operationalconsiderationsregardingon-demandairmobility ...odam has the po-tential to radically...

Review ArticleOperational Considerations regarding On-Demand Air Mobility:A Literature Review and Research Challenges

Xiaoqian Sun ,1,2 Sebastian Wandelt ,1,2 Michael Husemann,3 and Eike Stumpf3

1School of Electronic and Information Engineering, Beihang University, 100191 Beijing, China2National Engineering Laboratory for Multi-Modal Transportation Big Data, 100191 Beijing, China3Institute of Aerospace Systems, RWTH Aachen University, 52062 Aachen, Germany

Correspondence should be addressed to Sebastian Wandelt; [email protected]

Received 8 September 2019; Revised 25 November 2020; Accepted 22 January 2021; Published 5 February 2021

Academic Editor: Dongjoo Park

Copyright © 2021 Xiaoqian Sun et al. 0is is an open access article distributed under the Creative Commons Attribution License,which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

0e idea and development of on-demand air mobility (ODAM) services are revolutionizing our urban/regional transportationsector by exploring the third dimension: vertical airspace. 0e fundamental concept of on-demand air taxi operations is not new,but advances in aircraft design and battery/engine technology plus massive problems with congestion and increased traveldemands around the world have recently led to a large number of studies which aim to explore the potential benefits of ODAM.Unfortunately, given the lack of an established, formal problem definition, missing reference nomenclature for ODAM research,and a multitude of publication venues, the research development is not focused and, thus, does not tap the full potential of theworkforce engaged in this topic. 0is study synthesizes the recently published literature on operational aspects of ODAM. Ourcontribution consists of two major parts. 0e first part dissects previous studies and performs cross-comparison of report results.We cover five main categories: demand estimation methodology, infrastructure/port design/location problem, operationalplanning problem, operational constraints’ identification, and competitiveness with other transportation modes. 0e second partcomplements the report of aggregated findings by proposing a list of challenges as a future agenda for ODAM research. Mostimportantly, we see a need for a formal problem definition of ODAM operational planning processes, standard open datasets forcomparing multiple performance dimensions, and a universal, multimodal transportation demand model.

1. Introduction

0e rise of megacities/agglomerations together with a tre-mendous increase in travel demands leads to regulartransportation gridlocks in many urban areas all over theworld, e.g., in New York, Beijing, Sao Paolo, and Mumbai[1, 2]. Footage of congestions in these regions is well known;operators and governments try to find solutions towards abetter transportation future. 0e cost of congestion alone inthe US is quantified with 305 billion USD for the year 2016.Increasing the capacities further is not considered a viableoption in many cities any longer [3].

0e vision of opening the third dimension, i.e., altitude,for urban/regional transportation has gained substantialinterest in the last 5–10 years, with significant efforts toexplore the so-called on-demand air mobility (ODAM) [4].0e key idea behind ODAM is to use eVTOL (electric

vertical takeoff and landing) vehicles for inter- and intracitycommutes. 0ese small vehicles (usually anticipated to fly1–4 people) are envisioned to provide line-by-sight flights ataffordable prices, while providing a safe and enjoyable flightexperience. 0eir ability to perform vertical takeoff/landingenables them to be operated on small-size port infrastruc-ture. A successful implementation of ODAM requires notonly the development of new technologies, including vehicledesign and advances in engine/propulsion technology, butalso considerations of novel operational patterns, con-cerning its on-demand characteristic. ODAM has the po-tential to radically change our view of urban/regionalmobility and saves time in people’s daily commutes as well ason regional thin-haul connections [5, 6].

Industry and academia spend a large amount of re-sources on the development of ODAM. Uber, for instance, ispioneering to push mass adoption of VTOL services for

HindawiJournal of Advanced TransportationVolume 2021, Article ID 3591034, 20 pageshttps://doi.org/10.1155/2021/3591034

mailto:[email protected]://orcid.org/0000-0002-8713-142Xhttps://orcid.org/0000-0003-0088-9287https://creativecommons.org/licenses/by/4.0/https://creativecommons.org/licenses/by/4.0/https://doi.org/10.1155/2021/3591034

urban air transportation. See [7] for a detailed analysis onVTOL market feasibility across four dimensions: vehicle,infrastructure and operations, rider experience, and eco-nomics. Uber plans to launch the air taxi service in 2023,starting in Dallas, Texas, and Los Angeles, California [8].Lilium, a German startup company [9], has completed a testflight in 2017 and plans to provide a public air taxi service forcity transportation in 2025. Kitty Hawk, a flying-car startupcompany [10], also expects to bring air taxi services topassengers in 2023. Partnerships with the two major aircraftmanufacturer giants, Boeing and Airbus, further acceleratethe progress towards urban air mobility. Airbus is alsodeveloping its own air taxi City Airbus [11], which is em-bedded in the EU-initiative Urban Air Mobility. Boeingrecently acquired an aviation taxi developer, Aurora FlightSciences, which is developing a passenger VTOL vehicle.RWTHAachenUniversity develops a regional air taxi [12] incooperation with the University of Applied Sciences,Aachen, and the expected entry to service of the Silent AirTaxi is 2024. Future pandemics, such as COVID-19, couldbecome particular drivers of on-demand (personal) airmobility in the near future [13, 14]. While many companiesare racing to build flying prototypes, the scientific literatureon ODAM becomes fuzzy. 0ere is a rich body of researchfocusing on the technical feasibility of ODAM, such asvehicle design or propulsion system, with several recentreviews on ODAM vehicle design [15], current technologyfor electronic VTOL drones [16], distributed electronicpropulsion technology [17], and design of next-generationurban air mobility vehicles. Research on the operationalaspects of ODAM, however, is tremendously scattered, withmany studies published in different venues, using differentnomenclature, and investigating highly similar, yet distinctresearch problems.

In order to fill in the gap in the operational ODAM (Notethat our study subsumes urban air mobility (UAM), bydistinguishing urban ODAM from regional ODAM.) liter-ature, this study makes a twofold contribution. First, weprovide a comprehensive literature review on the opera-tional aspects of ODAM. We intentionally leave out bothtopics vehicle design and propulsion technology as there areexcellent surveys on these two topics, see above. Our reviewcovers five main categories: demand estimation methodol-ogy, infrastructure/port design/location problem, opera-tional planning problem, operational constraints’identification, and competitiveness with other trans-portation modes.0e second part of this study complementsthe report of aggregated findings by proposing a list ofchallenges as a future agenda for ODAM research. Onemajor challenge is to ensure reproducibility of previousODAM studies. We argue that there is an urgent need fordefining and agreeing upon concise research questions forODAM research, accompanied by benchmark/referencedatasets and methodology. 0e wide range of distinct resultson various regions in the world makes it necessary to de-velop a universal, multimodal ODAM framework, including

competition and cooperation with other modalities. Fur-thermore, we see a lot of potential for studies includinguncertainties into ODAM and regarding shared, profitableoperations.

0e remainder of this paper is organized as follows.Section 2 provides an overview of the state-of-the-art op-erational aspects of ODAM, followed by an analysis section.Section 3 discusses future research directions exclusively,and Section 4 concludes the study.

2. State-of-the-Art OperationalAspects of ODAM

0is section provides an overview of the operational aspectsof ODAM as reported in the literature. 0e overall goal is tosummarize/compare commonalities and differences amongprevious studies. Research on the operational aspects ofODAM, however, has appeared in many different venues,using different nomenclature and investigating highlysimilar, yet distinct research problems. 0is section aims toderive a consistent classification and unified terminology.Accordingly, we build a classification which is inspired bytraditional transportation systems wherever appropriate.Wecategorize the main results of research studies according tofive major categories: demand estimation methodology(Section 2.1), infrastructure/port design/location problem(Section 2.2), operational planning problem (Section 2.3),operational constraints’ identification (Section 2.4), andcompetitiveness with other transportation modes (Section2.5). Section 2.6 summarizes the major conclusions obtainedin this section and lays the foundations to identify oppor-tunities for the operational aspects of ODAM in Section 3.

2.1. ODAM Demand Estimation Methodology. As a newemerging transportation mode, demand is the fundamentaldriver for the development of ODAM, and it is critical forthe market orientation for ODAM operators at a strategiclevel. Table 1 provides an overview of the ODAM demandestimation, including the region under study, time frame,urban or regional scope, data usage and their availability,and a short summary. 0e studies are sorted by operationalscales (urban, regional, and urban + regional) and then byregion.

Regarding urban ODAM, we review the work for theregion in Europe and U. S. separately. Fu et al. [18] con-ducted a stated preference survey with 248 respondents fromMunich, Germany, investigating the mode choice behavioramong four transportation alternatives: public trans-portation, private car, autonomous taxi, and autonomous airtaxi. Results show that travel time, travel cost, and safety arethe critical determinants for the adoption of autonomoustransportation modes. Moreover, higher value of time andhigher income also favor the use of urban air mobility. For atypical European city, Decker et al. [19] used a substitutionrate of 10% of car traffic by a personal air vehicle, assuming

2 Journal of Advanced Transportation

Table 1: An overview of the ODAM demand estimation.

Ref. Region Time frame Scale Data type Dataavail. Short summary

[18] Munich, Germany (248respondents)

Feb.–Apr.2018 (survey

time)U Stated preference survey No

Travel time, travel cost, and safety arethe most critical determinants for the

adoption of autonomoustransportation modes. Moreover,

higher value of time and higher incomealso favor the use of urban air mobility.

