Journal of Air Transport Management 107 (2023) 102356

Contents lists available at ScienceDirect

Journal of Air Transport Management

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

3D-flight route optimization for air-taxis in urban areas with Evolutionary

Algorithms and GIS
Moritz Hildemann a ,∗, Judith A. Verstegen b
a University of Münster, Institute for Geoinformatics, Heisenbergstraße 2, 48149 Münster, Germany
b
Utrecht University, Department of Human Geography and Spatial Planning, Heidelberglaan 8, 3584 CS Utrecht, The Netherlands

ARTICLE INFO ABSTRACT

Keywords: Electric aviation is being developed as a new mode of transportation for the urban areas of the future. This
Urban air management requires urban air space management that considers these aircraft. Flight routes need to be determined that
Flight route optimization avoid no-fly areas, and minimize flight time, energy consumption and added noise. Yet, no method currently
Genetic Algorithms
exists for optimizing urban flight routes under multiple conflicting objectives while avoiding three-dimensional
GIS
restricted areas. In our work, this research gap is overcome by optimizing 3D-routes with the multi-criteria
Air-taxis
optimization technique called Non-dominated Sorting Genetic Algorithm II. We propose a novel procedure
in the optimization process to incorporate geographical representations. Furthermore, we include a seeding
procedure for initializing the flight routes and repair methods for invalid flight routes that may arise during
the optimization process. We apply the optimization to a case study in Manhattan (New York City) for two
different aircraft types, the Lilium Jet (vectored thrust) and the EHANG 184 (wingless multicoptor), under
three objectives concerning flight time, energy consumption and added noise. Compared to a least-distance
path, flight routes were obtained with maximum improvements of 38% in added noise, 65% in flight time and
52% in energy consumption for the EHANG 184. For the Lilium jet, maximum improvements of 43% in added
noise, 47% in flight time and 47% in energy consumption were obtained. Still, the obtained noise addition
levels by the aircraft in New York City exceed 5 dB, which is considered as long-term noise annoyance. We
illustrated that minimizing the added noise requires high search effort compared to the other two objectives.
Upon further analysis of the optimization results, we conclude that the Lilium jet as representative of the
eVTOL type vectored thrust is more sensitive to flight route changes than the multicoptor EHANG 184. This
information may help in air taxi type choice for a certain region as well as in flight route planning.

1. Introduction and Rakas, 2021). Air space segmentation into fly and no-fly zones,
depending on the location and the time of the day (Geister, 2017),
The traffic in today’s Megacities is congested and the commuting can ensure permanently secured locations of special safety concern like
times in these cities are high. Therefore, a new mode of transport hospitals as well as occasional no-fly zones in severe weather conditions
has been developed during the past years: The air taxi type Electrical or during public assemblages.
Vertical Take-Off and Landing vehicles (eVTOL). The expected potential After segmenting the air space, flight routes for the eVTOL need
demand in the highly competitive air taxi market in the United States to be formulated. However, the optimality of a route depends on
only is already 80.000 passengers per day Goyal (2018), and Rajendran the viewer’s perception. For eVTOL operator, the commuting time
and Srinivas (2020) expect air taxi operations in urban space within
and the energy cost for the flight route might be crucial. For the
the forthcoming years. Electrical Vertical Take-Off and Landing vehicles
commuter, the safety, commuting time and price might be the most
are air taxis with suitable flight characteristics for the urban air space:
relevant criteria (Baur et al., 2018), whereas for the other citizens, the
Vertical landing and take-off allows precise manoeuvring, and the
discomfort produced by the aircraft noise might be the most important
electric propulsion makes them quieter, safer, and less expensive to
operate compared to non-electrical helicopters (Garrow et al., 2021). criterion (Berglund et al., 1999). Therefore, a method is demanded that
In order to successfully integrate eVTOL into the urban air space, flight incorporates multiple objectives linked to spatially-varying attributes in
laws, other airspace users, and citizens have to be respected (Bauranov the search for optimal flight routes within the fly zones.

∗ Corresponding author.
E-mail addresses: [email protected] (M. Hildemann), [email protected] (J.A. Verstegen).

https://doi.org/10.1016/j.jairtraman.2022.102356
Received 30 June 2020; Received in revised form 29 December 2022; Accepted 29 December 2022
Available online 12 January 2023
0969-6997/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Geographical Information Systems (GIS) enable to compute cost- accidents will likely increase when the use of air taxis within the dense
surfaces based on weighted spatial attributes within fly zones (Hilde- urban air space is intensified. Besides, urban air mobility increases
mann and Delgado, 2019). With such a cost surface, the least cost path the already high noise levels in cities by up to 40 dB at the ports, as
between an origin and a destination can be found (Durmaz et al., 2019). results from Glaab et al. (2019) indicate. de Paiva Vianna et al. (2015)
In that case, the altitude is included in the cost value at each grid cell of show that exposure to high noise levels has a significant effect on the
the cost surface. The outcome is a two-dimensional line with each node annoyance and sleep quality of affected people, and the Environmental
extracted to the minimal flight height surface. Another GIS method is Protection Agency (1974) defines high noise levels as pollutant. The
3D-Routing along 3D-Networks. This solves the combinatorial problem implementation of no-fly zones at the right locations may help to
to travel along an existing network (Atila et al., 2013), which can be reduce the risk and annoyance to a minimum and hence to increase the
designed to avoid restricted airspace (Dai and Low, 2022). On street likelihood of acceptance. Geister (2017) propose a concept for no-fly
networks, the possible solutions are limited by the existing streets. The zones, and Hildemann and Delgado (2019) proposed an implementa-
problem of the cost-surface method is that the weights of the objectives tion of such three-dimensional no-fly zones in urban air space. The air
have to be predefined. Furthermore, both methods have the drawback space restrictions consider both flight regulations and a subset of land
that they do not search the full 3D-space. The cost-surface method uses and points of interest like parks, schools or recreational areas.
only finds routes on the minimal height surface, while the 3D routing In addition to restricting specific areas of the urban air space,
methods finds routes in a predefined 3D network. This is a shortcoming flight route planning plays an important role to reduce negative side
in the context of eVTOL because those aircraft can move in space effects (Rajendran and Pagel, 2020). Many trajectory concepts were
without being bound to a surface or network. proposed that reduce the flight time, e.g. Zhao et al. (2020), and
This work aims to fill the stated research gap to optimize routes for some include additional aspects like emissions (Okuniek et al., 2016),
eVTOL transportation in 3D fly zones with multiple objectives. Borrow- manoeuvrability aspects (Wang et al., 2021), or wind (Pradeep et al.,
ing techniques that have been applied to other domains, such as pipe 2022). Pradeep et al. (2020) state that wind has little impact on short
routing (Sandurkar and Chen, 1999) and drone mission planning (Peng flights in urban air space. Even though noise reduction for urban air
et al., 2011), we use a hybrid of GIS and Evolutionary Algorithms, mobility is identified as an important goal by researchers (Eißfeldt,
a set of search techniques for continuous optimization (Sivanandam 2020) and by space agencies (Rizzi et al., 2020), it has not been
and Deepa, 2008). The generalizable method is applied to a study case incorporated in planning flight routes so far.
of urban eVTOL routing in Manhattan (New York City). The routes
are evaluated for three different criteria, specific for the 3D-routing 3. Methods
in eVTOL transportation given the different stakeholders mentioned
above: The total flight time in minutes, the total energy consumption 3.1. Overview
in kilowatt per hour and the additional (to already existing background
noise) noise pollution in decibels. Our research questions are (1) Can The workflow of the used methods to optimize flight routes is
the three objectives be optimized while complying with the air space displayed in Fig. 1. The flight routes, individuals represented as 3D
restrictions and noise regulations? If yes, how much can a least-distance point arrays and 3D lines, are randomly initialized in a 3D channel
route computed with a GIS be improved for the objectives in the highly (Section 3.6.1). After creation, they are validated and repaired in
restricted air space of New York City for two different eVTOL? and (2) order to be conform with flight regulations (Section 3.6.2). All valid
What information for air space management can be derived from the flight routes are evaluated for the three objectives shortest flight time,
study case results? least energy consumption and least added noise, and are given a non-
domination rank (Section 3.6.3). Individuals are selected based on their
2. Background non-domination rank in order to produce children: the 3D points of the
parents are combined and mutated with the aim of producing a better
Air taxis are expected to be introduced in the urban air space in the flight route (Section 3.6.4). The children are validated and repaired,
future (Ahmed et al., 2021). From a limited competition of air taxis and the best flight routes from the parent and children generation
to medium-ranged rail commuting in year 2018 (Sun et al., 2018), combined form the next generation (Section 3.6.5).
the new eVTOL air taxis with different flight characteristics could In the next sections, the concept of evolutionary algorithms (Sec-
expand the market to shorter distances, competing with other means tion 3.2), the case study (Section 3.3), the used air-taxi representatives
of transport. Rajendran and Srinivas (2020) expect air taxi operations (Section 3.4) and the objective functions (Section 3.5) are explained
in urban space within the forthcoming years and identify multiple first. After that, each step in our proposed optimization approach (
actions that need to be taken for realizing air taxi operations, e.g. ride- Fig. 1) is described in detail in Section 3.6.
matching, pricing strategies, vehicle maintenance scheduling, and pilot
training and recruitment. 3.2. Evolutionary algorithm
Ahmed et al. (2021) analyse the willingness of citizens to use
and pay for air taxi operations. Their survey analysis shows that re- The principle of evolutionary algorithms mimics the biological evo-
spondents show the willingness to use and pay for such services if lution of cells, organs, individuals and populations (Sivanandam and
they expect benefits. Expected benefits are decreased commuting times Deepa, 2008). The Darwinian theory of natural selection proposes that
and fewer and less severe crashes compared to ground transportation organisms that are fitter than other organisms have a higher probability
options. They advise decision-makers to focus on developing regulatory to survive and thus to pass on the genes in reproduction processes.
frameworks that increase safety in air taxi operations. Ward et al. Evolutionary algorithms mimic this idea by analysing a population con-
(2021) analyse the willingness of passengers to fly in autonomous air sisting of several potential solutions to a given problem (Sivanandam
taxis. They found that introducing new vehicles without strong safety and Deepa, 2008). The evolutionary algorithm detects the individuals
records reduces the willingness. Ahmed et al. (2021) and Ward et al. with the highest fitness within this population, in other words the cur-
(2021) reflect on the willingness of citizens to use and pay for air taxis, rently best solutions for to problem (Holland, 1975). These individuals
but citizens not actively using the new mode of transport need to be have a higher probability of being selected for reproduction (selection
considered as well. process). In the reproduction, the crossover process determines how
Citizens at ground prefer not to be exposed to additional risk or elements of the parents are recombined to produce a child. In contrast,
noise by urban air taxi operations (McKinsey & Company, 2021). the mutation process generates characteristics that are not inherited
Despite declining accident rates (Budde et al., 2021), the number of by the parents by randomly perturbing elements (Holland, 1975). The

