Demand Oriented Mobility Solutions for Rural Areas Using Autonomous Vehicles

By Moritz von Mörner, Technische Universität Darmstadt, Institute for Transport Planning and Traffic Engineering

This study was funded by Opel Automobile GmbH and DB Regio AG

1.   Introduction

Conventional public transport in rural areas is facing major difficulties, especially in countries exhibiting demographic and societal changes. In rural areas in Germany demographic change and the accompanying decrease in population and pupil numbers endanger the already bristle financial situation of public transport.

Accordingly, in most rural areas only major links to and from cities are serviced by public transport. The offered services mainly are an extension to school bus service, adding only few additional trips to the schedule to assure a minimum of mobility for the public. Most of the mobility demand in rural areas is covered using private vehicles. Considering demographic change, more seniors will become dependent on mobility through private vehicles, who might not want to or should not drive anymore.

This paper explores the opportunities presented through the dawn of autonomous vehicles and their application in a new mode of public transport. This study is conducted exemplarily on a rural administrative district (AD) in Northern Bavaria in Germany. Using a household mobility survey available from that district, the public transport demand is modelled as trip requests and serviced with a new transport mode utilising shared autonomous vehicles (SAV).

Conducting this study, major trends are taken into consideration to estimate public transport demand and supply for rural areas in Germany in the future.

This study delves into answering questions about feasibility, vehicle utilisation and accessibility through SAVs. A model of a typical rural area in Germany is established to show the implementation of SAV service.

The focus of the model is not to represent all traffic in the rural area but only to show the new public transport mode SAV. The main results will be information about the number of vehicles needed to achieve a certain level of service during peak hours, average passenger detour per trip, overall vehicle kilometres driven daily and average billable trip-km per passenger and vehicle.

This paper sets the focus on the first results applying 4-seated SAVs to cover a demand of about 30% in the underlying administrative district. 

2.   Background (policy and society)

Flexible mobility solutions e.g. carpooling have been tried repeatedly in Germany. However, they have never caught on and most public trials featured major financial problems. Today most carpooling services supporting public transport are subsidized through local governments. They are merely an addition to regular taxi services mostly confined to one municipality. In some areas, partially flexible bus services are gaining momentum. This is especially evident in rural areas with high tourism potential.

However, today, conventional carpooling is picking up momentum through a rise in the sharing economy. Services like Uber and Lyft are in successful operation in urban areas in the United States and car sharing schemes are gaining attention in German cities as well [Shaheen, Mallery, and Kingsley 2012; Shaheen, Stocker, and Bhattacharyya 2016].0

With the example of Lisbon, it has already been shown that shared taxi services could reduce the number of cars in large cities by up to 80% when getting rid of private vehicles altogether [KLOTH Michael 2015]. Similar studies have been conducted in Hamburg [kirschner 2016] and Stuttgart [Friedrich and Hartl 2017].

This study however will focus on sharing rides in rural areas, where there are different conditions compared to inner city mobility. Average trip distance is longer and there are potentially fewer people that could share a ride [Statistisches Bundesamt 2016].

In the dawn of autonomous driving, flexible mobility solutions surface again. New economically viable options for carpooling are appearing. On the supply side, getting rid of personnel in transit vehicles is hoped to be linked to reduced operation costs and reduced fares. In Germany personnel costs in public transport make up of about 1/3 of all operational costs (reduction from 55 % of operational costs down to 22 % are thinkable) [Loos 2016; Bundestag 2016].

2.1 Definition and Characteristics of Rural Areas in Germany

Defining rural areas in Germany, one can resort to the BBSR (Bundesinstitut für Bau‑, Stadt- und Raumforschung) which distinguishes between four different spatial settlement structures [BBSR 2014].

  • City: A City has more than 100.000 inhabitants.
  • Urban districts: Districts with a at least a 50 % share of the population living in a density of at least 150 people/km² or districts with a density (without cities) of at least 150 people/km².
  • Rural districts with agglomeration tendencies: Districts with a 50 % share of the population in cities under 150 people/km2 and districts with an average density of 100 people/km2.
  • Scarcely populated rural districts: Districts with a population share in cities under 50 % and average densities lower than 100 people/km2.

Even though people are moving to cities [Statistisches Bundesamt 2016; SHELL 2014; Chilla, Morhart, and Braun 2008] still about 50 % of the population lives in rural areas in Germany. This part of the population could face major difficulties with declining public transport services. On the other hand, an increase in service quality could improve living conditions and reduce the dependencies on private car ownership.