[19] European cities Future U Scenario assumption No

For a typical European city, asubstitution rate of 10% of car traffic bya personal air vehicle is used, assumingthat daily car traffic is approximately300,000. Several requirements are noteasy to be met, and a broad range of

uncertainty remains.

[20] New York, U. S. Future U New York City taxi records No

Certain locations including largefacilities and smaller stops for New

York City are suggested. 0epercentage of time savings and

willingness to fly did not impact thelocation decision significantly, while

the on-road travel limit does.

[21]Northern California andWashington-Baltimore

Region, U. S.Future U

National household travelsurvey, LODES, ACS, and

StreetLight dataNo

0e demand is very sensitive to thepricing structure, and the costs have tobe kept between $1 per passenger mileand $1.25 per passenger mile to achievea potential market share of 0.5–4%,where a 4% market share represents

320,000 trips per day.

[22]

U. S. (2500 workers in 5cities: Atlanta, Boston,

Dallas, San Francisco, andLos Angeles

Apr.–Jun. 2018(survey time) U Stated preference survey No

No results from the survey wereprovided.

[23] U. S. (four focus groups) May 2017(survey time) U Stated preference survey No

Only survey for one focus group (6participants) was conducted with somedescriptive feedbacks. Results from twoonline platforms (Amazon MechanicalTurk and Qualtrics) were compared,and Qualtrics is recommended to use

in the future.

[24]

U. S. (1405 workers in 5cities: Atlanta, Boston,

Dallas, San Francisco, andLos Angeles

Mar.–May2019 (survey

time)U Stated preference survey No

0e survey settings are very similar to[22]. No results from the survey were

provided.

[25] Germany 2030 R 50 OD pairs’ samples inGermany No

With the ODAM door-to-door travelspeeds of 80–200 km/h, a willingness-to-pay value of 0.5–0.8€/km (monetaryvalue in 2015) for the year 2030 isderived for the German market.

[26] U. S. (3091 counties) 1995–2030 R DB1B and OAG Partial

More than 600 million on-demandsmall aircraft trips would compete withautomobiles in cost per passenger mile;such number of operations would have

negative impacts on the nationalairspace system, such as airport

congestion.

[27]A hypothetical network

with 5 nodes and 1–10 dailytrip demands

Future R Cirrus SR22 aircraft Yes

In order to maximize profit, expected57.7% of the daily demand should be

served/accepted, and an averageacceptance rate of approx. 90% mightstill allow for profitable operations.

Journal of Advanced Transportation 3

that daily car traffic is approximately 300,000.0e total rangefor this reference VTOL vehicle is 100 km, with maximumtwo seats and cruise speed 150–200 km/h. It was reportedthat several requirements are not easy to be met, and a broadrange of uncertainty remains. With New York City in theU. S. as a case study, Rajendran and Zack [20] estimated thepotential air taxi demand based on a subset of the currentregular taxi demand, i.e., approximately 300 million indi-vidual yellow and green taxi ride records from January 2014to December 2015. An air taxi service is eligible only if thetravel time of the air taxi is at least 40% shorter than that ofthe regular taxi, with the assumption that the maximum airtaxi travel distance is 120 miles (193 km) and with cruisespeed 170mph (273 km/h). With Northern California andWashington-Baltimore region as case studies, Syed et al. [21]proposed an integrated approach to estimate the ODAMdemand using the C-logit model, taking into account so-cioeconomic information, life cycle cost, fare structures,route constraints, and landing site requirements. Prelimi-nary results show that the demand is very sensitive to thepricing structure, and it is estimated that the ODAM costshave to be kept between $1 per passenger mile and $1.25 perpassenger mile to achieve a potential market share of0.5–4%, where a 4% market share represents 320,000 tripsper day. A 4-passenger-seat aircraft is subject to the as-sumption that an average of 2.4 passengers per aircraftwould generate 133,000 flights per day, even without takingreposition flights into consideration. Note that weatherconditions, public acceptance of a fully autonomous flyingvehicle, and its certification costs were not considered in thisstudy. In summary, existing work often assumes ODAM

demand comes from traditional ground transportation, andamong several uncertainty factors, travel time and travel costare the most critical for the profitability of ODAMoperations.

Furthermore, stated preference surveys were designed toask potential travelers in the U. S. which transportationmodes they would take under hypothetical situations. Binderet al. [22] presented details of a survey for estimatingcommuters’ willingness to pay for eVTOL flights in urbanareas in the U. S., collecting responses from approximately2,500 high-income workers with an average one-waycommuting time of 45 minutes or more by the individual.0ree typical modes (transit, car, and ride-share) arecompared with eVTOLs, where travel cost by typical groundtransportation modes ranges from $2.5 to $20, and the travelcost by eVTOLs ranges from $5 to $45, depending on thetravel distance. Maximum travel time is two hours for non-eVTOL modes. Access/egress/waiting times, transfer, andride guarantee availability of different modes are included.Personality, lifestyle characteristics, sociodemographics, andsocioeconomic information of the respondents are includedas well. Garrow et al. [23] used an online survey to identifyODAM demand segments and quantify willingness to payfor different ODAM alternatives; two online platforms,Amazon Mechanical Turk and Qualtrics, were used toconduct the surveys. Prior trip characteristics, traveleritinerary characteristics, and sociodemographics were alsoconsidered. 0ey also conducted a second survey with ap-proximately 100 questions [24], and the survey settings werevery similar to [22]. Four commuting modes, transit, tra-ditional car, self-driving car, and piloted air taxi, are

Table 1: Continued.

Ref. Region Time frame Scale Data type Dataavail. Short summary

[28] Germany Future U+R

Jeppesen airport databaseand census 2011 dataprovided by German

Federal Statistical Office

No

A market share of 19% or 235 milliontrips are estimated for Germany,

assuming passenger-specific costs of0.4€/km for ODAM services, 0.3€/km

for cars, and 0.32€/km for thecontemporary commercial CS-25

aircraft.

[29] Zurich, Switzerland Future U+R Simulation data No

Vehicle parameters (cruising speed,liftoff time, access time, and price) havea substantial impact on demand and

turnover.

[30] U. S. 2030 U+ROAG, American travel

survey, DB1B, FAA airportdatabase, and BADA

No

0e annual county-to-county personround trips for very light jets,

commercial airline, and automobile atone-year interval from 1995 to 2030

were predicted.

[31] Worldwide (4435 cities) 2042 U+R Forecast based on ADI 2012 No

26 potential markets were identified,regarding large population, limitedurban mobility, economic city, and

high passenger demand.U� urban; R� regional.


compared. No results from the survey were provided. Notethat all travel costs for these four modes in the survey wereassumed as $5. We can observe that findings from statedpreference surveys are rather constrained, and the impli-cations for ODAM operations are limited.

Regarding regional ODAM, Kreimeier et al. [25] per-formed an economical assessment of ODAM for the Germanmarket for the year 2030, with a door-to-door travel speed of80–200 km/h, and a willingness-to-pay value of 0.5–0.8€/km(monetary value in 2015) is derived. Smith et al. [26] ana-lyzed the potential impacts of autonomous, electric on-demand small aircrafts on the national airspace system for3,091 counties in the U. S. based on the projected demand forthe year 2030 by Baik et al. [30]. It was found that a largenumber of on-demand small aircraft trips would competewith automobiles in cost per passenger mile; such number ofoperations would have tremendous negative impacts on thenational airspace system, especially congestion around air-ports. Using a hypothetical network example with five nodesand 1–10 daily trip demands, Mane and Crossley [27]evaluated the impact of accepting different percentages ofpassenger-demanded trips and fleet size on the potentialprofitability of ODAM services. It was shown that, in orderto maximize profit, expected 57.7% of the daily demandshould be served/accepted; and an average acceptance rate ofapproximately 90% might still allow for profitableoperations.

Regarding urban and regional ODAM, Kreimeier et al.[28] estimated the potential market size of thin-haul ODAMservices in Germany, with a comprehensive analysis of theentire German population and their linear spatial distancesto feasible airfields. It was shown that a market share of 19%or 235 million trips are estimated for Germany, assumingpassenger-specific costs of 0.4€/km for ODAM services,0.3€/km for cars, and 0.32€/km for the contemporarycommercial CS-25 aircraft. Balac et al. [29] presented amethodology for demand estimation of personal aerial ve-hicles in urban settings, with the area of 30 km radius aroundthe city center of Zurich, Switzerland, as a case study. 0eminimum number of vehicles needed to serve all trips wasformulated as a minimum cost network flow problem andsolved by a Gurobi Optimizer. It was shown that the vehicleparameters (cruising speed, liftoff time, access time, andprice) have a substantial impact on the ODAM demand andturnover. However, in order to reduce computationalburden, only 10% of the population in the area of 30 kmradius circle around the Zurich city center was sampled. Itwas assumed that vehicles travel in a straight line, withoutconsidering the constraints of infrastructure (landing plat-forms or parking locations); the capacity of each vehicle wasset to be one passenger. Baik et al. [30] presented a modellingframework to predict the annual county-to-county personround trips for the air taxi, commercial airline, and auto-mobile at one-year intervals from 1995 to 2030. Based on atraditional gravity model to estimate ODAM demand andforecasting the air passenger demand between 4,435 set-tlements worldwide for the year 2042, Becker et al. [31]identified 26 interurban potential markets. However, themaximum range of ODAM was limited to 300 km.

