi, T.: Internet of autonomous vehicles:
Architecture, features, and socio-technological challenges. IEEE Wireless Commu-
nications 26(4), 21–29 (2019)
11. Jung, H., Silva, R., Han, M.: Scaling trends of electric vehicle performance: Driving
range, fuel economy, peak power output, and temperature effect. World Electric
Vehicle Journal 9(4), 1–14 (2018). https://doi.org/10.3390/wevj9040046
12. Li, B., Krushinsky, D., Reijers, H.A., Van Woensel, T.: The Share-A-Ride
Problem: People and parcels sharing taxis. European Journal of Opera-
tional Research 238(1), 31–40 (2014). https://doi.org/10.1016/j.ejor.2014.03.003,
http://dx.doi.org/10.1016/j.ejor.2014.03.003
13. Lin, Y., Li, W., Qiu, F., Xu, H.: Research on Optimization of Vehicle Routing
Problem for Ride-sharing Taxi. Procedia - Social and Behavioral Sciences 43, 494–
502 (2012). https://doi.org/10.1016/j.sbspro.2012.04.122
14. Louveaux, F.V., Salazar-Gonz´alez, J.J.: Solving the single vehicle routing prob-
lem with variable capacity. Transportation Science 50(2), 708–719 (2016).
https://doi.org/10.1287/trsc.2014.0556
15. Mourad, A., Puchinger, J., Chu, C.: A survey of models and algorithms for op-
timizing shared mobility. Transportation Research Part B: Methodological 123,
323–346 (2019)
16. Nguyen, N.Q., Tuan, K.L., Nghiem, N.V.D., Nguyen, M.S., Thuan, P.D., Mukai, N.:
People and parcels sharing a taxi for Tokyo city. In: Proc. of the Sixth International
Symposium on Information and Communication Technology. pp. 90–97 (2015).
https://doi.org/10.1145/2833258.2833309
17. Oskarbski, J., Birr, K., Miszewski, M., Zarski, K.: Estimating the average speed of
public transport vehicles based on traffic control system data. In: Proc. of the 2015
International Conference on Models and Technologies for Intelligent Transporta-
tion Systems. pp. 287–293 (2015). https://doi.org/10.1109/MTITS.2015.7223269
18. Ronald, N., Yang, J., Thompson, R.G.: Exploring Co-Modality Using On-
Demand Transport Systems. Transportation Research Procedia 12, 203–212
(2016). https://doi.org/10.1016/j.trpro.2016.02.059
19. Santos, D.O., Xavier, E.C.: Taxi and ride sharing: A dynamic dial-a-
ride problem with money as an incentive. Expert Systems with Appli-
cations 42(19), 6728–6737 (2015). https://doi.org/10.1016/j.eswa.2015.04.060,
http://dx.doi.org/10.1016/j.eswa.2015.04.060
20. Savelsbergh, M., Van Woensel, T.: 50th Anniversary Invited Article—City Logis-
tics: Challenges and Opportunities. Transportation Science 50(2), 579–590 (2016).
https://doi.org/10.1287/trsc.2016.0675
21. Toth, P., Vigo, D.: The vehicle routing problem. SIAM (2002)
22. United Nations; Department of Economic and Social Affairs; Population Divi-
sion: World Urbanization Prospects: The 2018 Revision. United Nations, New York
(2019)
The Share-A-Ride Problem with Integrated Routing and Design Decisions 15
23. Weiss, M., Cloos, K.C., Helmers, E.: Energy efficiency trade-offs in small
to large electric vehicles. Environmental Sciences Europe 32(1) (2020).
https://doi.org/10.1186/s12302-020-00307-8, https://doi.org/10.1186/s12302-020-
00307-8
24. Zhang, R., Spieser, K., Frazzoli, E., Pavone, M.: Models, algorithms, and evalua-
tion for autonomous mobility-on-demand systems. In: Proc. of the 2015 American
Control Conference. pp. 2573–2587. American Automatic Control Council (2015).
https://doi.org/10.1109/ACC.2015.7171122