2.2 Autonomous Driving (as utilised in this study)

This study takes for granted the ubiquity of safely functioning autonomous vehicles in a few years. Additionally, this study opts for the implementation of fully electric vehicles. However, in this step battery capacity, charging time and location of chargers are not considered.

2.3 Trends (changes in rural districts)

Public transport in Germany is a public service and supposed to facilitate mobility for everybody. However, due to cost constraints, public transport in rural areas is often merely an absolute minimum service, building on school bus service to obtain a certain economic feasibility.

Implementing a public transport built on autonomous shared vehicles might be the major step to an affordable, flexible, good quality service that might even be cost effective.

Conventional public transport is in peril in rural areas. Several societal trends are affecting people’s lives and the public transport customer base as well. The following figure depicts societal mega trends and their effects on demand and supply of mobility. These trends feed into changes in transport operation form, transport operation model and vehicle concepts.

Figure 1: Trends affecting rural districts in Germany

3.   Overall model structure

The model is written in Anylogic 8. Anylogic 8 is a java based multi method simulation environment. The simulation created for this research is an agent based model, representing a realistic simplification of a public transport operation using SAVs. The following agents were created for building the model:

  • Stops
  • Passengers
  • Dispatcher
  • Vehicle (SAV)

Furthermore, the model accounts for the following specifications:

  • Model duration 24+ hours (model runtime 20:00 day before till 03:00 day after)
  • Simulation of only the new transport mode
  • Not a complete traffic simulation (no delay, no traffic jams, no differences in travel times due to congestion)
  • Travel times and driving speeds are obtained from OSM (see data used)
  • Dispatcher tries to gain high vehicle occupancy rate
  • Maximum detour coefficient per passenger
    (actual travel time / optimum travel time per passenger)
  • Demand – all “public transport” trips except:
    • Trips with a duration < 5 minutes as these trips are most likely conducted on foot or by bike (duration < 1 minute for seniors)
    • School trips conducted via rail
    • School trips along major arteries towards the one major school site (demand for school trips would highly exceeds the overall demand peak, as all students must be at school at the same time)

4.   Data used

Stops

Theoretically, door to door service is possible using autonomous vehicles. However, for this model, a less dense stop distribution is chosen. Under the assumption that almost everybody can bear a small offset between door and nearest stop, stops with a 100 m catchment area were placed covering most of the settlements in the district excluding single houses and farms. The model incorporates 1,588 stops (a detail of the map is shown in Figure 2). However, in light of providing mobility for everybody, impaired citizens would need the option to be picked up or dropped off directly at the door. (Figure 2: Map showing stops in red (100 m catchment area). map data OSM/Geofabrik.de July 2017)

Demand

Demand is extrapolated from a mobility household survey conducted in an administrative district (AD) in Bavaria, Germany from 2014. The household survey showed an overall return rate of around 14 %, which is unusually high for such a survey conducted in Germany. Students showed some overrepresentation due to the survey being promoted in schools. The extrapolation and normalisation was done utilising demographic information for the administrative district for 2012 (status quo) and 2030 (prognosis) [Bertelsmann Stiftung 2017].

The trips requested in the model do not represent the whole extrapolated mobility demand. Different modal splits for SAV usage can be implemented through a uniformly distributed variable “transit marker” (shown in Figure 3). “Trip requests” depicts the time difference between reservation of the trip and actual requested departure time. “Flexibility” depicts the time interval in which the trip can begin (here +/- 30 minutes from requested departure time). All Stops are assigned a unique ID to be addressed in the model.

Passenger type Origin
ID
[-]
Age Trip
reason
Desti- nation
ID [-]
Trip
request [hours]
Flex-ibility [min] Requested departure [hh:mm] transit marker
[1-100]
Min. trip [min]
working 450307 58_ work 9901100 -2__ 30_ 06:45__ 62__ 5_
student 350307 8_ school 350110 -2__ 30_ 06:00__ 84__ 5_
other 150801 29_ supply 123201 -2__ 30_ 08:10__ 13__ 5_
senior 450307 73_ supply 9901100 -2__ 30_ 09:10__ 45__ 1_

Figure 3: Example of trip demand data utilised in the model

 

Supply

On the supply side, the focus lies on smaller vehicles than average public transport vehicles to maximise flexibility. This model can incorporate vehicles with 4 and/or 8 seats. The vehicle fleet size can be set as varying parameter with the goal of servicing all trip requests. Using the Agent Dispatcher that is adjusted for high efficiency in vehicle use, an excess supply of vehicles can be detected and can be reduced for further research, if needed. However, the vehicle fleet size should not be chosen too large as the Agent Dispatcher runs through all vehicles for every trip request resulting in longer model run times. The dispatcher prioritises the vehicle choice firstly for vehicles that already have passengers and secondly for the vehicle detour.