In summary, we can observe that most studies focus onthe U. S., while a few focus on Germany and Switzerland, aswell as one for the worldwide market. 0ere is no commonagreement on which market (urban or regional or both) ismore profitable for the ODAM operators. Often, there arefundamentally different parameter settings and differentassumptions on flight missions depending heavily on thecase study under consideration. Significant uncertainties anddiscrepancies in the demand estimation have been identified,and these study results cannot be compared in a consistentway. 0ere is no significant amount of data available fromthe previous studies in order to compare and understand theresults and their implications better.

2.2. ODAMInfrastructure/Port Design and Location Problem.In the last few years, several terminologies for ODAM in-frastructures have been proposed, such as vertiport, skyport,skypark, and airpark. In the following, we do refer to originalterminologies, whenever applicable. Table 2 provides anoverview of the ODAM infrastructure/port design and lo-cation problem, including the region under study, vehicletype, size of the problem, data usage, data availability, and ashort summary. 0e studies are sorted by vehicle types(STOL, VTOL, and eVTOL) and then by region. We discusseach of these studies in detail in the following.

Potential locations of airparks for the vehicle type STOLin two U. S. cities, Miami and South Florida, have beenrecently studied. Based on the FAA Obstacle Data Team’sdatabase, Robinson et al. [32] estimated the geodensity ofairparks suitable for short takeoff and landing ODAM op-erations. Four individual airpark construction types wereanalyzed: vacant land construction, barge construction,additive construction, and pre-existing airport incorpora-tion. Preliminary results for the Miami metropolitan areashow that an average airpark geodensity of 1.66 airparks persquare mile can be achieved with a 300-foot long runway.Somers et al. [33] investigated the impact of the performanceof the short takeoff and landing vehicle (ground roll andclimb/descent gradient) on the airpark geodensity in theSouth Florida region, taking into account the constraints ofobstacles and weather conditions. 0e Light Detection andRanging (LIDAR) data were used to detect the obstaclessurrounding vacant parcels, and weather data (wind speedsand wind directions) were taken from the National Oceanicand Atmospheric Administration. It was shown that, inSouth Florida, doublet runways are needed which make itcomplex to find suitable parcels due to poor weather con-ditions, and the climb/descent gradient has to be greaterthan six degrees.

For the vehicle type VTOL, capacity envelopes of po-tential vertiports in the U. S. have been analyzed as well.With New York City, U. S., as the case study, Rajendran andZack [20] proposed an iterative constrained clustering al-gorithm with the multimodal transportation warm-starttechnique to identify potential locations for ODAM. 0enumber of locations varies from 10 to 85. Certain locationsincluding large facilities and smaller stops for New York Cityare suggested. Surprisingly, they found that the percentage of


time savings and willingness to fly did not impact the lo-cation decision significantly, whereas the on-road travellimit (the maximum time for passengers spent on the road)does. Vascik and Hansman [34] used an integer program-ming approach to derive vertiport capacity envelopes andassess the sensitivity of ODAM vehicle throughput to ver-tiport topology variations and operational parameters. Fourclasses of topologies were included: linear topology, satellitetopology, pier topology, and remote apron topology. 0isapproach was applied to 156 different vertiports and 146different operational parameter settings. Results show thatthe ratio of gates to touchdown and liftoff pads at a vertiportis a key design factor. Moreover, the construction of vehiclestaging stands is recommended. Finally, vertiports withmultiple touchdown and liftoff pads which allow fully

independent operations or simultaneous arrivals and de-partures would be beneficial as well. Based on Seoul Met-ropolitan Areas as the case study, Lim and Hwang [35]selected the number of vertiports for three major routesbased on the K-means clustering algorithm over the com-muting data for the cases of 10–36 vertiports. Note that nooptimization strategy was used in the selection process of thevertiport location.

For the vehicle type eVTOL, the determination of op-timal locations of vertiports in a given region can bemodeledas a variant of standard hub location problems (HLPs): howto collect, transfer, and distribute travel demands betweenorigin nodes and destination nodes in a network, see recentreviews on network design in [39–42]. Rath and Chow [36]formulated the air taxi skyport location problem as an

Table 2: An overview of the ODAM port design and location problem.

Ref. Region Vehicletypes Problem size Data typeDataavail. Short summary

[32] Miami, U. S. STOL 15,666 potentialpark locationsFAA Obstacle DataTeam’s database No

0e Miami metropolitan area showed that anaverage airpark geodensity of 1.66 airparks persquare mile can be achieved with a 300-foot

long runway.

[33] South Florida,U. S. STOLUp to 10,310

parcels

U. S. Light Detectionand Ranging data and

NOAAPartially

Doublet runways are needed to find suitableparcels in South Florida due to poor weatherconditions; and the climb/descent gradient has

to be greater than six degrees.

[20] New York, U. S. VTOL 10–85 sites Taxi records No

Certain locations including large facilities andsmaller stops for New York City are suggested.0e percentage of time savings and willingnessto fly did not impact the location decisionsignificantly, while the on-road travel limit

does.

[34] U. S. VTOL 156 vertiports Simulation data No

0e ratio of gates to touchdown and liftoffpads at a vertiport is a key design factor;vehicle staging stands are recommended;

vertiports with multiple touchdown and liftoffpads which allow fully independent operationsor simultaneous arrivals and departures would

be beneficial as well.

[35] Seoul, SouthKorea VTOL10–36

vertiports, 3routes

Commuting data inSeoul, Incheon, and

GyunggiYes

0e number of vertiports was based on the K-means clustering algorithm, based on the

commuting data for three major routes in thecase study of Seoul.

[36] New York, U. S. eVTOL 10 skyportsTaxi and Limousine

Commission records forJan. 2018

Yes

When the number of skyports p> 6, there isnot much variation in the incoming demand;an even higher value (p> 9) is required toachieve at least 10% market penetration.

[37] San Francisco,U. S. eVTOL 1–8 vertiportsCensus tract data andpopulation and income

dataNo

0e objective function is to maximize thepackage demand served, with limits on thenumber of vertiports. Increasing the time

threshold for the allowable driving time fromthe vertiport to the customer would result in awider geographical distribution of vertiports.

[38]San Francisco

and Los Angeles,U. S.

eVTOL 10, 20, and 40vertiportsCensus data (LODES

and ACS) No

0e objective function is to maximize thecumulative time saved in commuting

throughout the network. Most appealing tripsare relatively short-range trips in San

Francisco; the strongest growth is in thecentral area for Los Angeles.

STOL� short takeoff and landing, VTOL� vertical takeoff and landing, and eVOTL� electric vertical takeoff and landing.


uncapacitated single allocation p-hub median problem (thenumber of hubs is p). 0e objective function is to minimizethe travel time for each OD pair in the skyport network, withthe consideration of air taxi transfer time and congestionfactor on the ground. Gurobi was used to solve this problemwith up to 10 skyports for travel demands among 144 taxizones and 3 airports. Experimental results for New York Cityshow that when the number of skyports p> 6, there is notmuch variation in the incoming demand; an even highervalue (p> 9) is required to achieve at least 10% marketpenetration. With San Francisco Bay area as the case study,German et al. [37] examined the use of eVTOL vehicles forsmall package delivery. 0e optimal location of the cargovertiports was formulated as an optimization problem:maximizing the package demand served, with limits on thenumber of vertiports. However, only 1–8 vertiports weresolved for this optimization. Results show that increasing thetime threshold for the allowable driving time from thevertiport to the customer would result in a wider geo-graphical distribution of vertiports. 0e vertiport locationproblem was further extended by Daskilewicz et al. [38],taking into account household population density as well aswork locations. With San Francisco and Los Angeles as twocase studies, the new objective function is to maximize thecumulative time saved in the network, with three differentnetwork sizes: 10, 20, and 40 vertiports. It was found thatmost appealing trips are relatively short-range trips in SanFrancisco; the strongest growth in eVTOL use is in thecentral area for Los Angeles. Moreover, Wawrzyniak et al.[43] analyzed the ODAM network based on complex net-work theory using real-world cargo aircraft movement datawithin the U. S. Central Command area of responsibility inthe year 2006.

In summary, similar with the case of demand estimation,most previous studies cover the U. S., except one study forSeoul, South Korea [35]. 0erefore, the universal applica-bility of these results to other regions in the world is verylimited since different metropolitan areas have distinctsocio-economic-demographic characteristics. 0e vehicletype covers STOL, VTOL, and eVTOL, among which themajority of studies in the literature focuses on VTOL. 0eproblem size for the ODAM infrastructure design and lo-cation problem is rather limited in the literature because ofthe computational complexity.0ere are some studies whichcompute optimal solutions for very small size of theproblem, for instance, less than 10 skyports/vertiports[36, 37], while a few studies consider more than 10,000potential locations [32, 33] using heuristics and rules ofthumb to obtain the solutions. Tradeoffs between the level ofaccuracy and the computational complexity need to befurther explored. Moreover, there is no well-defined refer-ence benchmark dataset as well as solution techniques forthe ODAM infrastructure location problem; therefore, it isdifficult to reproduce the results and perform comparativeanalysis and further extensions.

2.3. ODAM Operational Planning Problem. 0is sectionsummarizes the literature on operational planning ofODAM. Table 3 provides an overview of the ODAM

operational planning problem, including the region understudy, vehicle type, size of the problem, mathematical for-mulation, solution technique, and a short summary. 0estudies are sorted by vehicle types (general vehicles, small jet,very light jet, and eVTOL) and then by region.

