Scholarly article on topic 'Exploring Co-Modality Using On-Demand Transport Systems'

Exploring Co-Modality Using On-Demand Transport Systems Academic research paper on "Social and economic geography"

Share paper
Academic journal
Transportation Research Procedia
OECD Field of science
{co-modality / "demand-responsive transportation" / simulation}

Abstract of research paper on Social and economic geography, author of scientific article — Nicole Ronald, Jie Yang, Russell G. Thompson

Abstract Integrating passenger and freight transport systems, known as co-modality, is becoming more feasible due to recent developments in information and communication technologies (ICT) such as smart phones and global position systems (GPS). This paper uses simulation of an on-demand transportation scheme in which passengers and parcels can travel together to explore the benefits of co-modality when compared to existing schemes. It is shown that, depending on the demand, co-modality can provide improved experiences for both operators and passengers/customers.

Academic research paper on topic "Exploring Co-Modality Using On-Demand Transport Systems"

Available online at


Transportation Research Procedia 12 (2016) 203 - 212


The 9th International Conference on City Logistics, Tenerife, Canary Islands (Spain), 17-19 June

Exploring Co-Modality Using On-Demand Transport Systems

Nicole Ronald a, Jie Yang a, Russell G. Thompson a

aDepartment of Infrastructure Engineering, The University of Melbourne, Melbourne, Australia


Integrating passenger and freight transport systems, known as co-modality, is becoming more feasible due to recent developments in information and communication technologies (ICT) such as smart phones and global position systems (GPS). This paper uses simulation of an on-demand transportation scheme in which passengers and parcels can travel together to explore the benefits of co-modality when compared to existing schemes. It is shown that, depending on the demand, co-modality can provide improved experiences for both operators and passengers/customers.

© 2016 The Authors.PublishedbyElsevierB.V. Thisis an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organising committee of the 9th International Conference on City Logistics Keywords: co-modality; demand-responsive transportation; simulation

1. Introduction

Co-modality involves the use of alternative modes for increasing the efficiency and sustainability of transport systems (Commission of the European Community, 2006). Recently public transport systems have been considered for urban freight transport. Integrating passenger and freight transport systems is becoming more feasible due to recent developments in information and communication technologies (ICT) such as smart phones and global position systems (GPS).

A combination of passenger and freight transport can be realized using buses or taxis for carrying goods as well as passengers. Passenger transport companies can benefit from carrying goods by utilising space on less crowded vehicles and shippers benefit by having a convenient courier service as an option (Thompson and Taniguchi, 2014).

Demand Responsive Transport (DRT) is a user-focused form of public or private transport with flexible scheduling and flexible routing of small- or medium-sized vehicles to pick-up and drop-off passengers to their desired locations in a shared-ride mode. Traditionally DRT services require pre-booking (e.g., bus on demand), and/or are designed to provide transport from/to a hub (e.g., airport or hotel shuttles), with varying degrees of automated routing and

2352-1465 © 2016 The Authors. Published by Elsevier B.V. This is an open access article under the CC BY-NC-ND license (http://creativecommons.Org/licenses/by-nc-nd/4.0/).

Peer-review under responsibility of the organising committee of the 9th International Conference on City Logistics doi:10.1016/j.trpro.2016.02.059

dispatching. DRT can provide the accessibility and personal security benefits of taxi services at a reasonable cost to the user.

Simulation of DRT combines both optimisation and transport modelling to explore the performance of different schemes under different demands. A review of individual-based DRT simulation can be found in Ronald et al. (2015b).

This paper uses simulation of a parcel/passenger on-demand transport scheme to explore its performance in different scenarios, particularly varied demand. An evening scenario is demonstrated, in which passengers would like to travel to a restaurant to eat, and customers would like to order take-away to be delivered to their home. The vehicles operate as either passenger, parcel, or co-modal vehicles. Household travel survey data from Melbourne, Australia is used to generate demand for both passengers and customers. Our hypothesis is that the utilisation of shared vehicles for parcels and passengers results in improved performance.

Section 2 describes related work, focusing separately on on-demand delivery and passenger services, before moving to co-modal services. The model and experimental design is described in section 3. Results are discussed in section 4, before conclusions and future work are presented in section 5.

2. Related Work

2.1. On-demand passenger services

The classic example of an on-demand passenger service is a taxi service. Traditionally, these vehicles have been either hailed on the street or booked over fixed-line phones.

Some traditional public transportation has an on-demand option. For example, the Telebus service in the outer eastern suburbs of Melbourne has a fixed route and timetable (with some flexibility), but can deviate to passengers' home if they pre-book via telephone (before the bus leaves its first stop) or they request upon boarding at a stop (see Public Transport Victoria (2013) for an example of one of the services).