 

Routing

Routing is done via the street network available through OSM-Maps [© OpenStreetMap Creative Commons BY-SA 2.0]. The maps are obtained through Geofabrik GmbH. Fastest routes in between all stops are then determined utilising speed limits for driving speed. These driving durations are used in an O-D-Matrix covering roughly 2.5 million trip combinations. This O-D-Matrix is utilised in the Agent Dispatcher for the determination of a good travel time for each trip request.

Fleet Size

Fleet size is used as a parameter to determine good operation conditions for servicing most of the modelled trip requests.

Vehicle Size

Vehicle sizes are chosen in similarity to vehicles available today. Focusing on a rural area, with a disperse population a high mean occupancy in SAVs is very unlikely. 4- and 8-seated SAVs are therefore chosen as suitable vehicles.

5.   Model results

For this paper, different scenarios were examined. Fleet size was varied after initial testing from 350 vehicles to 800 vehicles in steps of 50 vehicles. Three different maximum detour coefficients were used as well as two different values for the parameter “flexibility” in trip starting time (shown in Figure 4). Trip requests and therefore the modal split of SAVs are aimed at around 30 % of all trips in 2030. However, only trips within the AD are modelled. The user groups are modelled with different modal split values (estimation for 2030). Trips with a duration under a certain threshold (shown in Figure 5) were excluded under the assumption that short trips are being conducted on foot or by bike. After identifying a reasonable parameter range 60 model runs were conducted.

    Average quality service High quality service Taxi Service
Max. Detour coefficient 1.4 X X        
1.2     X X    
0         X X
Flexibility [minutes] 30 X   X   X  
10   X   X   X

Figure 4: Scenario overview

 

 

User group Working Student Apprentice Senior Other
Modal split in SAVs
base: all trips
35 % 15 % 50 % 45 % 40 %
Minimum travel duration

[minutes in a vehicle]
5 5 5 1 5

Figure 5: Utilised modal split for each user group

 

5.1 Extreme Model Results

In this study only 4-seated SAVs were utilised. After initial testing, model results showed that larger vehicles would only shortly be occupied with more than 4 passengers, as can be seen in Figure 6.

Additionally, in Figure 7 a taxi service is shown with a maximum detour coefficient of 1. This leads to a very low utilisation of carpooling. Most vehicles are occupied by just 1 passenger, as detours are prohibited. Only passengers with the same origin and destination can share a ride.

Figure 6: Vehicles’ occupancy for 30% demand and trips restricted to AD (500 8-seated vehicles;
max. detour coefficient 1.4; flexibility +/- 30 minutes)

 

Figure 7: Vehicles’ occupancy for 30% demand and trips restricted to AD (500 4-seated SAVs;
max. detour coefficient 1.0; flexibility +/- 30 minutes)


 

5.2 Detailed Model Results

In Figure 8 to Figure 11 results of a single model run are shown. This run was conducted with a maximum detour coefficient of 1.4, a flexibility of +/- 30 minutes, a fleet size of 500 SAVs and the restriction of only modelling trips with origin and destination inside the administrative district.

All trip requests are serviced, the mean detour coefficient settles at 1.22 and not all vehicles are utilised at the same time. High occupancy rates are found especially during peak hours, as can be expected. At maximum 470 vehicles are utilised at the same time in the afternoon peak.

The fleet covers 246,458 veh-km, with 37,820 veh-km travelled without passengers. The sum of all optimum trip-km (all trip requests without detours) totals at 357,391 km. In a kilometre based tariff, this would be the total billable distance.