Traditionally, there are three horizons in the airlineplanning process for scheduled air transportation: long-termstrategic planning, midterm tactical planning, and opera-tional planning [54]. Strategic planning involves customerdemand forecasting, market segmentation, and brandingand marketing strategies; tactical planning addresses flightscheduling (which determines the optimal daily schedule offlight operations) and fleet assignment (where an aircrafttype is assigned to each flight in the schedule); operationalplanning deals with aircraft scheduling (which decides asequence of flight legs to an individual aircraft, and it is oftendecomposed into aircraft routing and tail assignment) andcrew scheduling (the pairing of crew resources according tothe flight schedules). 0e traditional aircraft schedulingproblem has always been the main focus of the airline in-dustry since fuel consumption, crew salaries, and aircraft-related cost typically represent the largest expense for anairline. For large commercial airlines, this problem is usuallycombined with flight scheduling [55, 56], fleet assignment[57], and crew scheduling [58]. A wide range of research hasbeen proposed to tackle this problem such as the classicflight string model [59].

0e business model of ODAM shares large similaritieswith existing on-demand aviation services, especially frac-tional management companies. Among on-demand aviationservices, there are several types of programs: fractionownership, timeshare, and joint ownership [60]. Whilefraction ownership allows different stakeholders to useaircraft resources for a fraction of time at different levels,much research has been done for the scheduling of thefractional operation business model [61–64]. See [65] for anexcellent review on the vehicle routing problem with pickupand delivery, with four areas of applications, the dial-a-rideproblem, the urban courier service problem, the dial-a-flightproblem, and the emergency vehicle dispatch problem, fo-cusing on the problem formulation and solution algorithms.

Few studies investigate the operation planning problemfor general vehicles. Van der Zwan et al. [44] adopted the setpartition model [61] and provided a detailed description ofthe aircraft routing problem for per-aircraft air taxi operator.Contrary to the column generation solution methodologyapplied in [61], they used a k-shortest path algorithm togenerate feasible routing pools for the set partition. With anumber of 19–25 air taxis for 225 airports over the 72-hourtime horizon, CPLEXwas applied to solve the set-partitioningmodel. However, only virtual test environment and virtual airtaxi operator data were used due to the lack of real air taxioperational data. Kohlman and Patterson [45] presented anurban air mobility network model for rapid simulation,consisting of four parts: network definition and missionmodel, vertiport model, vehicle model, and demand model.0e concurrent on-demand vehicle design and vehicle lo-cation problem has been investigated as well. In order to solveconcurrent aircraft design and operation planning problems,


Table 3: An overview of the ODAM operational planning problem.

Ref. Region Vehicletype Problem sizeMathematicalformulation Solution technique Short summary

[44] Fly Aeolus,Belgium General

225 airports, 19–25vehicles, 7–15 flightdaily requests, and 15closest trips for a route

Set-partitioningmodel

Exact solution withCPLEX

0e developed aircraft routingsystem consisted of model creator,CPLEX, and the solution module.0e number of closest trips that areappended to a trip to a route is 3–15,and this parameter has a large effect

on the feasible routes and thecomputational time.

[45] Hypothetical General N/A N/A N/A

An urban air mobility networkmodel for rapid simulation was

presented, consisting of four parts:network definition and missionmodel, vertiport model, vehiclemodel, and demand model.

[46] Hypothetical General 10 airports Integer program Exact solution withCPLEX

Concurrent aircraft design andoperation planning problems werecombined together; its integerprogramming model entailedrevenue-generating trips, non-revenue-generating trips, and

charter flights.

[47] FlySmart,Norway Small jet12 airports, 10 vehicles,and 200 daily bookings

Dial-a-flightproblem model

Heuristics(insertion + local

search)

0e required air taxi fleet to handleapproximately 100 bookingrequests is close to 15, andconfirming the pickup times

immediately to the customers canprovide as good solutions as

postponing the confirmation thenight before.

[48] Etirc Aviation,EuropeVery

light jet10 airports, 38 routes,and 270 weekly flights

Dial-a-rideproblem model

Dynamicprogramming

An average of 5.5% cost advantageis achieved on the cost of emptylegs; the addition of a two-hour timewindow in operations increases thecustomer acceptance rate with

2–3%.

[49] NorthCarolina, U. S.Very

light jet

5 airports, 10 routes,and 50–250 weeklypassengers per route

Discrete-eventmodel and flow

modelGradient algorithm

An application of these two modelson air taxi pricing showed that theflow model could be used for a

pricing application.

[50] U. S. Verylight jet12–75 vehicles (test

scenario) N/A N/A

A fast-time simulation tool wasdeveloped to understand thecomplex dynamics of air taxi

networks.

[51] Hypothetical eVTOL 3 vertiports, 20 vehicles,and 6 routes

Mixed-integerquadraticprogram

Gurobi solver (planto use)

0ree variants of basic formulationmodels for different types of

ODAM services were presented.However, these models were not

being implemented, and nonumerical experiment results were

provided.

[52] Hypothetical eVTOL N/A N/A N/A

An open-source multiagenttransportation simulationframework was developed,

considering the effects of variationsin vehicles as well as infrastructure

locations.


Mane and Crossley [46] combined aircraft allocation andaircraft design together; its integer programming modelentailed revenue-generating trips, non-revenue-generatingtrips, and charter flights. Due to the model complexity, aninstance of 10 city nodes was tested through direct CPLEXimplementation.

For the small jet vehicle type, Fagerholt et al. [47]presented a simulation methodology to support strategicdecisions for an air taxi company in Norway (FlySmart),including fleet size, price differentiation strategies, andbooking confirmation policies. A heuristic with insertionand local search was used to optimally distribute waitingtime along a single aircraft schedule. A fleet of ten aircraftswith approximately 200 daily bookings and 1–4 passengersfor each booking were used as case studies.

For the very light jet vehicle type, de Jong [48] modeledthe air taxi dispatch problem (which is defined as a time-scheduled trip being part of a single aircraft route) as a dial-a-ride problem based on a case study of ten airports, 38routes, and 270 flight requests per week, representing the airtaxi network of Etirc Aviation (a company which operatespersonalized business flights in Europe and Russia). Oneselected case study showed that an average of 5.5% costadvantage is achieved on the cost of empty legs, and theaddition of a two-hour time window in operations increasesthe customer acceptance rate by 2–3%. Note that thestandard dial-a-ride problem assumes a homogeneous fleetwith a fixed size. A single-provider air taxi service model wasused to characterize the week-to-week flow of passengersand the aircraft [49] for the case of 5 airports, 10 routes, and50–250 weekly passengers per route [66]. Results show thatthe flow model could be used for a pricing application.Bonnefoy [50] built a fast-time simulation tool to under-stand the complex dynamics of air taxi networks, includingdemand modelling with the gravity model, trip generationbased on Monte Carlo simulation, aircraft routing, and pilotassignment, as well as unscheduled maintenance events.

For the eVTOL vehicle type, based on the simulatedODAM market demand, Shihab et al. [51] formulated threebasic models of scheduling types, fleet dispatching, and fleetplanning, with the goal to maximize profit. 0ree types ofservices were compared as well: on-demand service,scheduled service, and amix of both services. However, thesemodels were not implemented, and no numerical experi-ment results were provided. Rothfeld et al. [52] presented anopen-source multiagent transportation simulation frame-work, MATSim, with the extension of considering the effectsof variations in personal aerial vehicles as well as dedicated

infrastructure locations. Kleinbekman et al. [53] proposed asequencing and scheduling algorithm for ODAM arrivals,and experiment results showed that their algorithm can besolved within 79 seconds for up to 40 incoming vehicles. Anovel concept of airborne trajectory management for highODAM traffic densities in a congested urban area has beenproposed as well [67].

In summary, we can observe that there are a few studiesconducted from the air taxi operator’s point of view, in-cluding a Belgian air taxi company Fly Aeolus [44], aNorwegian air taxi company FlySmart [47], and the Lux-embourg-based air taxi company Etirc Aviation [48], whilemost studies used hypothetical network operators. Very lightjet and electronic vertical takeoff and landing (eVTOL)vehicles are the chosen types. 0e size of the operationalplanning problem is very limited, including the number ofvehicles, the number of airports (3 vertiports [51] and 5airports [49]), and the number of served routes (10 routes[49]). 0e scale of the problem in existing research is farfrom the envisioned operational scale of ODAM, with thepredicted trip demand in a magnitude of million [26, 28].Furthermore, different mathematical models have beenformulated with distinctive solution techniques; for verysmall-scale problems, the exact solution with CPLEX wasoften used [44, 46, 53], while heuristics are frequently used tosolve medium-size problems [47]. Similar with the case ofthe ODAM infrastructure/port design and location problem,there are no well-defined reference benchmark datasets aswell as solution techniques for the ODAM operationalplanning problem, and thus, it is extremely difficult to re-produce and understand the results from existing work.

2.4. ODAM Operational Constraints’ Identification.Although there is a large body of research focusing on theassessment of the technical feasibility of ODAM, operationalconstraints during the implementation process of theODAM network at large scale are the bottleneck forachieving market success. Table 4 provides an overview ofidentified operational constraints for ODAM, including theregion under study, vehicle type, fleet size of ODAM ve-hicles, operational scale (urban vs. regional), data usage andtheir availability, and a short summary. 0e studies aresorted by vehicle types (VTOL, personal air vehicle, andgeneral vehicles) and then by region.

For the VTOL vehicle type, Rothfeld et al. [68] analyzedthe operation performance of a potential ODAM imple-mentation with the artificial Sioux Falls Network using a

Table 3: Continued.