2
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Table 1
Vertical and horizontal restrictions in meters.
Landuse Vertical restriction Horizontal restriction
Airport 600 7620
Hospitals 300 300
Universities 200 300
Parks 300 100
Graveyards 300 100
Recreational Areas 300 100
Rooftops 152.4 –

the city yields one of the highest market potentials (Goyal, 2018).
Moreover, most of the air space is likely to be regulated into fly and no-
fly zones, where the latter can be divided into restricted and protected
air space (Fig. 2).

• Restricted air space due to flight regulations. Areas around air-

ports within the horizontal distance of 25.000 ft (7620 m) belong
to this category, as well as the space below the height of 500 ft
(152.4 m) and above the height of 700 ft (213.36 m) measured
from the rooftops (Federal Aviation Administration, 2017).
• Protected airspace to consider the possible negative side effects
to the citizens (Geister, 2017). Flying above these specified land
uses can lead to annoyance or discomfort: examples are schools,
graveyards, recreational areas.

We assume that the vertical and horizontal minimum distance to

be maintained by aircraft depends on the land use (Table 1), which
is retrieved from OpenStreetMap (OpenStreetMap contributors, 2019).
Furthermore flight obstacle maps from Federal Flight Administration
(2016) and rooftop heights from Open Data NYC (2018) are obtained.
To assess the performance of the method described in the next
sections, i.e. as a reference route, a least-cost path on the minimal flight
height plane, where the cost is the accumulated euclidean distance
along all raster cells. The least cost flight route from Hildemann and
Delgado (2019) was chosen as reference route, which is ≈6 kilometres
long (Fig. 2).
Fig. 1. Optimization process flowchart.
3.4. Selected eVTOL

next generation, consisting of old (parents) and new (children) individ- Many eVTOL configurations exist with different thrust types, among
uals, is expected to have a higher fitness. The algorithm runs until it of them ‘‘Thrust, Lift and Cruise, Wingless, Hover Bikes and Personal
reaches a predetermined termination criterion. The evolutionary algo- Flying Devices, Electric Helicopters’’ (Electric VTOL News, 2019). The
rithm searches for so called non-dominated solutions. Non-dominated flight characteristics of the aircraft have an impact on the objective
solutions are optimal compromises between the objectives (Sivanan- values and thereby on the optimal flight routes. Therefore, in our
dam and Deepa, 2008), i.e. solutions that cannot be improved for one case study, the wingless multicoptor eVTOL EHANG 184 (EHANG,
objective without becoming worse for the other two objectives (Reddy 2019) is compared to the vectored thrust eVTOL Lilium Jet (Lilium
and Kumar, 2015). All non-dominated solutions can be plotted together Aviation, 2019). We chose these two because they were found to vary
to form a so-called Pareto front. Each point on the Pareto front is the most in short-distance and long-distance performance (Bacchini
optimal, but the ‘‘importance’’ or weight of the objectives varies for and Cestino, 2019). The wingless multicoptor is more energy efficient
each point. As such, the Pareto front visualizes trade-offs between the in short distance missions, whereas the Lilium Jet is more energy
objectives. efficient and faster in long-distance missions. We obtained the aircraft
In our study, we use an evolutionary algorithm to produce valid parameter values needed to calculate the energy consumption and
optimal flight routes for the objectives shortest flight time, lowest speed from Bacchini and Cestino (2019) (Table 2).
energy consumption and minimal added noise level (Fig. 1). In this
study, the Non-Dominated Sorting Genetic Algorithm II (NSGA II) de- 3.5. Objectives for flight route optimization
veloped by Deb et al. (2002) is used as evolutionary algorithm strategy.
This strategy allows to find multiple pareto-optimal solutions in one 3.5.1. Flight time
simulation run. It is well known in the field of optimization and the The first objective is to minimize flight time. To compute the flight
details are outlaid by its developers. time, the maximum flight speed at any position of the flight route needs
to be known. The maximum flight speed is limited by three different
3.3. Case study and reference route factors: The maximum allowed flight speed by the flight authorities,
the maximum possible flight speed by the aircraft and the maximum
The urban area of Manhattan, New York City, (Fig. 2) is an often flight speed that is still comfortable for the passenger given the flight
used study area for air taxi marketing (Lilium Aviation, 2019) and manoeuvre. The maximum cruise speed for the Lilium Jet is 252 km∕h,
air taxi market studies (Goyal, 2018), as the traffic is congested and whereas the EHANG 184 is slower with a maximum speed of 100

3
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 2. Maps of the study area New York City: (a) Overview of the study are with the protected and restricted air space, (b) Overview of the least cost path to be optimized, (c)
Detail view of least cost path.

km∕h (Bacchini and Cestino, 2019). We assume, that the flight agencies Table 2
limit the maximum velocity to 100 km∕h. To guarantee flight comfort, Aircraft parameters for calculation of energy consumption from Bacchini and Cestino
(2019). Cruise power variable for the Vectored Thrust eVTOL: 28 kW in optimal flight
a maximum gravitational force equivalent (𝑔𝐹 𝑜𝑟𝑐𝑒 ) of 1.25 is aimed for. speed, 36 kW at 100 km/h, variable (x) below 100 km/h flight speed. 𝐶𝐿 in drag polar
The formulas for the calculation are provided in Appendices B.1 and is the lift coefficient of the aircraft wings.
B.2. E-Hang 184 Lilium
eVTOL type Wingless Multicoptor Vectored Thrust
3.5.2. Energy consumption Cruise power in kW 34.6 28/36/x
The second objective is to minimize energy consumption. With Hover power in kW 42.7 187
Acceleration energy in kW 42.7 187
the variables flight velocity and the flight angle, the required power Deceleration energy in kW 42.7 187
can be calculated for each flight route segment, when the aircraft Acceleration speed in ms
2
2 2
flight characteristics are known (Table 2). These flight characteristics Deceleration speed in m2
−2 −2
s
change per aircraft type, therefore the energy consumption is aircraft- Wing area in m2 0 3.6
dependent. The vectored thrust aircraft can generate more lift with Empty weight in kg 168 490
increasing flight velocity with their wings. The wingless multicoptor Drag polar – 0.0163 + 0.058𝐶𝐿2

generates less additional lift (Sadraey, 2009). As aircraft with wings

have a better lift/drag ratio with increasing flight speeds, the energy
consumption is speed dependent for the Lilium. The multicoptor from height of the aircraft and the noise pressure of 100 dB, we get the
EHANG on the other hand has no wings and can only produce lift with perceived noise level at ground from the aircraft. Next, we combine
its rotors. The required energy for landing, hovering, acceleration and the noise emission of the aircraft with the background noise at the
deceleration are constant (Table 2). The formulas for the calculation ground, to compute the added noise. The formulas for the calculation
are provided in Appendix B.3. are provided in Appendix B.4.