Figure 8: Trip Requests for 30% demand and trips restricted to AD (500 4-seated SAVs;
max. detour coefficient 1,4; flexibility +/- 30 minutes)

Figure 9: Distances for 30% demand and trips restricted to AD (500 4-seated SAVs;
max. detour coefficient 1,4; flexibility +/- 30 minutes)

Figure 10: Number of vehicles for 30% demand and trips restricted to AD (500 4-seated SAVs;
max. detour coefficient 1,4; flexibility +/- 30 minutes)

Figure 11: Vehicles’ occupancy for 30% demand and trips restricted to AD (500 4-seated SAVs;
max. detour coefficient 1,4; flexibility +/- 30 minutes)

 

Figure 12: Km per vehicle per day + operation hours per vehicle per day for 30% demand and trips restricted to AD (500 4-seated SAVs; max. detour coefficient 1,4; flexibility +/- 30 minutes)

5.3 Parameter Variation

For identifying good parameter combinations, variations of different fleet sizes, the maximum detour coefficient and flexibility were modelled. Figure 13 shows the model results for a maximum detour coefficient of 1,4, a flexibility of +/- 30 minutes and varying fleet sizes (fleet size = 500 in bold). With a fleet of 500 SAVs or larger, all trip requests can be served. Increasing fleet size leads to reduced unoccupied vehicle kilometres. Rerouting vehicles to the next trip origin is not necessarily needed, as more vehicles are available to service trip requests. Additionally, the share of billable km increases, seemingly converging at around 800 vehicles. The average vehicle kilometres travelled by each vehicle decreases rapidly from almost 500 veh-km/veh (500 SAVs) to around 300 veh-km/veh (800 SAVs), a distance that could be covered by today’s electric vehicles. Simultaneously, the average operating hours per vehicle drop from 10 hours (500 SAVs) to 6 hours per vehicle (800 SAVs). However, these are just average values. Lowest and largest value vary greatly (as shown in Figure 15). For this model, adding 300 vehicles (from 500 to 800) leads to a reduction of about 10,000 km of total distance covered.

As can be expected, taxi services with a maximum detour coefficient of 1.0 cannot surpass the vehicle kilometres driven with the billable distance. Here, less than 80 % of the vehicle kilometres could be billed to passengers in a km based tariff.

Parameters Demand: 30%, restricted to administrative district, max. detour coefficient: 1,4, flexibility +/- 30 minutes

4-seated SAVs

Fleet size 350 400 450 500 550 600 650 700 750 800
Trip requests 29,922 trip requests
Trip requests served 28,312 29,365 29,816 29,922
Trip requests not served 1,610 557 106 0
Trip requests
not served
5.4% 1.9% 0.4% 0%
Distances
Vehicle kilometres [km] 234,960 242,885 248,069 246,458 243,908 242,705 240,362 237,362 235,397 236,075
Vehicle kilometres occupied [km] 195,353 202,642 207,578 208,638 208,538 208,429 208,366 207,609 207,743 208,706
Billable distance [km] 338,151 350,873 356,106 357,391 357,391 357,391 357,391 357,391 357,391 357,391
Driven without pax [km] 39,607 40,243 40,492 37,820 35,371 34,276 31,996 29,753 27,654 27,369
Vehicle kilometres unoccupied 16.8% 16.5% 16.3% 15.4% 14.5% 14.1% 13.3% 12.5% 11.8% 11.6%
Billable kilometres 143.9% 144.4% 143.5% 145.0% 146.5% 147.3% 148.7% 150.6% 151.8% 151.4%
Average results per vehicle
Vehicle kilometres [km] 671 607 551 493 443 405 370 339 314 295
Operating hours [hh:mm] 13:25 12:09 11:02 9:53 8:53 8:07 7:25 6:48 6:18 5:55
Pax served 81 73 66 60 55 50 46 43 40 37
Billable distance [km] 966 877 791 715 650 596 550 511 477 447
Average results per passenger
Detour 1,21
Distance covered [km] 8.30 8.27 8.32 8.24 8.15 8.11 8.03 7.93 7.87 7.89
Billable distance [km] 11.94 11.95 11.94 11.94 11.94 11.94 11.94 11.94 11.94 11.94

Figure 13: Parameter variation of fleet size 350-800 vehicles for 30% demand and trips restricted in AD
(4-seated SAVs; max. detour coefficient 1.4; flexibility +/- 30 minutes) (fleet size = 500 in bold)

 

 

Figure 14: Share of billable trip km (sum of all services trip requests / km covered by vehicles)

 

Figure 15: Average billable trip-km with the vehicle with the lowest and highest number of veh-km

 

6.   Findings

The model results show a high mean value for vehicle kilometres per vehicle and day. With around 500 veh-km per vehicle and day, every vehicle would cover almost 200,000 km per year. This could lead to a potentially short life span for each vehicle, possibly needing repairs or replacements every year.