Ref. Region Vehicletype Problem sizeMathematicalformulation Solution technique Short summary

[53] Hypothetical eVTOL 10–40 vehicles Mixed-integerlinear programExact solution with

CPLEX

A sequencing and schedulingalgorithm for ODAM arrivals wasproposed, and it can be solvedwithin 79 seconds for up to 40

incoming vehicles.eVOTL� electric vertical takeoff and landing.


multiagent simulation tool, MATSim [52]. Varying pa-rameters have been tested: vehicle cruise speed (50–450 km/h), vehicle takeoff and landing speed (5–20m/s), ground-based process time (0.5–20min), passenger capacity (1–12),fleet size (50–300), and the number of vertiports (4–10).Simulation results show that the adoption of ODAM isstrongly influenced by the potential travel time reduction.However, only homogeneous vehicles are simulated, and theprice of using ODAM is assumed to be three times the carprice.

For personal air vehicles, Liu et al. [69] provided anoverview of research activities with the focus on the U. S. andEurope. It was found that, despite dramatic technologyinnovation, several challenges still remain in the ultimateapplication of personal air vehicles, especially safety, in-frastructure availability, and public acceptance.

Most studies on the identification of the ODAM oper-ational constraints are for general vehicles. With the LosAngeles Basin as the case study, Vascik and Hansman [70]identified several critical operational constraints for ODAMservices: noise and public acceptance, accessibility of takeoffand landing sites, interaction with air traffic control, ground

infrastructure, and flight density. 0ey also discussed po-tential mitigation strategies against the first three key op-erational constraints. 0e authors further investigated threemajor cities in the U. S. (Los Angeles, Boston, and Dallas)[71]. With 32 reference missions within these three cities, itwas found that the three most stringent constraints are noiseacceptance, availability of takeoff and landing sites, andscalability of air traffic control.

0e routes and procedures for urban air mobility basedon the existing helicopter routes have been investigated aswell. With Dallas-Fort Worth area as the case study, Vermaet al. [72] investigated routes and procedures for urban airmobility based on the existing helicopter routes along withdifferent communication procedures. 0ree different levelsof traffic were evaluated: low (115 flights), medium (167flights), and high (225 flights). Field test data showed thathelicopter routes and communication procedures on thecurrent day can support near-term urban air mobility op-erations, but may not be scalable due to extra workload forair traffic controllers.

Moreover, the integration of the urban air mobility withthe city transportation has been studied recently.

Table 4: An overview of the ODAM operational constraints’ identification.

Ref. Region Vehicletype Fleet size Scale Data typeDataavail. Short summary

[68] Artificial Sioux FallsNetwork VTOL 50–300 U Simulation data No0e adoption of ODAM is strongly influenced

by the potential travel time reduction.

[69] U. S. and Europe Personal airvehicle N/AU

and R N/A N/A

It was found that, despite the dramatictechnology innovation, several challenges stillremain in the ultimate application of personalair vehicles, especially safety, infrastructure

availability, and public acceptance.

[70] Los Angeles General N/A U FAA 5010database, etc. No

Identified several critical operationalconstraints for ODAM services: noise and

public acceptance, accessibility of takeoff andlanding areas, interaction with air trafficcontrol, ground infrastructure, and flight

density; potential mitigation strategies againstthe first three key operational constraints were

discussed.

[71]3 cities in the U. S.

(Los Angeles,Boston, and Dallas)

General N/A UCensus data,

consumer wealthdata, etc.

No

With 32 reference missions within these threecities, it was found that the three most

stringent constraints are noise acceptance,availability of takeoff and landing areas, and

scalability of air traffic control.

[72] Dallas-Fort Wortharea General115–225flights U

Field experiment/test data No

It was found that helicopter routes andcommunication procedures in the current daycan support near-term urban air mobility

operations, but may not be scalable.

[73] General General N/A U N/A N/A

Several relevant factors regarding theintegration of ODAM to the urban city

transportation environment were discussed,such as ownership structure, adaptability of

schedules, and demand drivers.

[74] Munich, Germany General N/A U N/A N/A

Preliminary results for the case study ofMunich show that rents in all city centers

increase, while the household’s average utilitydecreases.

U� urban; R� regional.


Straubinger and Raoul [73] discussed several relevant factorsregarding the integration of urban air mobility to the citytransportation environment, such as ownership structure(private vs. rental), adaptability of schedules (scheduled vs.on-demand), and demand drivers (job vs. housing). Fur-thermore, the authors developed an urban spatial com-putable general equilibrium model to assess the welfarechanges for the urban air mobility integration [74]. Pre-liminary results for the case study of Munich show that rentsin all city centers increase, while the household’s averageutility decreases.

In summary, there seems to be a consensus that thepublic acceptance, infrastructure availability, and scalabilityof ODAM operations are critical operational constraints.However, different problem definitions have been consid-ered in the literature, for instance, personal air vehiclesindicating the ownership of the vehicle itself [69] and urbanair mobility being part of the city transportation system[73, 74], inducing different operational concepts and busi-ness models. Furthermore, the projects in the field of ODAMresearch have not been commonly planned or coordinated,which makes the accessibility to the in-depth findings andanalysis results difficult.

2.5. Competitiveness of ODAM with Existing TransportationModes. As a new emerging transportation mode, ODAMhas the potential to relieve the grid-locked ground trans-portation in megacities by leveraging the third spatial di-mension. Table 5 provides an overview of thecompetitiveness of ODAM compared with existing trans-portation modes, including the region under study, oper-ational scale (urban vs. regional), data usage and theiravailability, competition modes, evaluation criteria, and ashort summary. 0e studies are sorted by scale (urban andurban + regional) and then by region.

Most studies focus on urban ODAM; stated preferencesurveys were often used to identify the mode choice behavioramong different transportation modes. Fu et al. [18] con-ducted a stated preference survey with 248 respondents inMunich, Germany, regarding the mode choice behavioramong four transportation alternatives: public trans-portation, private car, autonomous taxi, and autonomous airtaxi. Five different criteria were applied: total travel time,total travel cost, safety, inconvenience, and multitaskingpossibility. Results show that the first three are the mostcritical determinants for the adoption of autonomoustransportation modes. Moreover, higher value of time andhigher income also favor the use of urban air mobility.Moreover, there are some studies on the U. S. regions.Binder et al. [22] presented details of a survey for estimatingcommuters’ willingness to pay for eVTOL in 5 cities,Atlanta, Boston, Dallas, San Francisco, and Los Angeles inthe U. S., collecting responses from 2,500 high-incomeworkers whose average one-way commuting time is 45minutes or more by the individual. Four transportationmodes are compared, transit, traditional personal car, self-driving car, and piloted air taxi, regarding multiple criteria:travel time, travel cost, access/egress/waiting times, transfer,

and ride guarantee availability of different modes. Fur-thermore, Garrow et al. [24] conducted a second surveyamong 1,405 respondents in the same five cities in the U. S.as in [22], and the survey settings with approximately 100questions are very similar to [22], so as the same fortransportation mode alternatives and evaluation criteria.However, no results from both surveys mentioned abovewere provided.

Travel time, specifically door-to-door travel time, is themost critical factor for the competitiveness of differenttransportation modes. 0ere are two studies regarding thecompetitiveness of ODAM: one for the U. S. and the otherone for Europe. Wei et al. [75] proposed a methodology toanalyze the door-to-door commuting time of short takeoffand landing vehicles, compared with personal cars. A total of4,529 different commutes between 1,035 unique censustracts in South Florida were used as the case study. Asensitivity analysis of the door-to-door travel time againstthe vehicles’ takeoff distance, cruise speed, ground handlingtime, climb rate, and cruise altitude was performed. It wasfound that takeoff distance and cruise speed are the twomostimportant parameters for the benefits of ODAM. Reducingthe takeoff distance is important in order to be able to find asufficient number of vertiport locations. For South Florida,ODAM operations should target for ground commutesmore than 45min, and the vehicles should be able to take offfrom runways of 500 ft, with cruising speeds over 160 kts.Note that commutes shorter than 30min are filtered outbecause of lack of incentives to switch to an ODAM service.

For the scale of urban and regional ODAM, Sun et al.[76] designed and implemented a door-to-door travel timeestimation framework to analyze the potential competi-tiveness of ODAM when competing with existing trans-portationmodes (car, railway, and aircraft), and 29 countriesin Europe were used as case studies. Results show that whilethe major competitor for ODAM in Europe is railway, thecompetitiveness of air taxi service, however, largely dependson the region. 0ese 29 European countries have a largepotential to benefit from air taxi service up to 450 km be-cause the periphery of Europe has the largest potential forODAM services.

Although travel time is appealing for the adoption ofODAM, travel cost is also one decisive factor for thecompetitiveness of ODAM. Uber estimated that the price forpersonal air vehicles is at the level of Uber Black or opti-mistically as low as its cheapest service UberX price [77].Regarding the potential market size of thin-haul ODAMservices in Germany, Kreimeier et al. [28] estimated amarketshare of 19% or 235 million ODAM trips, with the as-sumption that passenger-specific costs will be 0.4€/km forODAM services, 0.3€/km for cars, and 0.32€/km for con-temporary commercial CS-25 aircrafts. With NorthernCalifornia and Washington-Baltimore region, U. S., as twocase studies, Syed et al. [21] estimated that the ODAM costshave to be kept between $1 per pax mile and $1.25 per paxmile to achieve a potential market share of 0.5–4%, where a4% market share represents 320,000 trips per day. ODAM isused for different range classes, and this alone leads toseemingly inconsistent results (e.g., urban air mobility $2 per


pax km and regional air mobility less than $0.65 per pax km).See [78] for vehicle configurations up to 100 km, as well asthe Silent Air Taxi and Lilium for ranges beyond 100 km.