3.5.3. Added noise

3.6. Optimization with Genetic Algorithms and GIS
The last objective is to minimize the noise added to the existing
background noise at the ground. The Environmental Protection Agency
(1974) and the World Health Organisation (Berglund et al., 1999) 3.6.1. Initialization of individuals and population
support the approach of planning flight routes above areas with high One individual is one 3D-flight route and all individuals together
noise levels. This leads to less added noise in quiet residential areas. form the population. The representation of the individuals is a sequence
We compute the added noise perpendicular from the aircraft to the of georeferenced 3D-points (Fig. 1). Each point consists of the 𝑥 and 𝑦
surface. First, the sound pressure arriving at the ground is calculated coordinates and the height above the ground in meters. Additionally,
with the noise emission of the aircraft (Kinsler, 2000). As the noise each route consists of a smoothed line that connects the 3D-points
emission values are not made public by the aircraft manufacturers, (Fig. 1). The line is needed to validate a route because straight connec-
the target noise levels of Uber Elevate (2016) are used. Their goal is tions between the points can intersect with the restricted air space, even
a maximum of ‘‘62 dB at 500 ft altitude’’. This translates to a noise if the points are positioned outside of the restricted air space. The array
pressure of 100 dB in a reference distance of one meter. Using the flight representation, the 3D-points and the 3D-line are intertwined (Fig. 1). If

4
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

one of these three representations of one individual changes, the other differ per solution, so the crossover algorithm is slightly adapted (see
two representations are updated accordingly. Appendix C). The used mutation technique for the optimization of
The initial population is formed using a seeding approach (Friedrich 3D-flight routes bases on the idea of Peng et al. (2011). They split
and Wagner, 2015). That means that prior knowledge of the problem a mutation process into deletion, insertion and disturbance of points.
and its solutions exists and is used for generating the initial popula- These processes were slightly adapted to increase the probability of
tion for an evolutionary algorithm. Seeding is considered especially generating valid solutions: The insertion and deletion of points is done
useful when the initialization and evaluation of the solutions are ex- deterministically instead of randomly. For the insertion, points are
pensive (Friedrich and Wagner, 2015). If we would make the initial sought with the highest distance to the neighbouring points. For the
population completely random, a very large part of the population deletion, points are sought with the lowest distance to the neighbouring
would not connect start and end point, be very long, and insect with points. The reason for this approach is the high proportion of restricted
the restricted airspace. Therefore, we use the existing knowledge of airspace. Higher distances between the points increase the probability
of a line intersecting with the restricted and protected airspace, which
the least-cost path (Section 3.2, Fig. 2), as that path is known to
makes a route invalid.
efficiently connect the start and end point and does not interfere with
the restricted airspace. The initial population is formed by individuals
3.6.5. Selection of the next generation
randomly created within an 1800-meter-wide 3D channel around the
After producing and evaluating the children in every generation, the
least-cost path. The height of the channel depends on the distance from non-domination rank and the crowding distance are reused for selecting
the minimal height to the maximal height at each position. If a point is the individuals that form the next generation. In the NSGA II, the
created in the restricted airspace, the point is deleted. The initialization individuals that replace the old population are selected from the parents
of individuals is repeated until the population size of 16 is reached. and children. The solutions with the best ranks are selected until the
population size of 16 is reached again. By doing test of convergence,
3.6.2. Validation and GIS repair processes we selected a number of 20 produced generations as the termination
An individual is invalid when it violates one of the defined con- criterion to stop the optimization.
straints. In our case, the following conditions need to be fulfilled for
an individual to be valid: (1) No 3D-line segment intersects with the 3.6.6. Data and software availability
minimal flight height plane, and (2) No 3D-line segment intersects with In this project, the Python libraries of the licensed GIS desktop appli-
the restricted and protected airspace. cation ArcGIS Pro 2.4.11 are used for processing and storing the spatial
The covered area of restricted and protected airspace is ≈ 41% of data. The optimization algorithm is developed in Python 3.68.2 The
the study area, and the flight route is ≈19 kilometres long. In case of repository containing the optimization implementation (Hildemann,
a constraint violation at any point of the optimization, it is inefficient 2020) and the required spatial data (Hildemann and Delgado, 2020)
to discard the individual, because the computation effort for producing are permanently available at Mendeley Data.
a solution is high and the probability of violating a constraint is high.
4. Results
Therefore, we first try to ‘repair’ the individual. A 3D point is relocated
when it is inside a restricted or protected air space. When a line
4.1. The extremes of the Pareto front
interferes with a restricted or protected air space, we add points to
make the line circumvent the restricted or protected air space (see
The flight routes at the Pareto front of the Lilium jet have a flight
Appendix A for more details). If the individual is still invalid after a time between 221 and 261 s, an energy consumption between 13.4 and
repair trial, it is removed. 30.6 kWh, and added noise between 3.61 and 4.56 dB (Fig. 3). On the
other hand, the flight routes at the Pareto front of the EHANG 184 have
3.6.3. Evaluation and ranking of the individuals a flight time between 222 and 254 s, an energy consumption between
All valid individuals are evaluated by applying the objective func- 3.76 and 6.73 kWh, and added noise between 3.62 and 4.77 dB.
tions for the flight time, the energy consumption and the noise addition. The extremes of the Pareto front (red, blue and green dot in Fig. 3),
As such, each individual is assigned three values that express its qual- represent the solutions with all weight on a single objective, i.e. the
ity. These values are used to compute a non-domination rank, which route with the shortest flight time, the route with the lowest energy
was introduced and explained by Deb et al. (2002). All solutions not consumption and the route with the lowest added noise. These three
dominated by any other solution are assigned the first rank. The second routes are found to resemble each other closely at the start and end
rank is assigned to solutions that are not dominated by any solution not point of the route for both the EHANG 184 (Fig. 4a) and the Lilium
belonging to the first rank. This process continues until every solution (Fig. 4b). In the middle of the route, with several scattered restricted
got assigned a non-domination rank. or protected air spaces, the differences are larger. For both eVTOL
types, the route with the shortest flight time goes through a corridor
3.6.4. Selection, crossover and mutation of individuals between two small restricted air spaces, while both least-energy routes
The selection of the individuals for the crossover and mutation op- bypass the two restricted or protected airspaces on the western side.
erations is done with a tournament selection (Sivanandam and Deepa, The routes with the least added noise have in common that they are
2008). From the population, a subset of four solutions is randomly composed of long route segments arranged in parallel with the main
selected, the Tournament Group. From the Tournament Group, the street on the north western area of the map, because the added noise
individual with the lowest non-domination rank (i.e., the best score) from the aircraft is lower, where the transportation noise is louder. The
wins (Deb et al., 2002). If two solutions have the same non-domination other routes are straighter and not aligned with the road. Following the
roads is often a detour, which is time- and energy consuming, so this
rank, an additional metric is used to choose the superior solution, the
characteristic is unique for the least added noise flight routes.
crowding distance. The crowding distance is the distance to neighbour-
The total flight distances in the height profiles (Fig. 4c, d) show,
ing points in the objective space (Deb et al., 2002). The solution with
that the total travel distances of the least added noise routes are the
the highest crowding distance wins to maintain the maximum diversity
longest of the three different objectives for both eVTOL types, even
in the population. The Tournament Winner proceeds to the crossover
if the distances are marginally greater compared to the shortest flight
and mutation.
time routes.
For the crossover, the genes of two Tournament Winners from the
selection process are combined with a k-point crossover (Sivanandam
and Deepa, 2008). The produced children inherit the combined el- 1
https://www.esri.com/de-de/arcgis/products/arcgis-pro/resources
2
ements of the parents. In our approach, the number of points can https://www.python.org/downloads/release/python-368/

5
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 3. Objective values of all generated non-dominated solutions in optimization, (a) aircraft EHANG 184, (b) Lilium jet. The grey-scale refers to the generation the solutions
were produced. The extreme solutions for each objective are highlighted and the reference point of the least cost path is added. (For interpretation of the references to colour in
this figure legend, the reader is referred to the web version of this article.)