Furthermore the results show, that carpooling can work, even in rural areas. Increasing flexibility and maximum detour coefficient seem to have a positive effect on vehicle utilisation. However, the differences are rather small.

8-seated SAVs do not seem suitable for this rural area, as vehicles would barely reach a capacity of 4 passengers. However, trips to and from school would need to be organised separately in larger vehicles. The sheer number of 4-seaters could not be handled in front of schools. Additionally, there is the chance that vehicles utilised for school trips need to be cleaned after each use.

Looking at the distribution of operating hours per day (in Figure 12), it becomes obvious that with more than 50 % of all vehicles driving more than 8 hours, an operation with driver would most likely not be feasible.

The model runs show that – depending on fleet size – some vehicles exhibit a high value for veh-km/day. These distances are larger than today’s average electric vehicle’s range. The current model does not account for battery capacity and charging duration. However, this information is crucial for the determination if an economically sound service would be possible.

Small fleet sizes lead to long operating hours for each vehicle. Not considered in this study are service and cleaning times. However, with operating hours of 15 hours and more, this should be considered in a next step.

It is very likely that the household survey does not cover all trips conducted in this administrative district and trip chains were not implemented in the survey. Additionally, the extrapolation towards a mobility demand for the year 2030 might be imprecise and changes in behaviour can only be predicted so far. Therefore, trip demand and trip duration has to feature some inaccuracies.

The model does not incorporate an optimisation of vehicle allocation or the possibility to change passengers to a vehicle with a better fit, after the vehicle has been set. Through an optimisation of all allocated trips, a lower number of vehicles might be achievable.

Lastly, the scope of this paper only covers trips inside the AD. Including trips out of and back into the AD will lead to more complex traffic patterns, a different allocation of passengers to vehicles and probably to a larger vehicle fleet.

7.   Future work

The limitations mentioned before are a good starting point for further research. Different parameter combinations need to be examined.

This study shows in a plausible way that more kilometres can be billed than are covered by SAVs. Economic feasibility needs to be examined with kilometre costs and possible production and operation costs.

As this study does not include a complete optimisation, different optimisation parameters need to be researched, especially with school trips and school starting times in mind.

For now, trips are scheduled at a fixed time before the requested starting time. However, this is not very realistic. Distribution in request times should be incorporated into the model, even considering spontaneous requests.

However, most trips will most likely be daily trips to and from work and to and from school, that could get optimised days before.

8.   SUMMARY AND CONCLUSIONS

Shared autonomous vehicles can not only be utilised in cities. SAVs can cover parts of the demand in a very efficient manner in rural areas, as well. This could decrease the number of cars in households considerably. Even though, additional demand will probably be induced, especially from those, who are not able to drive a vehicle themselves.

References

Bertelsmann Stiftung. 2017.
Demographiebericht. Bertelsmann Stiftung.

Bundestag, Deutscher. 2016.
Bericht Der Bundesregierung Über Die Entwicklung Der Kostendeckung Im Öffentlichen Personennahverkehr.

Chilla, Tobias, Tobias Morhart, and Boris Braun. 2008.
“Beyond the Commuter Belt: Graduates’ Plans for Migration and Demographic Change in Rural Areas.” Raumforschung Und Raumordnung 66 (3): 260–70. doi:10.1007/BF03183161.

Friedrich, Markus, and Maximilian Hartl. 2017.
Megafon.

Kirschner, Benjamin. 2016.
“Erneuerung Der Städtischen Mobilität,” November, 1–30.

Kloth, Michael. ITF COM. 2015.
“Urban Mobility System Upgrade,” April, 1–36.

Loos, P. 2016.
“Vergleichende Wirtschaftliche Bewertung Des Einsatzes Konventioneller Und Selbstfahrender Fahrzeugeinheiten Im ÖPNV.”

Shaheen, Susan A, Mark A Mallery, and Karla J Kingsley. 2012.
“Personal Vehicle Sharing Services in North America.” Research in Transportation Business & Management 3 (August): 71–81. doi:10.1016/j.rtbm.2012.04.005.

Shaheen, Susan, Adam Stocker, and Abhinav Bhattacharyya. 2016. “Multimobility and Sharing Economy,” July. Washington, D.C., 1–22.

SHELL. 2014.
Shell PKW-Szenarien Bis 2040.

Statistisches Bundesamt. 2016.
Datenreport 2016. Bonn.