In summary, it is instrumental to compare the com-petitiveness with existing transportation modes upon theintroduction of the ODAM in order to predict the mode

choice behavior of travelers more accurately and align themarket position for the ODAM service more realistically inthe future. From the existing literature, we can observe thatin comparison with traditional transportation modes, onemajor advantage of ODAM is its travel time, both for thecase of the U. S. and Europe. However, travel costs estimated

Table 5: An overview of the competitiveness of ODAM with existing transportation modes.

Ref. Region Scale Data type Dataavail. Competitive modes Evaluation criteria Short summary

[18] Munich,Germany UStated preference

survey No

Public transport,private car,

autonomous taxi,and autonomous

air taxi

Total travel time, totaltravel cost,

inconvenience, safety,and multitasking

possibility

Travel time, travel cost, andsafety are the most critical

determinants for theadoption of autonomoustransportation modes.

Moreover, higher value oftime and higher incomealso favor the use of urban

air mobility.

[22]

U. S. (2500workers in 5cities: Atlanta,Boston, Dallas,San Francisco,

and Los Angeles)

U Stated preferencesurvey NoTransit, traditionalpersonal car, andride-share car

Travel time, travel cost,access/egress/waiting

times, transfer, and rideguarantee availability of

different modes

No results from the surveywere provided.

[24]

U. S. (1405workers in 5cities: Atlanta,Boston, Dallas,San Francisco,

and Los Angeles)

U Stated preferencesurvey NoTransit, traditionalpersonal car, andself-driving car

Travel time, travel cost,access/egress/waiting

times, transfer, and rideguarantee availability of

different modes

No results from the surveywere provided.

[75] South Florida,U. S. U

Censustransportation

planning products of2000 (survey data)

No Personal car Door-to-door traveltime

Takeoff distance and cruisespeed are the two mostimportant parameters forthe benefits of ODAM. Forthe South Florida area,

ODAM operations shouldtarget for ground

commutes more than45min, and the vehiclesshould be able to takeofffrom runways of 500 ft,with cruising speeds over

160 kts.

[76] 29 countries inEurope U+R

OpenStreetMap,openAIP, and

gridded populationof the world

YesConventional

aircraft, rail, andpersonal car

Door-to-door traveltime

0e major competitor forODAM in Europe israilway, and the

competitiveness of air taxiservice largely depends onthe region. Furthermore,

these 29 Europeancountries have a large

potential to benefit fromair taxi service up to

450 km, and the peripheryof Europe has the largestpotential for ODAM

services.U� urban; R� regional.


for ODAM with different range classes (urban vs. regional)seem completely inconsistent; also, market shares werederived incomparably. 0ese inconsistencies make it ex-tremely difficult to judge the profitability of ODAM oper-ations. 0e unavailability of the large amount of data makesthe comparative analysis and assessment even more difficultand less reliable.

2.6. Discussion. We have reviewed recent literature on op-erational aspects of ODAM, covering five major categories:demand estimation methodology, infrastructure/port de-sign/location problem, operational planning problem, op-erational constraints’ identification, and competitivenesswith other transportation modes. Research on the opera-tional aspects of ODAM is widely dispersed and published indifferent scientific venues, using different nomenclature andinvestigating highly similar, yet distinct research problems.0e major findings of our survey are summarized in thefollowing.

2.6.1. Lack of Formal Problem Definition. While small air-crafts have been used and operated privately for businesspurposes (general aviation) for decades, the idea of a broadand ubiquitous air taxi network as a mobility service hasbeen promoted, especially since the emergence of new de-sign concepts and propulsion technologies. With this hype,many recent research projects aim at analyzing and opti-mizing a part of air taxi flight operations, however, withoutdefining a fully established and standardized nomenclaturefor research projects first. For instance, some studies con-sider the design and planning of ODAM flight operationsfrom the perspective of typical routing and planningproblems and therefore adopt the corresponding vocabularyfrom standard planning problems of ground traffic (e.g., see[20]). In other studies, on the contrary, assumptions arebased on already known general aviation procedures, whichis why terms and definitions from this area are used (e.g., see[44]). Furthermore, the aircrafts used for such flight servicesare sometimes referred to as “air taxis” (e.g., see [27]), butelsewhere also as “personal air vehicles” (e.g., see [5]),suggesting ownership by the passenger himself, and thus acompletely different operational concept. 0is lack ofestablished phraseology, the misuse of language, and, aboveall, the lack of a real, uniform problem definition often leadto confusion and make it considerably more difficult tounderstand the current state of air taxi research, let alone torecognize similarities and differences.

2.6.2. Coverage and Inconsistency. Analogous to the non-uniform formulation of the actual research problem, ODAMliterature makes different assumptions regarding flightmissions, especially depending on the case study (i.e., initialconditions of the area of investigation) and individualproject focus. An important distinction is, for example, theintended use and the associated range capacity of the ODAMaircraft. 0is fundamental difference in initial mission as-sumptions alone seems to lead to inconsistent study results

in terms of operating costs per passenger kilometer, a keymeasure for comparing study results. In studies with urbanapplications (“urban air mobility”), flight costs of approx.2$/pax/km were calculated; for regional air mobility with aflight distance of up to 500 km, however, less than 0.65$/Pax/km were calculated (see [28]). In addition, despite manyserious efforts of recent ODAM studies to estimate accuratedemand in several regions of the world, we have identified acollection of significant uncertainties and discrepanciesbetween the results of these studies. For example, Kreimeieret al. [28] estimated a market share of 19% or 235 milliontrips per year for the potential market size of ODAM thin-haul services in Germany, assuming passenger-specific costsof 0.4€/km for ODAM services, 0.3€/km for cars, and 0.32€/km for modern commercial CS-25 aircrafts. In contrast,Decker et al. [19] used a substitution rate of 10% of car trafficby ODAM for a typical, average European city instead,assuming that daily car traffic is around 300,000 trips. Syedet al. [21] estimated that ODAM costs have to be kept be-tween $1 per pax mile and $1.25 per pax mile to reach apotential market share of 0.5–4%, with a market share of 4%representing 320,000 trips per day for Northern Californiaand the Washington-Baltimore region in the U. S. as casestudies. As particular projects aim at completely differentstudy areas with heterogeneous characteristics, it is im-portant for the comparison of the study results to have a listof initial study parameters. As mentioned before, individualassumptions of the passenger and range capacity of the studyvehicles result in different substitute transport capacities andthus diverging information regarding a possible transportdemand. In addition to technical parameters, standardizedboundary conditions must also take into account opera-tional factors such as mission time, mission start and endlocation, possible connections, pricing systems, and acces-sibility when setting the mission studies.

2.6.3. Applicability of Research Results. As mentioned be-fore, many research projects base their studies on a specificstudy area (urban or regional areas), adopt its specificcharacteristics as input values for parameter studies, andalign the ODAM network accordingly. However, mostprevious ODAM research projects focused on large U. S.metropolitan areas (e.g., Los Angeles as in [22] or [70] or SanFrancisco as in [37]) so that the universal applicability ofthese results is very limited. Different metropolitan areas allover the world with their demographic, geographical, ortopographical characteristics, different demand patterns forair taxi services, and air traffic management systems facedifferent operational challenges. A systematic, comparativeanalysis of hypothetical ODAM operations in representativeregions of the world, which highlights essential differences,is therefore lacking.

2.6.4. Optimality vs. Scalability. Analogous to conventionalproblems with the design and optimization of networks,ODAM optimization is inherently difficult. With increasingscenario size and additional parameters as the input, thecomputational complexity and thus the computation time


also increase significantly so that large computer clusters arerequired. 0is often results in a compromise between thelevel of detail/accuracy and handling/calculation time.Hence, previous studies can generally be assigned to twogroups:

(a) Studies that calculate optimal solutions/assignmentsfor very small study areas with a small fleet (3 routes[35]) and only a few requirements, for instance, lessthan 10 vertiports/skyports/airports [36, 37, 46, 51].0e level of detail and significance is to be regardedas low, but the calculation time is comparativelyshort.

(b) Studies that consider many heuristics and carry outrules of thumb for large input problems, for example,more than 10,000 potential locations [32, 33]. 0isincreases the significance; thus, the study results canbe classified as comparatively reliable. Handling, onthe contrary, is very demanding and requires suffi-cient computational power and time.

A particular problem here is the fact that there is almostno favorable, useful transition or overlap between the twotypes of studies as they are often carried out and published atdifferent institutions with different intentions, e.g., in op-erational research or air traffic management, consideringdifferent publication criteria.

3. Future Research Directions

As our literature review shows, there are enormous effortsmade to understand the potential of ODAM and to use it torevolutionize the way we think about transportation. Nev-ertheless, based on the above results, we see a whole set of linesfor future research on the operational aspects of ODAM.0ese proposals for research are discussed in the following.