Fig. 4. Mapped 3D-routes for optimal solutions per objective, for eVTOL (a) EHANG 184 and (b) Lilium jet. The height profiles for all routes are illustrated for eVTOL (c) EHANG
184 and (d) Lilium jet.

6
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 5. Pairwise density distributions (diagonal), contour plots (lower-left) and scatter plots with correlation coefficients (upper-right) for all three objectives, for (a) the EHANG
184 and (b) the Lilium Jet.

4.2. Comparison of the extremes of the Pareto front with the least-cost path these objectives with added noise are highly negative (Fig. 5). This
implies that lower added noise comes with longer flight times and more
The first part of the first research question ‘‘Can all objectives be energy consumption. A linear regression of the final Pareto front shows
optimized while complying with the air space restrictions and noise that, on average, a decrease of 1 s of flight time comes with a reduction
regulations?’’ can be answered with a yes. The best solutions, compared in energy consumption of 0.48 kWh for the Lilium jet and 0.1 kWh for
to the initial least cost path, were improved for both the EHANG 184 the EHANG 184. A decrease of 1 kWh therefore comes with a decrease
and the Lilium jet for all three objectives (Fig. 2). As this part can of 2.08 s flight time for the Lilium jet and 9.09 s for the EHANG 184.
be answered with yes, the second part of the research question is An increase in added noise of 1 dB comes with an additional flight time
addressed: How much can a least-distance route computed with a GIS of 16 s for the Lilium jet and 26.9 s for the EHANG 184. In parallel to
be improved for the objectives in the highly restricted air space of New that, a decrease in flight time of one second comes with an additional
York City for two different eVTOL? Comparing the three extremes of the noise of 0.06 dB for the Lilium and 0.037 dB for the EHANG 184. A
Pareto front to the least-cost path (Fig. 2), the flight time has decreased decrease of 1 kWh of energy consumption comes with an additional
from 331 s to 115 s (65%), the energy consumption has decreased noise level of 0.14 dB with the Lilium jet and 0.4 dB with the EHANG
from 4.93 kWh to 2.38 kWh (52%), and the added noise has decreased 184, which equals an additional 6.99 kWh (Lilium) and 2.48 kWh for
from 9.19 dB to 5.74 dB (38%) for the EHANG 184. For the Lilium jet, every reduced noise level of 1 dB.
the flight time has decreased from 220 s to 116.8 s (47%), the energy The reason for the positive correlation between the objectives en-
consumption has decreased from 9.99 kWh to 5.25 kWh (47%), and the ergy consumption and flight time is the common component of the
added noise has decreased from 9.19 dB to 5.28 dB (43%). However, flight distance, which appears in the computation of both the flight time
note that these three improvements cannot all be reached at the same and energy consumption. In general, higher flight distances cause worse
time. In other words, there are trade-offs between the objectives. fitness values in energy and flight time together. On the other hand, the
only distance in the objective function of the least added noise is the
4.3. Trade-offs between the three objectives distance to the ground. A larger distance to the ground opposes short
flight times and low energy consumption because the eVTOL is required
The energy consumption values of the Pareto front have a relatively to travel further.
narrow range for the EHANG 184 with values from 2.4 kWh to 12 kWh, The frequency distributions of non-dominated solutions differ per
compared to 5.3 kWh to 27 kWh for the Lilium jet. Similar behaviour aircraft type (Fig. 5, panels at the diagonal). While the frequency
can be observed for the objective shortest flight time. Only the value distribution for the EHANG 184 is evenly spread, the non-dominated
ranges for the added noise are similar between the aircraft types. As the solutions for the Lilium jet are partitioned into two clusters (Fig. 5b).
optimization procedure is the same, this indicates a higher sensitivity One cluster has high flight times, high energy consumption and low
to changes of the flight route structure for the Lilium jet: changes to noise levels, the second cluster the opposite. There are no trade-off
the flight route structure of a certain magnitude have higher impacts solutions in between these clusters, implying that there are no flight
on the route quality for the Lilium jet. routes with medium performance for all three objectives.
Our second research question was: What information is gained The eVTOL characteristics in the objective functions explain these
by the study case results? For both eVTOL aircraft types the energy clusters. Sharp turns force both aircraft types to slow down, which
consumption and the flight time of the optimal solutions are highly increases the flight time. But, for the Lilium jet only, sharp turns
positively correlated, whereas the correlation coefficients of both of also increase energy consumption. The Lilium Jet can produce lift

7
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

when travelling with higher flight speeds, which decreases energy noise are more difficult to find, because they are far away from the
consumption. The energy consumption in the hovering, acceleration current position, not clustered, and in different directions from the
and deceleration is higher than in cruise mode (Bacchini and Cestino, current position. The best position for the lowest added noise is the
2019). The energy consumption is therefore affected by route segments corner at the bottom right of Fig. 6. The best approach to get to this
that have sharp turns, which forces the aircraft to shift from cruise solution would be a totally random positioning, so the solution can only
mode to deceleration or hover mode. The required power increases be found by chance. An adaptation for making the search easier to find
from 36 to 187 kW, if the flight speed is 100 km/h. This does not apply the noisy areas can be done by smoothing the input noise data. Filtering
to the multicoptor aircraft EHANG 184, as its energy consumption is the processes, for example a Gaussian smoothing (Longley et al., 2015), can
same in hover, acceleration, and deceleration. And with an increase in cause the raster file to produce a more continuous search space, but at
energy consumption from 34.6 to 42.7 kW, it is a smaller difference to the same time this would make the detected solutions less precise.
shift from cruise mode to the other modes. Therefore, the EHANG 184
is not as affected by sharp turns along the route as the Lilium jet. 5.2. Implications for air space management
The flight routes that produce the lowest added noise follow the
noisy parts of the city, which are mostly highly trafficked streets. Street If a goal of air space management is that the yearly average back-
networks in general tend to have many sharp turns. This also means ground noise change shall not increase by more than 1 dB (Uber
that the best flight routes for the lowest added noise tend to have a Elevate, 2016; Berglund et al., 1999) (because this the smallest change
higher negative impact on the lift-generating aircraft types than on the that can be detected by a person), our results show that this goal cannot
multicopter aircraft. These opposing requirements for good solutions be reached. The added noise we computed, even for the route with the
for the different objectives explain the two observed clusters for the lowest added noise at the extreme of the Pareto front, has an average
Lilium jet. added noise of ≈ 5 dB. A potential solution would be vary the route
The same flight characteristics also explain, why the flight routes over the course of the day, adapting it to traffic, or to select a less
for the Lilium jet were observed to have higher value ranges in the ob- noisy aircraft than the ones evaluated in this work. It is also advisable,
jectives shortest flight time and least energy consumption (Fig. 4b). The to make the objective noise more important for flight routes meant
initial points of the flight routes are random and have many sharp turns
for night-time flights and possibly to include residential areas into the
because connecting these random points leads to many sharp direction
no-fly zones at night.
changes. Each of these turns has negative impacts on the objectives
One important consideration in our method with respect to flight
shortest flight time and least energy consumption. But, as discussed
management is the level of detail of the flight route representation.
above, the negative impact is higher for lift-generating aircraft. The
The result of our current optimization is a 3D-line with an accuracy
sharp turns are likely to be smoothed out from generation to generation
of less than a centimetre. But in reality, aircraft are not able to follow
because the idea of the evolutionary algorithm is to produce better
the line with such precision, even if the maximum speeds are preserved.
children than the parents. The positive impacts of removing sharp turns
Wind gusts or minor piloting errors, just as two examples, can lead to
are higher with the same reasoning. This sensitivity is also underlined
a misallocation of the aircraft from the flight route. Therefore, if this
by the higher improvement rates of optimizing the flight routes for the
work is to be used for air space management it is advisable to compute
Lilium jet.
the objective values on 3D pipes around the flight route lines. One
possibility for this transformation is adding three different boundaries
5. Discussion
around the line, which can be equivalent to a geofence safety boundary
management developed by NASA (Dill et al., 2016).
5.1. Algorithm performance