With the advent of smartphone technology, taxi services can now use apps to take bookings. The most well-known at the moment is Uber (Uber, 2015), which uses an app to connect drivers and passengers directly.

On-demand services can also be shared. The Kutsuplus service in Helsinki, Finland, takes bookings up to 60 minutes in advance (Rissanen, 2014). Passengers ride together on a minibus, but do not necessarily share origins nor destinations.

Similar services such as car- or bike-sharing could also be seen as on-demand.

2.2. Delivery on-demand

Uber Rush (Uber, 2014b) is an environmental-friendly delivery platform that Uber launched in April, 2014. It uses bike messengers and on-foot messengers to bring small items (no more than 30 pounds) from point A to point B in Manhattan and its neighbourhood Brooklyn and Queens. Using the Uber app, the delivery progress can be tracked, and the location of package can be easily traced and shared with the recipient. Uber Rush provides an innovative courier delivery system, which could be used effectively and efficiently in heavy congested urban areas.

Uber Essentials (Uber, 2014a) was another ambitious experiment undertaken in Washington D.C. As a pilot project everyday products were delivered by registered Uber cars. Uber users could make an order (from a list of most popular grocery items) in the app and have it delivered within 10 minutes. Even though this pilot ran for only a few weeks from December 2014 to January 2015, Uber claims that this experiment was highly praised by its customers. This trial allows the operators to identify the value of the service, providing an opportunity to improve with a better business model in the future.

2.3. Co-modality

The use of passenger vehicles for freight transport is becoming more common. Three axes have been recommended for improving mobility in modern transportation networks, firstly, "to improve the sharing of road space between passenger's flows and goods flows", secondly, "to shift passengers and goods flows from private motorised road transport to other urban transport modes" and finally, "to introduce distribution facilities" (Trentini and Malhene,

2010). They also reviewed some experimental solutions such as shared buses, shared subway, shared tramway, car sharing.

This conceptual design was later developed into practical operations (Trentini et al., 2012). A case study performed in the urban area of La Rochelle in France confirmed that it is efficient to use spare capacity on buses for the distribution of goods. Their results are based on the assumption that goods are loaded from one single CDC (City Distribution Centre) located at the beginning of the bus routes, and delivered to the bus stops along the bus route, the bus route and stops are pre-fixed locations, the last mile delivery is then completed by tricycles. In their demonstration, all the bus stops are used as transfer points, which is not very realistic and as a result makes their conclusion questionable. However, it provides some valuable information on the formulation of the mathematic model (objective function is to minimize the number of city freighters used and to minimize the total time travelled) and the algorithm (an Adaptive Large Neighbourhood Search method to solve the routing problem).

Another initiative for passenger and goods ridesharing involved the development of two mathematical models based on the conceptual design of using the same taxi network to handle people and parcels in an integrated way (Li et al., 2014). In this study, MILP (Mixed-integer Linear Programming) formulations are presented for both models, and numerical studies using real taxi trail data are performed respectively.

The first model SARP (Share-a-Ride Problem) assumes passenger requests are random and parcels can be picked up and delivered while passengers are on board. Consequently, the model allows multiple parcels and single passenger in one taxi at the same time, but does not accept two or more passengers to be served concurrently by one taxi. The objective of this SARP model is to maximize the total profit from the taxi company's point of view. This model is solved under several constraints: the pickup and delivery time for people is more critical than for parcels, there is a threshold for acceptance of detour, there is a different rate of cost/benefits for people and parcels, taxis have a certain capacity for parcels, rejections for both people and freight may occur when the network cannot serve all the requests, etc. The SARP model to a great extend reflects the dynamic environment, however, it is significantly increases the difficulty of computation.

The second model FIP (Freight Insertion Problem) is then proposed with reduced complexity. FIP assumes that the routing for handling people requests are given; only the scheduling for inserting parcel requests needs to be decided.

In some real-world cases, vehicles are able to dynamically change between carrying passengers and goods. Uber, a transportation network company which launched in 2010 offering chauffeured rides, frequently runs special offers to deliver certain items within a service area, sometimes themed (e.g., roses on Valentine's Day and icecreams in summer). Uber has moved into offering short-distance cycle and foot courier services in New York City (Uber, 2014b).

In May 2014, a pizza shop in Melbourne's inner southeast began using Uber X drivers (that is, drivers using their own vehicles, not commercial vehicles) to deliver pizzas to customers instead of using designated delivery drivers (White, 2014).

3. Method And Experimental Design

3.1. Simulation framework

The simulation was created using MATSim (MATSim, 2014), a large-scale agent-based transport simulation tool. Outside logistics, MATSim has been used to simulate vehicle and public transit in locations such as Zurich and Singapore (Erath et al., 2012), since its inception in the early 2000s.