3.1. Formulation of a Common Lexicon and Adequate ODAMResearch Problems. Due to the comparably young field ofresearch, ODAM research problems are loosely defined,especially in comparison to well-established counterparts inground traffic and commercial (scheduled) air trans-portation. 0is makes it rather challenging to qualitativelycapture and evaluate the contributions of specific studies,let alone compare studies and their results on a larger scale.0erefore, we believe that future ODAM projects will benefitsignificantly from a number of specially aligned ODAMresearch problems. Established studies can inspire some ofthese problems, but the focus must be on key differences andpeculiarities. In light of the studies reviewed as part of thissurvey, we propose to use the basic terminology ODAM taxi(covering air taxi, personal air vehicle, etc.) and ODAM port(covering vertiport, skyport, and airfield). Moreover, wepropose the following ODAM operational research prob-lems; while these problems interact with each other, they arelisted in a chronological order.

3.1.1. ODAM Demand Estimation. Estimate the demandpotential for ODAM operations in a given region. Tradi-tional models, such as gravity or radiation model, are good

for estimating the large demands. Accordingly, out of thebox, these models have limited applicability since ODAMoperations cannot be assumed to fully capture such esti-mates. ODAM, by design, is targeting special trips, oftenwith lower demand (premium passengers, urgent trans-portation requests, or those without any reasonable trans-portation alternatives). Accordingly, there is a need fordemand estimation including (a) uncertainty, (b) presenceof existing transportation infrastructure, and (c) temporalinformation (time of day).

3.1.2. ODAM Port Location Problem. Optimize the locationsof ODAM ports in a given region. 0is problem has to relyon accurate estimations from the previous stage (ODAMdemand estimation). 0e optimization function for theODAM port location problem cannot simply be trans-portation time or cost. Instead, this is naturally multi-objective: it needs to further include ground access time forpassengers, environmental costs, and other societal costs.

3.1.3. ODAM Scheduling. Design schedules for ODAMtaxis. 0ree types of problems are conceivable: first, a simplemodel where all ODAM taxis are grounded until there is atransportation request. In this case, there is no special needfor scheduling between ODAM ports, second, a modelwhere all ODAM taxis are airborne in anticipation of triprequests (as induced by ODAM demand estimation) inorder to reduce air access time and maximize vehicle in-usetime, and third, a model combining both previous modelsinto a general one.

3.1.4. ODAM Dispatching. Assign actual trips to ODAMtaxis. 0is case is specifically interesting for ODAMscheduling operations, where some ODAM taxis aregrounded and others are in flight. Choosing an appropriateODAM vehicle (or, in an extreme case, rejecting thetransportation request) involves decisions regarding ODAMtaxi capabilities, estimated profit, passenger perception, andmore.

3.1.5. ODAM Routing. Assign actual routes to ODAM taxis.0is includes trajectory planning and collision avoidance, allunder safety considerations and within operational limits.0is stage has potential interactions with ODAM dis-patching, in case of in-flight, on-demand requests, whichlead to changes in planned trajectories.

Future research studies should carefully distinguishODAM operations from traditional airline scheduling. Forexample, in scheduled air transportation, flight rotationsconsist mainly of revenue-generating trips between airports,whereas commercial airlines often analyze aircraft sched-uling and other related optimization issues one or moremonths in advance. 0e ODAM service features door-to-door travel patterns, resulting in much more dynamic andcomplex operations. In addition, nonrevenue trips due torepositioning flights must also be taken into account inplanning to meet dynamic demand, an event that is more


likely to be avoided in conventional scheduled air transport.Since the demand mechanisms for ODAM services are oftenshort term, less predictable, and less reproducible, theexisting conventional air transportation planning process forODAM services must be adapted [44], taking into accountdifferent variants of delay [79] and disruptions [80].Moreover, most studies in the literature so far focus onODAM taxis with space for only one passenger. 0e first realbusiness model directly linked to ODAM operation istherefore likely to be a single-seater model, covering thetransport needs of a single person. As a direct extension, avariant is conceivable in which more than one passengershare an ODAM taxi for a flight on the same route and thushead for the same departure and destination airfield/station.0e problem with such operations, however, is that there is acompromise between the passenger pickup and the demandsituation (on-demandness): in order to maximize the loadfactor of an ODAM taxi, the operator is tempted to wait forfurther requests, while additional waiting times let thepassengers lose the benefits of on-demand operation. 0iscompromise and the possibility of rejecting unprofitabletransport requests are largely unexplored in the ODAMliterature and deserve further attention. In this context, thebehavior, aspirations, and acceptance of passengers shouldalso be examined in detail. Studies that formally define andverify such standard problem formulations in trans-portation, e.g., in terms of traffic assignment [81], facilitylocation [82], and location routing [83], are among the mostfrequently cited papers in their fields.

3.2. Creating Reference Benchmark Datasets for ODAM.When formulating appropriate ODAM research problems,there is a need to evaluate and compare required solutiontechniques, design decisions, and heuristics. 0is impliesthat once such research problems are clear and standardized,it must be possible to understand the state of the art, which ismuch easier when the same terminology is used. Otherresearch areas, for example, have benefited massively fromthe establishment of reference datasets. For example, theexistence of small-scale databases such as CAB [84] in hublocation research [85], the equivalent of vertiport locationproblems in ODAM research projects, led to fair perfor-mance evaluations among the exact solution techniques andis considered the indispensable gold standard for a wholerange of hub location problem types. Although the networkconsists of only 25 nodes, it has advanced this area further asits optimal solutions are well known to the community andeasily accessible at all times. Over time, the desire to solveeven larger problems has led to the development of addi-tional standard datasets of increasing size, e.g., Turkishpostal system dataset [86] and Australian postal dataset [87].

A starting point for ODAM datasets is to build onexisting ground transport datasets, which usually containinformation about locations (often in zones) as well asdistances and requirements between these locations. Forexample, Transportation Networks for Research [88]maintains a collection of such datasets for several regions ofthe world. Table 6 summarizes the network data for some

selected datasets. It becomes evident that these networkscover a broad spectrum of regions and scales. Together withpublicly available reference contracts and jointly agreedbenchmark measures (gap, execution time, loss of profit,etc.), we expect that such and other benchmark datasets,such as [89], will further shift the boundaries of successfulODAM research.

Regarding the creation of reference datasets, approachesin the literature on simulated data should be combined withreal-world data. We propose Guangzhou, China, as an ex-cellent candidate for such reference benchmarks. Guangz-hou is the capital of Guangdong Province in southern China.First of all, Guangzhou is at the heart of the Pearl Rivermetropolitan area, one of the largest urban agglomerationsin the world, with an urban population of approximately 15million in the year 2018. While Guangzhou has a rather well-established public transportation infrastructure, with bus/subway/taxi networks and the third largest airport in China,yet, induced by its tremendous travel demands, it facesregular grid locks along its highways, i.e., there is a strongneed for novel transportation solutions. 0e government ofGuangzhou and China are pushing towards modern tech-nologies, which make the implementations/simulations ofODAM taxis much easier than in conservative regions, suchas, for instance, Germany. Moreover, EHang, the Chineseautonomous aerial vehicle company, has tested its firstODAM taxi network in Guangzhou on August 9, 2019. Withthis pilot program, EHang will help the Guangzhou gov-ernment to set up a command-and-control center and tobuild up basic infrastructures to support ODAM, and moreflight routes and ODAM ports would be tested as well beforethe real commercial operations. 0is makes Guangzhou anexcellent testbed for preliminary scientific studies and earlyprototypical implementation in practice.

3.3. Development of a Universal, Multimodal ODAM Trans-portation Competition/Cooperation Model. Previous studieson ODAM services preferably focus on a single urban/re-gional area and report on specific results. Especially becauseregions around the world have their own specifications, it isnot clear how universal the results of such studies are andhow general conclusions could be transferred to other re-gions. 0e aim should therefore be to derive a universalmodel that, for example, provides predictions about thecompetitiveness of ODAM services in the presence of aspecific standardized transportation infrastructure. In thisway, general findings can also be adapted to other regions.Also, the understanding of the infrastructure as multimodalis decisive for the interpretability and validity of the studies.In our review, we found that most studies assume thatODAM services will compete with only one mode, especiallywith cars. In reality, however, ODAM must compete withthe sum of all transport options in a given study area, whichincludes a well-developed public transport infrastructure,novel means of transportation such as electric scooters, andhigh-speed alternatives such as high-speed rail in the re-gional case. Without taking into account all the facets of theavailable modalities, the significance of the study results is


rather limited. While the results of such analysis are highlycity-/region-specific, depending on its geography, thereshould be the goal of producing universal conclusions thatcan be easily transferred. Naturally, researchers have totarget a tradeoff between being too specific and operationallytoo far from being applicable. Clustering similar regions is astarting point for this research.

One reason for limiting coverage in previous studies isthat researchers often conduct studies in regions for whichdata are available. However, given the recent progress inOpen/Linked Data, there is an abundance of data. Moreimportantly, organizations, researchers, and volunteersaround the world have made enormous efforts to make dataavailable at the planetary level [90, 91]. A few examples ofsuch datasets are shown in Table 7, covering a wide range oftransport modes, routing services, and economic data. Withthese data, it is possible to consistently compare the potential

of ODAM services across the planet and identify com-monalities and dissimilarities, with the ultimate target of auniversal multimodal transportation model that makesODAM competition accessible.

4. Conclusions

0e focus of our review is on the operational issues ofODAM. A large number of studies have appeared in theliterature addressing this topic; our survey concerns a total of100 references published in recent years. While there areseveral other issues in ODAM, e.g., vehicle design or pro-pulsion system, there exist excellent reviews in these areas,e.g., on vehicle design [15], current technology for electronicVTOL drones [16], and propulsion technology [17]. Webelieve that our work on the operational issues in ODAMcomplements these existing reviews, helps to advance the

Table 6: An overview of selected transportation networks for ODAM research at https://github.com/bstabler/TransportationNetworks.