In Section 4.1, it was shown that the algorithm managed to find bet- 5.3. Generalizability and limitations
ter solutions than the least-cost path. The quality of the non-dominated
solutions improved over the generations, and the improvements be- In this study, we focus on computing flight routes for air taxis
came smaller with each generation: The average improvements from only. However, other airspace users than air taxis will populate the
the first to the second generation were 13.9% (Lilium) and 5.3% urban airspace e.g. drones operating in the lower airspace (Merkert and
(EHANG), but only 1.4% (Lilium) and 0.0% (EHANG) in the last gener- Bushell, 2020). This yields the risk of collisions and airspace conges-
ation (Fig. 3). This is a sign of the desired outcome, that the solutions tion. For this reason, airspace management systems will be required
converge to an optimum. Furthermore, the number of non-dominated that manage the operations of all airspace users. A solution to control
solutions increased over time. The last generation contained on average all airspace operations is a variable pricing strategy for the different
220% more non-dominated solutions than the first generation. The airspace users (Merkert et al., 2021). For temporal permissions to access
more non-dominated solutions are produced, the better we can derive specific flight corridors, variable charges can be utilized that may con-
the shape of the Pareto front, and the more options the decision maker sider various factors. Such factors can be, for example, the aircraft type,
has for the selection of a single solution (Deb and Chaudhuri, 2007). the maximum speed, the demand for operating at the desired flight
The convergence behaviour of the evolutionary algorithm was found height, but also externalities such as added noise or added risk. The
to differ per aircraft, with smaller improvements per generation for the variable charges can be considered in flight route planning by defining
EHANG (Fig. 3). an alternative objective function that quantifies the total charges per air
The relative flight route quality improvement in the objective of taxi operation. Since some factors change over time, e.g. the demand
least added noise was lower compared to the relative quality improve- or the added noise, this strategy would result in temporally changing
ment in the other two objectives for both aircraft types. This can optimal flight routes. A flight route optimization that considers all
be explained by the difference in the solution-space complexity for airspace users yields many challenges and thus potential for future
each objective, which we have illustrated with an example in Fig. 6. research.
For an easier understanding, we remove the height dimension from In this work, the objective functions are approximations of the real
the objective functions and only consider two-dimensional position flight time, energy consumption, and added noise. We considered the
changes. For the energy and flight time, the positions that improve the added noise directly underneath the aircraft’s position. In future work,
fitness, because the course of the flight route straightens (green marked the added noise to the neighbouring areas could be computed too.
area), are close to the current position of the point and clustered Furthermore, it is not clear how noisy the aircraft are in reality, because
together. On the other hand, positions that improve the minimal added the air-taxi manufacturers did not supply any information about the

8
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 6. Illustration of solution-space complexity. Green area displays regions of better positions compared to the current yellow point for (a) the objectives least energy consumption
and shortest flight time, and (b) the objective least added noise. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of
this article.)

sound pressures or about the sound frequencies. This is a limitation consumption by 47% from 9.99 kWh to 5.25 kWh, and the added noise
that cannot be solved without the cooperation of the manufacturers. decreased by 43% from 9.19 dB to 5.28 dB.
Besides, the impact of wind can be considered. Depending on the Our second research question was: What information for air space
wind direction and wind speed, the wind conditions can change rapidly management can be derived from the study case results? The produced
and can be distributed unevenly in the city area (Bornstein and John- non-dominated solutions showed high positive correlations between the
son, 1977). High wind speeds from specific directions can lead to objectives flight time and energy consumption, while they are both
differences in energy consumption for different flight directions, un- highly negatively correlated to the objective of least added noise. One
equal propagation of noise, as well as in dangerous areas, which would lesson learned was that lift-generating aircraft are more sensitive to the
need to be modelled in time-dependent no-fly zones (Stevens and flight route characteristics, such as bends, while more trade-off solu-
Atkins, 2019; Geister, 2017). tions exist for multicoptor aircraft. Another information gain regards
The origin and destination chosen in Manhattan, New York, was one the hardness of finding an optimum for the objective of least added
of many possible ones. To demonstrate the robustness of our results, we noise. It is easier to find better solutions for lower energy consumption
selected a second origin–destination pair, computed a second least-cost and shorter flight times because the fitness landscape is smoother with
path with a length of 18 km, and redid the optimization. The results fewer local optima.
are displayed in Appendix D (Fig. 8 and Fig. 9). For any different Another thing we can conclude is that without changing the flight
path in the same study area, only the input least-cost path needs to routes over time it is not possible to comply with the noise regula-
be changed. For the application of 3D-flight route optimization in a tions not to increase the long-term noise by more than 1 dB. The
different study area, the method is generally applicable if the data average noise additions were more than 5 dB, which is considered an
requirements are met. Information to compute the restricted flight areas unacceptable long-term annoyance.
needs to be available, as well as the surface with the minimal flight
CRediT authorship contribution statement
height. Suppose the flight route optimization is intended to be applied
to different aircraft. In that case, the only required adaptation is to alter
Moritz Hildemann: Conceptualization, Methodology, Software,
the aircraft specifications.
Writing – original draft, Visualization. Judith A. Verstegen:
Supervision, Conceptualization, Methodology, Writing – review &
6. Conclusion
editing.

In this work, we proposed a 3D-route optimization method with Declaration of competing interest
novel GIS procedures like seeding and repair. The method was applied
to a case study in Manhattan in New York for two different eVTOL The authors declare that they have no known competing finan-
Lilium Jet and EHANG 184. The proposed method is also applicable cial interests or personal relationships that could have appeared to
to other urban areas, different aircraft types and other 3D-routing influence the work reported in this paper.
applications. Appendix A. Repair geo-processes
Our first research question was: Can all objectives be optimized
while complying with the air space restrictions and noise regulations? If a 3D-point lies within a restricted or protected air space, ad-
If yes, how much can least distance flight routes computed with a GIS ditional reference points are created before the optimization process
be improved for multiple objectives in the highly restricted air space of begins. The reference points were created in 10-meter distance from
New York City for two different eVTOL? Our results confirmed the first the border lines of the restricted or protected air space. All 3D-points,
part of this question. Compared to the least cost path (Fig. 2), the flight that are positioned in the restricted air space, are relocated to the x-
time has decreased by 65% from 331 s to 115 s, the energy consumption and y-coordinate of the nearest reference point as illustrated in Fig. 7
by 52% from 4.93 kWh to 2.38 kWh, and the added noise decreased by (left). The line repair function repairs intersections of 3D-lines with the
38% from 9.19 dB to 5.74 dB for the EHANG 184. For the Lilium jet, restricted airspace. Even if the 3D-points are not positioned in the 3-
the flight time has decreased by 47% from 220 s to 116.8 s, the energy dimensional restricted or protected air space after the point relocation,