MATSim has previously been used for simulation of the VRP problem (Maciejewski and Nagel, 2012) and taxi scheduling (Padgham et al., 2014, Maciejewski and Nagel, 2013). Logistics operations have been incorporated into MATSim (Joubert et al., 2010, Zilske et al., 2012). As MATSim requires all plans at the start of the run, a separate module has been developed for on-demand vehicle routing (dvrp). This has previously been used to simulate different schemes for a dial-a-ride bus (Ronald et al., 2015a). Unlike traditional MATSim models, the dvrp module does not make use of iterations, where individuals make changes to their travel plans depending on how they assessed their performance in the previous iteration; only one iteration is performed.

3.2. Model

The users of the system are called passengers and customers. Passengers need to travel from their home to an activity location, stay at the activity location for a period of time in order to carry out an activity, and then travel back home again (that is, two separate trip requests). Customers place an order for a parcel to be delivered to them, originating at an activity location with their home as the destination. Parcels take a short time to be prepared at the activity location. In the latter case, the request time is the time of the original order plus the time required to prepare the order, and the wait time is measured as the time between the activity location requesting transportation and pickup, not from the time of the original order.

The model contains a number of vehicles, which could be either taxis or vans. For the purposes of this simulation, both types have a capacity of 4 (not including the driver). In the co-modal scenarios, both types of vehicles can carry both passengers and parcels. In the scenarios with no co-modality, taxis can only carry passengers and vans can only carry parcels. Note that, unlike in the Uber X pizza delivery case study described in the related work, vehicles cannot change roles during the simulation. The vehicles can perform two tasks:

• Stay: the vehicle is stationary at a location. This could be waiting for the next task, waiting for the next passenger/parcel, or performing drop-offs or pickups.

• Drive: the vehicle is in motion between two nodes.

Passengers and customers make requests in similar ways, but with some minor differences. Passenger p makes a request at time tp, to travel from home op to an activity location dp. They then make a second request, but this time op is the activity location and dp is home. Customers make a request at time tr for a delivery to be made from the activity location op to their home dp. However, the shop taking the request takes some time to fulfill the order (t0), before issuing a request for transportation at time tp = tr + to.

Our simulation inserts requests as they arrive, making use of an insert heuristic. The objective function aims to minimise travel time of passengers/parcels (Equation 1):

y estimated arrival at d^-tp (1)

direct travel time(Op,dp)

This calculates the sum of the ratio of total travel time using the on-demand vehicle (consisting of wait time and travel time) and the total travel time taking the shortest route to the destination (for example, in a single passenger car). This is a simplification in the customer's case, as the "direct" trip in a car would require driving to the activity location and back.

As in Ronald et al. (2015a), our model has a few limitations. Vehicles cannot stop in the middle of a link; they can only stop at a node. A vehicle performing a drive task between an origin and destination cannot be re-routed until it has reached the destination, nor can an unscheduled stop be made en-route.

3.3. Experimental design and setup

The simulation takes place in a grid world, 15km by 15km. Main streets are located 2.5km apart and have a speed of 60km/h; in between these are minor streets spaced 0.5km apart with a speed of 40km/h. In the centre of the grid is an activity location; this is where deliveries are made from and where passengers would like to travel. The activity location contains three shops.

The simulation takes place from 6pm - 10pm, representing evening activities such as eating out, purchasing take away food or getting home delivery.

Three different delivery scenarios are simulated:

• Shop-specific: each shop has their own van(s); parcels are placed in the shop-specific vehicle. There are also taxis

which carry passengers.

• Collaborative: all shops share a number of vans; taxis carry passengers separately.

• Co-modal: passengers and parcels travel together in vans and taxis.

Input data. In order to populate the simulation with realistic data, survey data from the Victorian Integrated Survey of Households and Activities (VISTA) was used. This is a self-completion household travel survey, undertaken in 2009/10 in Melbourne and regional cities in Victoria (The Urban Transport Institute, 2011). VISTA participants report their trips over a 24 hour period, including the purpose, mode, and length of the trip, and duration of the activity performed at the destination.

The intention behind using VISTA data is not to replicate a particular suburb, but to create realistic input data. Passenger data is extracted from trips beginning at home and with the purpose of "eat and drink" or "socialise", while parcels are extracted from trips beginning at home with a destination of a fast food location or restaurant, and the purpose of "buy something". Only trips between 6pm and 9pm are extracted, and only trips within the same Statistical Local Authority (Note that SLAs were part of the old Australian Standard Geographical Classification and have now been superseded. However, the VISTA data from 2009/10 is still segmented using SLAs) are extracted. These are then averaged to produce demand within an average SLA (the average size of a SLA is 92km2, slightly smaller than our grid environment, however note that SLAs are not square). This is considered the base data; the distribution of activities is shown in Figure 1. Expected demands and durations are shown in Table 1.