Network Zones Links Nodes FolderAnaheim 38 914 416 TransportationNetworks/tree/master/AnaheimAustin 7388 18961 7388 TransportationNetworks/tree/master/AustinBarcelona 110 2522 1020 TransportationNetworks/tree/master/BarcelonaBerlin Center 865 28376 12981 TransportationNetworks/tree/master/Berlin-CenterBirmingham, England 898 33937 14639 TransportationNetworks/tree/master/Birmingham-EnglandChicago-sketch 387 2950 933 TransportationNetworks/tree/master/Chicago-SketchChicago 1790 39018 12982 TransportationNetworks/tree/master/chicago-regionalEastern Massachusetts 74 258 74 TransportationNetworks/tree/master/Eastern-MassachusettsGold Coast, Australia 1068 11140 4807 TransportationNetworks/tree/master/GoldCoastHessen-asymmetric 245 6674 4660 TransportationNetworks/tree/master/Hessen-AsymmetricPhiladelphia 1525 40003 13389 TransportationNetworks/tree/master/PhiladelphiaSydney, Australia 3264 75379 33837 TransportationNetworks/tree/master/SydneyTerrassa-asymmetric 55 3264 1609 TransportationNetworks/tree/master/Terrassa-AsymmetricWinnipeg 147 2836 1052 TransportationNetworks/tree/master/Winnipeg

Table 7: An overview of global data sources for ODAM research.

Data Organizations Links

Air transportation

Official Airline Guide (OAG) http://www.oag.com/International Civil Aviation Organization (ICAO) http://www.icao.int/

Airport Council International http://www.airports.org/Sarbe Airport Data Intelligence http://www.sabreairlinesolutions.com/

Open Flights http://openflights.org/Innovata Flight Schedules http://www.innovata-llc.com/

Railway transportationInternational Union of Railways (UIC) http://www.uic.org/

OpenRailwayMap https://www.openrailwaymap.org/Railway Directory http://www.railwaydirectory.net/

Road transportation OpenStreetMap https://www.openstreetmap.org/

Routing services

Open Source Routing Machine https://project-osrm.org/Google Maps https://maps.Google.com/Baidu Maps https://map.baidu.comTransitland https://transit.land/GTFS Router https://atfutures.github.io/gtfs-router/

Population/economic data

Gridded Population of the World https://sedac.ciesin.columbia.edu/data/collection/gpw-v4/World Bank Open Data https://data.worldbank.org

UN World Population Prospects https://population.un.org/wpp/International Monetary Fund https://www.imf.org/en/data/Global Change Data Lab https://global-change-data-lab.org/


https://github.com/bstabler/TransportationNetworkshttp://www.oag.com/http://www.icao.int/http://www.airports.org/http://www.sabreairlinesolutions.com/http://openflights.org/http://www.innovata-llc.com/http://www.uic.org/https://www.openrailwaymap.org/http://www.railwaydirectory.net/https://www.openstreetmap.org/https://project-osrm.org/https://maps.Google.com/https://map.baidu.comhttps://transit.land/https://atfutures.github.io/gtfs-router/https://sedac.ciesin.columbia.edu/data/collection/gpw-v4/https://sedac.ciesin.columbia.edu/data/collection/gpw-v4/https://data.worldbank.orghttps://population.un.org/wpp/https://www.imf.org/en/data/https://global-change-data-lab.org/

understanding of ODAM, and paves the way towards itssuccessful implementation.

Eventually, the three research challenges mentionedabove lead to the overarching goal of research reproduc-ibility. With existing, well-defined reference benchmarksand the availability of large amounts of data, it would bepossible to reproduce and understand results from previousstudies. A crucial prerequisite for this would be that authorsmake their additional data and codes available to the publicas has become common in other areas in recent years, e.g.,health care [92], geography [93], public sector [94], andeducation [95].0e Open Science Initiative [96] also calls forsuch an approach by making scientific data collected directlyas part of an experiment or indirectly as a byproduct of adownstream analysis available to the public and online [97].In addition to objective reproducibility, the provision ofresearch data also benefits the overall effectiveness of sci-entific processes [98–101], which, for instance, also results ina robust citation advantage [102].

In addition to pure access to raw data, future projects inthe field of ODAM research should be commonly planned,coordinated, and carried out. 0is also implies the will-ingness to make more in-depth findings and analyses ac-cessible, as well as voluntary support for subsequent projects.One possibility for this is, for example, the establishment andfounding of interest and working groups, which are com-mitted to the ongoing observance of these requirements andwhose members are representatives of various researchinstitutions from different countries. At central conventions,conferences, and association meetings, essential progressshould be recorded and evaluated so that work steps basedon this can be derived. We are therefore committed toremoving barriers to the availability and reusability of sci-entific data, codes, and study results used throughoutODAM research. 0is is crucial for reliable and long-termdecisions about our future multimodal transport systems.

Data Availability

No data were used as the input for this review. 0e data asreported from other studies are subject to availability in thedata tables and original publications.

Conflicts of Interest

0e authors declare that they have no conflicts of interest.

Acknowledgments

0is study was supported by the National Natural ScienceFoundation of China (Grant nos. 61861136005,61851110763, and 71731001).

References

[1] H. Wen, J. Sun, and X. Zhang, “Study on traffic congestionpatterns of large city in China taking beijing as an example,”Procedia-Social and Behavioral Sciences, vol. 138, pp. 482–491, 2014.

[2] A. Anas and G. R. Timilsina, Impacts of Policy Instruments toReduce Congestion and Emissions from Urban Trans-portation: =e Case of Sao Paulo, Brazil, 0e World Bank,Washington, DC, USA, 2009.

[3] R. Arnott and K. Small, “0e economics of traffic conges-tion,” American Scientist, vol. 82, no. 5, pp. 446–455, 1994.

[4] B. J. Holmes, M. H. Durham, and S. E. Tarry, “Small aircrafttransportation system concept and technologies,” Journal ofAircraft, vol. 41, no. 1, pp. 26–35, 2004.

[5] J. Schippl, M. Decker, and T. Fleischer, “Personal air vehiclesas a new option for commuting in europe: vision or illusion?”in Proceedings of the 41st European Transport Conference(ETC 2013), Frankfurt, Germany, September 2013.

[6] B. J. Holmes, “A vision and opportunity for transformationof on-demand air mobility,” in Proceedings of the 16th AIAAAviation Technology, Integration, and Operations Conference,AIAA AVIATION Forum, Atlanta, GA, USA, June 2016.

[7] J. Holden and N. Goel, “Fast-forwarding to a future of on-demand urban air transportation,” Tech. Rep., Uber Elevate,San Francisco, CA, USA, 2016.

[8] U Elevate, =e Future of Urban Mobility, U Elevate, SanFrancisco, CA, USA, 2019, https://www.uber.com/in/en/elevate/.

[9] Lilium, Lilium Webpage, 2019, https://lilium.com/.[10] Kittyhawk, Kittyhawk Webpage, 2019, https://kittyhawk.

aero/.[11] Airbus, My Kind of Flyover, 2019, https://www.airbus.com/

newsroom/news/en/2016/12/My-Kind-Of-Flyover.html.[12] R. W. T. H. Aachen, Silent Air Taxi, 2019, https://airspace-

aachen.de/.[13] X. Sun, S. Wandelt, and A. Zhang, “How did covid-19 impact

air transportation? a first peek through the lens of complexnetworks,” Journal of Air Transport Management, vol. 89,Article ID 101928, 2020.

[14] X. Sun, S. Wandelt, and A. Zhang, “Dissection of a singu-larity: the impact of covid-19 on aviation,” Journal of In-ternational Logistics and Trade, vol. 18, no. 3, pp. 113–125,2020.

[15] E. Stumpf, M. Kreimeier, P. Strathoff, J.-M. Lueckhof, andK.-U. Schroeder, “Small aircraft concept for regional on-demand air mobility,” in Proceedings of the 31st ICASCongress, Belo Horizonte, Brazil, September 2018.

[16] N. Polaczyk, E. Trombino, P. Wei, and M. Mitici, “A reviewof current technology and research in urban on-demand airmobility applications,” in Proceedings of the Vertical FlightSociety Autonomous VTOL Technical Meetings and ElectronicVTOL Symposium, Mesa, Arizona, January 2019.

[17] H. D. Kim, A. T. Perry, and P. J. Ansell, “A review of dis-tributed electric propulsion concepts for air vehicle tech-nology,” in Proceedings of the 2018 AIAA/IEEE ElectricAircraft Technologies Symposium (EATS), pp. 1–21, IEEE,Cincinnati, OH, USA, July 2018.

[18] M. Fu, R. Rothfeld, and C. Antoniou, “Exploring preferencesfor transportation modes in an urban air mobility envi-ronment: munich case study,” Transportation Research Re-cord, vol. 2673, no. 10, pp. 427–442, Article ID0361198119843858, 2018.

[19] M. Decker, T. Fleischer, S. Meyer-Soylu, and S. Jens, “Per-sonal air vehicles as a new option for commuting in Europe:vision or illusion,” in Proceedings of the European TransportConference, pp. 1–20, Frankfurt, Germany, October 2013.

[20] S. Rajendran and J. Zack

operationalconsiderationsregardingon-demandairmobility ...odam has the po-tential to radically...

Documents