9
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Algorithm 1 Line repair

1: [𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠] = [𝑃 𝑜𝑖𝑛𝑡𝑠] ⊳ Points of flight route
2: 𝑖𝑝1, 𝑖𝑝2 ← 𝐼𝑛𝑡𝑒𝑟𝑠𝑒𝑐𝑡(𝐹 𝑙𝑖𝑔ℎ𝑡𝑅𝑜𝑢𝑡𝑒, 𝑅𝑒𝑠𝑡𝑟𝑖𝑐𝑡𝑒𝑑𝑜𝑟𝑃 𝑟𝑜𝑡𝑒𝑐𝑡𝑒𝑑𝐴𝑖𝑟𝑠𝑝𝑎𝑐𝑒) ⊳
Intersection points
3: 𝑏𝑝1, 𝑏𝑝2 ← 𝑁𝑒𝑎𝑟(𝑖𝑝1, 𝑃 𝑜𝑖𝑛𝑡𝐵𝑜𝑢𝑛𝑑𝑎𝑟𝑦) ⊳ Two nearest boundary
points
4: [𝑤𝑎𝑦1], [𝑤𝑎𝑦2] ← [𝑃 𝑜𝑖𝑛𝑡𝑠] ⊳ Initialize empty arrays
5: while 𝑖𝑝2 ∉ [𝑤𝑎𝑦1, 𝑤𝑎𝑦2] do ⊳ Stops if the second intersection
point is reached
6: 𝑛𝑝1 ← 𝑁𝑒𝑎𝑟(𝑏𝑝1)
7: if 𝑛𝑝1 ∉ [𝑤𝑎𝑦1, 𝑤𝑎𝑦2] then
8: 𝑏1 ← 𝑛𝑝1
9: [𝑤𝑎𝑦1].𝑎𝑑𝑑(𝑏𝑝1)
10: end if
11: 𝑛𝑝2 ← 𝑁𝑒𝑎𝑟(𝑏𝑝2)
12: if 𝑛𝑝2 ∉ [𝑤𝑎𝑦1, 𝑤𝑎𝑦2] then
13: 𝑏2 ← 𝑛𝑝2
Fig. 7. Illustration of repairing points and lines intersecting with restricted or protected 14: [𝑤𝑎𝑦2].𝑎𝑑𝑑(𝑏𝑝2)
air spaces.
15: end if
16: if [𝑤𝑎𝑦1].𝑙𝑎𝑠𝑡 = 𝑖𝑝2 then
17: [𝑅𝑜𝑢𝑡𝑒].𝑎𝑑𝑑([𝑤𝑎𝑦1])
the connecting line segments might do intersect with the restricted air 18: end if
space, as can be seen in Fig. 7 (right). 19: if [𝑤𝑎𝑦2].𝑙𝑎𝑠𝑡 = 𝑖𝑝2 then
At first, the intersection points of the 3-line with the 3D-shape are 20: [𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠].𝑎𝑑𝑑([𝑤𝑎𝑦2])
computed. From these intersection points, the nearest two neighbour- 21: end if
ing points from the restricted or protected air space point boundary 22: end while
are added to two arrays. For each point, the nearest neighbour will 23: [𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠] ← 𝑅𝑒𝑜𝑟𝑑𝑒𝑟([𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠])
be added to the array. Now, for every last point of each array, the 24: 𝐿𝑖𝑛𝑒 ← 𝑃 𝑜𝑖𝑛𝑡𝑠𝑇 𝑜𝐿𝑖𝑛𝑒([𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠])
nearest neighbour point that does not yet exist in any of the two 25: return [𝑅𝑜𝑢𝑡𝑒𝑃 𝑜𝑖𝑛𝑡𝑠], 𝐿𝑖𝑛𝑒 ⊳ Returns flight route with new
arrays is added. This procedure is repeated until the second intersection elements
point is added to one of the two arrays. The points in the arrays are
inserted into the point representation. In this way, the restricted or
protected air space is surrounded in both directions and the shorter 2
𝑎𝑐 Centripetal acceleration in ms
way is used to be inserted to the point representation. If the 3D-
point representation is now converted to a line again by connecting 𝑣 Velocity in ms
the points, the line is not intersecting with the restricted or protected 𝑟 Circle radius in m
air space anymore. The repaired line is illustrated in Fig. 7 (right, 𝑔𝐹 𝑜𝑟𝑐𝑒 Gravitational force equivalent
Relocated 3D-Line). In order to achieve this point-to-line conversion 𝑔𝐹 𝑜𝑟𝑐𝑒𝑀𝑎𝑥 Maximum allowed g-force
2
without crossings, the points need to be in the correct order. This is 𝑔𝑚 Gravitational acceleration of 9.81 ms
achieved by reordering all 3D-points after all point and line repairs, 𝑣𝑚𝑎𝑥 Maximum allowed speed at the given radius and
from the starting point to the end node. After this, the old 3D-line maximum g-force
representation is replaced by the repaired 3D-lines. All changes are
The output is the maximal flight speed at each position, which is
propagated to all three representations of the solution.
the minimum value of the maximal allowed legal flight speed, the
maximum speed of the aircraft and the maximum speed to have a lower
Appendix B. Cost function computations
𝑔𝐹 𝑜𝑟𝑐𝑒 than 1.25.

B.2. Flight time computation

B.1. Maximum flight speed

The total flight time with the specified flight constraints is the
To compute the maximum flight speed, the maximum flight speed
added-up flight time of all line segments:
for each position is computed with its centripetal forces. As the flight
route is no single circle, the centripetal force changes at each point ∑𝑛
𝑑𝑖
of the route. In order to compute the centripetal acceleration at each 𝑡𝑡𝑜𝑡𝑎𝑙 = (4)
𝑣
𝑖=1 𝑖
position, an imaginary circle is computed. This imaginary circle is
constructed at every point with its two neighbour points. The radius where
of the imaginary circle is used to compute the centripetal acceleration 𝑡𝑡𝑜𝑡𝑎𝑙 Total flight time
𝑎𝑐 and the resulting maximal velocity 𝑣𝑚𝑎𝑥 for the maximum 𝑔𝐹 𝑜𝑟𝑐𝑒 of 𝑛 Number of line segments
1.25. The formulas for the computation are expressed in Eqs. (1)–(3). 𝑑𝑖 Euclidean distance from the current point to next point
𝑣2 𝑣𝑖 Flight velocity at current segment
𝑎𝑐 = (1)
𝑟 To calculate the flight time of each segment, we use the distances
𝑎 between the points and actual flight speeds. If the aircraft is slower than
𝑔𝐹 𝑜𝑟𝑐𝑒 = 𝑐 (2)
𝑔𝑚 the allowed flight speed at the next position, the aircraft accelerates
2
√ with 2 ms . If the allowed flight speed of the next position is lower
𝑣𝑚𝑎𝑥 = 𝑔𝐹 𝑜𝑟𝑐𝑒𝑀𝑎𝑥 ⋅ 𝑟 ⋅ 𝑔𝑚 (3) 2
than the current flight speed, the aircraft decelerates with −2 ms .
where Furthermore, it needs to be assured, that the distance to the next point

10
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 8. Mapped 3D-routes for optimal solutions per objective, for eVTOL (a) EHANG 184 and (b) Lilium jet.

is sufficiently big in order to decelerate to the allowed flight speed. If 𝐿𝐶 Required lift
the distance is not long enough, the flight speeds of the previous points 𝑊 Weight of aeroplane in N
are readjusted. 𝛾 Angle of climb/descent in ◦
𝐿𝐶𝐿 Lift coefficient in climb
𝑆 Wing area of aircraft in m2
B.3. Energy consumption computation 𝑉 Flight speed in ms
kg
𝑝 Min. Air density m3
Now, with the variables flight velocity and the flight angle, the 𝐶𝐷𝐶 Drag coefficient in climb
𝐾𝐶 Drag factor
required power can be calculated for each flight route segment by
𝐶𝐷0 Total zero-lift drag
following the calculus of Eq. (5)–(11) (Sadraey, 2009). The velocity is
𝐷𝐶 Drag in climb in N
derived from the previous step and the ascending/descending angle is
𝑇𝑅𝐶 Thrust required in N
computed for each point to the next point. By summing up the required
𝑃𝑅𝐶 Power required in climb in kW
power of all segments, the desired objective function for getting the
𝑃𝑅𝐶𝑡𝑜𝑡𝑎𝑙 Total required power of flight route
total energy consumption in kWh is complete (Eq. (12)). 𝑛 Number of line segments
𝐿𝐶 = 𝑊 ∗ 𝑐𝑜𝑠(𝛾) (5) 𝑑𝑖 Euclidean distance from current line segment i
𝑣𝑖 Flight velocity at current segment i
𝐿𝐶 B.4. Noise addition computation
𝐿𝐶𝐿 = (6)
1 2
2
𝑝 𝑆
The first step is to calculate the sound pressure arriving at the
𝐶𝐷𝐶 = 𝐶𝐷0 + 𝐾 ∗ 𝐶𝐿2 (7) ground (Kinsler, 2000). For this calculation, the noise emission of the
aircraft is required. These values are not made public by the aircraft
1 2 manufacturers, therefore the targeted noise level of Uber Elevate is used
𝐷𝐶 = 𝑝𝑉 𝑆𝐶𝐷𝐶 (8) as a replacement. Uber Elevate (2016) is the only available source to
2
mention the noise level targets. Their goal is a maximum of ‘‘62 dB
at 500 ft altitude’’. This translates to a noise pressure of 100 dB in a
𝑇𝑅𝐶 = 𝑊 sin(𝛾) + 𝐷𝐶 (9)
reference distance of one meter. By filling the variable r in Eq. (13)
with the current flight height we get the perceived noise level at ground
𝑇𝑅𝐶 = 𝑊 sin(𝛾) + 𝐷𝐶 (10) from the aircraft.
The next step is to combine the noise emission by the aircraft with
𝑇𝑅𝐶 the background noise on the ground. The used data for the background
𝑃𝑅𝐶 = (11) noise is the average daily sound level from aviation and interstate road
1000
noise. The data was made available by the U.S. Department of Trans-
∑
𝑛
𝑑𝑖 portation (Bureau of Transportation Statistics, 2014). By combining the
𝑃𝑅𝐶𝑡𝑜𝑡𝑎𝑙 = 𝑃𝑅𝐶𝑖 ∗ (12) aircraft and background noise by inserting corresponding noise levels
𝑖=1
𝑣𝑖
at variable 𝐿𝑃 .𝑖 in Eq. (14), the total noise level 𝐿𝑃 .𝑇 𝑜𝑡 is computed.
where At this point, it is possible to determine how much noise the aircraft

11
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Fig. 9. Convergence behaviour, for eVTOL (a) EHANG 184 and (b) Lilium jet.

added at ground (Eq. (15)). In this equation can be observed, that the 4. The points of the children are reordered. All changes are propa-
noise addition is higher if the background noise is lower. Therefore, the gated to the array, point and line representations.
proposed objective function complies with the idea of preferring areas
with a high background noise.
Appendix D. Evaluation of a second origin–destination pair
𝑟
𝐿𝑃 = 𝐿𝑚 − 20 log10 (13)
𝑟𝑚
( ) See Figs. 8 and 9.
∑
𝐿𝑃 .𝑇 𝑜𝑡 = 10 log 100.1𝐿𝑃 .𝑖 (14)
𝑖 References
( )
𝐿𝑃 .𝑆𝑜𝑢𝑟𝑐𝑒 = 10 log 100.1𝐿𝑃 .𝑇 𝑜𝑡𝑎𝑙 − 100.1𝐿𝑃 .𝐵𝑎𝑐𝑘𝑔𝑟𝑜𝑢𝑛𝑑 (15)
Ahmed, S.S., Fountas, G., Eker, U., Still, S.E., Anastasopoulos, P.C., 2021. An ex-
where ploratory empirical analysis of willingness to hire and pay for flying taxis and
𝐿𝑃 Sound pressure at the grounda shared flying car services. J. Air Transp. Manag. 90, 101963. http://dx.doi.org/10.
1016/j.jairtraman.2020.101963.
𝑟 Distance to ground
Atila, U., Karas, I., Rahman, A., 2013. A 3D-GIS implementation for realizing 3D
𝑟𝑚 Reference distance (1 m from aircraft) network analysis and routing simulation for evacuation purpose. In: Pouliot, J.,
𝐿𝑚 Sound level at reference distance (100 dB) Daniel, S., Hubert, F., Zamyadi, A. (Eds.), Progress and New Trends in 3D
𝐿𝑃 .𝑖 Each noise from source 𝑖 measured at ground Geoinformation Sciences. In: Lecture Notes in Geoinformation and Cartography,
𝐿𝑃 .𝑇 𝑜𝑡 Combined noise pressure from background and aircraft noise vol. 29, Springer Berlin Heidelberg, Berlin, Heidelberg, pp. 249–260. http://dx.
doi.org/10.1007/978-3-642-29793-9-14.
𝐿𝑃 .𝑆𝑜𝑢𝑟𝑐𝑒 Added noise pressure from aircraft at ground
Bacchini, A., Cestino, E., 2019. Electric VTOL configurations comparison. Aerospace 6
a All sound pressure units are expressed in decibel (dB). (3), 26. http://dx.doi.org/10.3390/aerospace6030026.
Baur, S., Schickram, S., Homolenko, A., Martinez, N., Dyskin, A., 2018. In: Roland
The average added noise level is calculated by averaging all added
Berger (Ed.), Urban Air Mobility: The Rise of a New Mode of Transportation. In:
noise levels at each point at the flight route. Focus, URL: https://www.rolandberger.com/publications/publication_pdf/Roland_
Berger_Urban_Air_Mobility.pdf.
Appendix C. Adaptation of the n-point crossover Bauranov, A., Rakas, J., 2021. Designing airspace for urban air mobility: A review
of concepts and approaches. Prog. Aerosp. Sci. 125, 100726. http://dx.doi.org/10.
1016/j.paerosci.2021.100726.
1. The N randomly chosen positions are chosen in a range between Berglund, B., Lindvall, T., Schwela, D., 1999. Guidelines for Community Noise.
1 and the length of the shorter parent −1. The last position and Stockholm University.
first position are not exchanged as they are always the starting Bornstein, R.D., Johnson, D.S., 1977. Urban-rural wind velocity differences. Atmos.
and ending point of the flight route. Environ. (1967) 11 (7), 597–604. http://dx.doi.org/10.1016/0004-6981(77)90112-
3.
2. For each recombination point N of the first parent, the nearest
Budde, D., Hinkelbein, J., Boyd, D.D., 2021. Analysis of air taxi accidents (20042018)
flight route point of the second parent is sought with a GIS Near and associated human factors by aircraft performance class. Aerosp. Med. Hum.
Point Analysis from ArcGIS Pro. These points are the beginning Perform. 92 (5), 294–302. http://dx.doi.org/10.3357/AMHP.5799.2021.
points of the point exchanges. Bureau of Transportation Statistics, 2014. National transportation noise map. URL:
3. With both recombination positions of the parents, the children https://www.bts.gov/newsroom/bts-releases-national-transportation-noise-map.
are produced by exchanging the parent segments. Up to the first Dai, W., Low, K.H., 2022. Conflict-free trajectory planning for urban air mobility based
on an airspace-resource-centric approach. In: AIAA Aviation 2022 Forum. American
recombination point, children 1 gets the segments of parent 1,
Institute of Aeronautics and Astronautics, Reston, Virginia, http://dx.doi.org/10.
and children 2 gets the segments of parent 1. After the first 2514/6.2022-4077.
recombination point children 1 gets the segments of parent 2 de Paiva Vianna, K.M., Cardoso, M.R.A., Rodrigues, R.M.C., 2015. Noise pollution
and children 2 gets the segments of parent 2. This changes at and annoyance: an urban soundscapes study. Noise & Health 17 (76), 125–133.
each recombination point N. http://dx.doi.org/10.4103/1463-1741.155833.

12
M. Hildemann and J.A. Verstegen Journal of Air Transport Management 107 (2023) 102356

Deb, K., Chaudhuri, S., 2007. I-MODE: An interactive multi-objective optimization and McKinsey & Company, 2021. Study on the societal acceptance of urban air mobility in
decision-making using evolutionary methods. In: Obayashi, S., Deb, K., Poloni, C., Europe. URL: https://www.easa.europa.eu/sites/default/files/dfu/uam-full-report.
Hiroyasu, T., Murata, T. (Eds.), Evolutionary Multi-Criterion Optimization. In: pdf.
Lecture Notes in Computer Science, vol. 4403, Springer Berlin Heidelberg, Berlin, Merkert, R., Beck, M.J., Bushell, J., 2021. Will it fly? Adoption of the road pricing
Heidelberg, pp. 788–802. http://dx.doi.org/10.1007/978-3-540-70928-2_59. framework to manage drone use of airspace. Transp. Res. A 150, 156–170. http:
Deb, K., Pratap, A., Agarwal, S., Meyarivan, T., 2002. A fast and elitist multiobjective //dx.doi.org/10.1016/j.tra.2021.06.001.
genetic algorithm: NSGA-II. IEEE Trans. Evol. Comput. 6 (2), 182–197. http: Merkert, R., Bushell, J., 2020. Managing the drone revolution: A systematic literature
//dx.doi.org/10.1109/4235.996017. review into the current use of airborne drones and future strategic directions for
Dill, E.T., Young, S.D., Hayhurst, K.J. (Eds.), 2016. SAFEGUARD: An assured safety net their effective control. J. Air Transp. Manag. 89, 101929. http://dx.doi.org/10.
technology for UAS, thirty fifth ed. DASC, In: 2016 IEEE/AIAA 35th Digital Avionics 1016/j.jairtraman.2020.101929.
Systems Conference, IEEEC, http://dx.doi.org/10.1109/DASC.2016.7778009. Okuniek, J.N., Gerdes, I., Jakobi, J., Ludwig, T., Hooey, B.L., Foyle, D., Jung, Y.C.,
Durmaz, A., Ünal, E., Aydın, C., 2019. Automatic pipeline route design with multi- Zhu, Z., 2016. A concept of operations for trajectory-based taxi operations. In:
criteria evaluation based on least-cost path analysis and line-based cartographic 16th AIAA Aviation Technology 2016. http://dx.doi.org/10.2514/6.2016-3753.
simplification: A case study of the mus project in Turkey. ISPRS Int. J. Geo-Inf. 8 Open Data NYC, 2018. Open data of New York City: New York City population
(4), 173. http://dx.doi.org/10.3390/ijgi8040173. by neighborhood tabulation areas. In: NYC.gov (Ed.), New York, URL: https:
EHANG, 2019. Ehang 184 specifications. URL: http://www.ehang.com/ehang184/specs. //opendata.cityofnewyork.us/.
Eißfeldt, H., 2020. Sustainable urban air mobility supported with participatory noise OpenStreetMap contributors, 2019. https://planet.osm.org. URL: https://www.
sensing. Sustainability 12 (8), 3320. http://dx.doi.org/10.3390/su12083320. openstreetmap.org.
Electric VTOL News, 2019. Evtol Classifications. World eVTOL Directory, URL: https: Peng, Z., Wu, J., Chen, J., 2011. Three-dimensional multi-constraint route planning of
//evtol.news/classifications/. unmanned aerial vehicle low-altitude penetration based on coevolutionary multi-
Environmental Protection Agency, 1974. Information on levels of environmental noise agent genetic algorithm. J. Cent. South Univ. Technol. 18 (5), 1502–1508. http:
requisite to protect public health and welfare with an adequate margin of safety: //dx.doi.org/10.1007/s11771-011-0866-4.
EPA: U.S. URL: http://www.nonoise.org/library/levels74/levels74.htm. Pradeep, P., Chatterji, G.B., Lauderdale, T.A., Sheth, K., Lai, C.F., Erzberger, H.,
Federal Aviation Administration, 2017. Pilot’s Handbook of Aeronautical Knowledge: Sridhar, B., 2022. Wind-optimal lateral trajectories for a multirotor aircraft in urban
2016, First Skyhorse Publishing ed. Skyhorse Publishing, New York, NY. air mobility. Front. Aerosp. Eng. 1, http://dx.doi.org/10.3389/fpace.2022.1064142.
Federal Flight Administration, 2016. Digital obstacle file: Flight obstacle maps. URL: Pradeep, P., Lauderdale, T.A., Chatterji, G.B., Sheth, K., Lai, C.F., Sridhar, B., Ed-
https://www.faa.gov/air_traffic/flight_info/aeronav/digital_products/dof/. holm, K., Erzberger, H., 2020. Wind-optimal trajectories for multirotor eVTOL
Friedrich, T., Wagner, M., 2015. Seeding the initial population of multi-objective aircraft on UAM missions. In: AIAA Aviation Forum 2020. http://dx.doi.org/10.
evolutionary algorithms: A computational study. Appl. Soft Comput. 33, 223–230. 2514/6.2020-3271.
http://dx.doi.org/10.1016/j.asoc.2015.04.043. Rajendran, S., Pagel, E., 2020. Recommendations for emerging air taxi network
Garrow, L.A., German, B.J., Leonard, C.E., 2021. Urban air mobility: A compre- operations based on online review analysis of helicopter services. Heliyon 6 (12),
hensive review and comparative analysis with autonomous and electric ground e05581. http://dx.doi.org/10.1016/j.heliyon.2020.e05581.
transportation for informing future research. Transp. Res. C 132, 103377. http: Rajendran, S., Srinivas, S., 2020. Air taxi service for urban mobility: A critical review
//dx.doi.org/10.1016/j.trc.2021.103377. of recent developments, future challenges, and opportunities. Transp. Res. E 143,
Geister, D., 2017. Integrating UAS into the Future Aviation System: A Flexible Approach 102090. http://dx.doi.org/10.1016/j.tre.2020.102090.
Enabling Large-Scale UAS Operations: German Aerospace Center (DLR). Institute Reddy, M.J., Kumar, D.N., 2015. Elitist-mutated multi-objective particle swarm op-
of Flight Guidance, URL: http://www.dlr.de/fl/desktopdefault.aspx/tabid-11763/ timization for engineering design. In: Khosrow-Pour, D.B.A. (Ed.), Encyclopedia
20624_read-48305/. of Information Science and Technology, third ed. IGI Global, pp. 3534–3545.
Glaab, P., Wieland, F., Santos, M., Sharma, R., Tamburro, R., Lee, P.U., 2019. http://dx.doi.org/10.4018/978-1-4666-5888-2.ch346.
Simulating fleet noise for notional UAM vehicles and operations in New York. In: Rizzi, S.A., Huff, D.L., Boyd, D.D., Bent, P., Henderson, B.S., Pascioni, K.A., Sar-
2019 IEEE/AIAA 38th Digital Avionics Systems Conference. DASC, IEEE, pp. 1–10. gent, D.C., Josephson, D.L., Marsan, M., He, H.B., Snider, R., 2020. Urban air
http://dx.doi.org/10.1109/DASC43569.2019.9081670. mobility noise: Current practice,gaps, and recommendations. NASA Sci. Tech. Inf.
Goyal, R., 2018. Urban Air Mobility (UAM) market study: Presented to: National (STI).
Aeronautics and Space Administration - Aeronautics Research Mission Directorate. Sadraey, M., 2009. Aircraft Performance Analysis. VDM Verlag Dr. Müller, Saarbrücken,
In: Booz-Allen and Hamilton, Inc. (Ed.), In: Air Transportation and Safety, NASA Germany.
Headquarters and Washington, DC, United States, URL: https://ntrs.nasa.gov/ Sandurkar, S., Chen, W., 1999. GAPRUS - genetic algorithms based pipe routing
search.jsp?R=20190001472. using tessellated objects. Comput. Ind. 38 (3), 209–223. http://dx.doi.org/10.1016/
Hildemann, M., 2020. 3D flight route optimization for air-taxis in urban areas (Python S0166-3615(98)00130-4.
+ arcpy repository), Mendeley Data, http://dx.doi.org/10.17632/kzk3j8xhhm.1, Sivanandam, S., Deepa, S., 2008. Introduction to Genetic Algorithms. Springer, Berlin.
https://data.mendeley.com/datasets/kzk3j8xhhm. Stevens, M.N., Atkins, E.M., 2019. Generating airspace geofence boundary layers in
Hildemann, M., Delgado, C., 2019. An adaptable and scalable least cost network for wind. J. Aerosp. Inf. Syst. 1–12. http://dx.doi.org/10.2514/1.I010792.
air-taxis in urban areas study area: Manhattan, New York. In: Accepted Short Papers Sun, X., Wandelt, S., Stumpf, E., 2018. Competitiveness of on-demand air taxis
and Posters from the 22nd AGILE Conference on Geo-Information Science. AGILE. regarding door-to-door travel time: A race through Europe. Transp. Res. E 119,
Hildemann, M., Delgado, C., 2020. Spatial input data for computing air-taxi flight routes 1–18. http://dx.doi.org/10.1016/j.tre.2018.09.006.
in restricted and protected air spaces in New Yerk City (USA). http://dx.doi.org/ Uber Elevate, 2016. Fast-forwarding to a future of on-demand urban air transportation:
10.17632/b3n53j747h.1. White paper. URL: https://www.uber.com/us/en/elevate/.
Holland, J.H., 1975. Adaptation in Natural and Artificial Systems: An Introductory Wang, M., Diepolder, J., Zhang, S., Söpper, M., Holzapfel, F., 2021. Trajectory
Analysis with Applications to Biology, Control, and Artificial Intelligence. University optimization-based maneuverability assessment of eVTOL aircraft. Aerosp. Sci.
of Michigan Press, Ann Arbor, Michigan. Technol. 117, 106903. http://dx.doi.org/10.1016/j.ast.2021.106903.
Kinsler, L.E., 2000. Fundamentals of Acoustics, fourth ed. Wiley, New York and Ward, K.A., Winter, S.R., Cross, D.S., Robbins, J.M., Mehta, R., Doherty, S., Rice, S.,
Chichester, URL: https://www.who.int/occupational_health/publications/noise1. 2021. Safety systems, culture, and willingness to fly in autonomous air taxis:
pdf. A multi-study and mediation analysis. J. Air Transp. Manag. 91, 101975. http:
Lilium Aviation, 2019. Mission: Aircraft for everyone. URL: https://lilium.com/ //dx.doi.org/10.1016/j.jairtraman.2020.101975.
mission/. Zhao, N., Li, N., Sun, Y., Gao, Z., 2020. 4D trajectory planning of aircraft taxiing
Longley, P., Goodchild, M.F., Maguire, D., Rhind, D.W., 2015. Geographic Information considering time and fuel. Math. Probl. Eng. 2020, 1–12. http://dx.doi.org/10.
Science & Systems, fourth ed. John Wiley & Sons, Hoboken, New Jersey. 1155/2020/9603968.

13