On top of this base demand, sensitivity testing was performed. The ratio of parcels between the shops is varied, and the ratio of parcels and passengers is also varied.

Parameters. The model makes use of a number of parameters, shown in Table 1. It is acknowledged that the pickup and dropoff for parcels is longer than for passengers, as the driver usually needs to leave the vehicle to pickup/deliver the parcel, or needs to wait in the car for the shop owner to bring the parcel out or for the customer to come to the vehicle. Within the time period distributions, the precise value is uniformly drawn. Within the distance distributions, a link within the specified distance range from the shopping strip is uniformly chosen.

Table 1. Values of parameters used in the simulation

Parameter description Value

Pickup and dropoff dwell times for passengers 60 secs

Pickup and dropoff dwell times for parcels 120 secs

Activity start time for passengers 6-7pm: 73.5% 7-8pm: 19.3% 8-9pm: 7.2%

Activity start time for parcels 6-7pm: 66.8% 7-8pm: 26.1% 8-9pm: 7.1%

Duration of activity for passengers 0-1 hours: 23.7% 1-2 hours: 30.2% 2-3 hours: 46.1%

Time needed to prepare parcel 0-10 minutes: 45.2% 10-20 minutes: 26.4% 20-30 minutes: 28.4%

Distance distribution of passengers to shops 0-1km: 25.2% 1-2km: 47.1% 2-5km: 20.3% 5-10km: 5.8% 10km+: 1.6%

Distance distribution of parcels to shops 0-1km: 50.9% 1-2km: 23.0%

2-5km: 20.0% 5-10km: 6.1% 10km+: 0%

Outputs. Outputs from the simulation include VKTs, trips/kilometre, and the average occupancy of each vehicle from the operational viewpoint; this can also be measured separately for van and taxis. From the passenger viewpoint, wait, excess travel and total travel times are measured, separately for passengers and parcels.

4. Results And Discussion

The model described in the previous section was used to perform three experiments. Results are averaged over 10 runs, with demand generated randomly for each run.

4.1. Base demand

In the first experiment, we explore the effects of each delivery scenario, using real-world inputs. It is expected that the co-modal scenario, in which parcels and passengers travel together, is more efficient than the other two scenarios.

In looking at the operational outputs, it can be seen that the co-modal scenario results in a slightly lower trips per kilometre outcome than the other two scenarios. This is due to the increased success rate in the mixed scenario: more passengers and parcels can be handled. The average occupancy of vans and taxis also changes markedly between scenarios: the shop-specific and collaborative scenarios show an imbalanced occupancy between vehicle types, whereas the co-modal scenario provides a more even balance (see Figure 1).

Fig. 1. Average occupancy for different vehicle types in the three scenarios (base demand)

From the passenger and customer perspective, the mixed scenario provides a much better service. Wait time drops from over 8 minutes in the shop-specific and all-shops scenarios to just over 6.5 minutes in the mixed scenario. The amount of excess time spent travelling also drops from around 1.5 times (shop-specific and collaborative) to less than 1.2 times (co-modal) the direct travel time.

These results show that the mixed scenario is currently more beneficial for passengers and customers, who gain from being able to ride in formerly underutilised parcel-designated vehicles.

4.2. Altering shop demands

The second experiment explores the effects of changing the ratio of deliveries from each shop. If shops agree to participate in a co-modal (or collaborative) scheme, is there an advantage for shops with smaller or larger demands? In this experiment, the demand for the three shops was changed from being all even (that is, 1/3 each, known as case 1) to the third shop having double the demand of the first two shops (that is, 1/4 , 1/4, 1/2 respectively; case 2) and also to the third shop having half the demand of the first two shops (that is, 2/5, 2/5, 1/5 respectively; case 3). In this case, we expect to see the most change occurring in the shop-specific scenario in those three cases, and between the shop-specific scenario and the other two scenarios for each case.

Firstly, the shop-specific scenario shows a slight increase in wait times for parcels when moving from case 1 to case 2 and 3. The excess time also increases significantly between case 1 and case 2, but only slightly between case 1 and case 3. Figure 2 shows the total travel time for parcels. This shows that an uneven distribution of shop demand can affect the delivery of parcels from all shops.

Fig. 2. Total travel time for parcels in each case

Secondly, looking at cases 2 and 3, the variation in van occupancy in the collaborative and the co-modal scenarios are smaller than in the shop-specific scenario. This shows that the collaborative scenarios produce a more stable service when demand is uneven.

4.3. Varying ratio of passengers and parcels

The final experiment builds on the previous experiment and explores the effects of changing the ratio of passengers and customers. In this experiment, the demand was changed to 400 users in total, and the number of vehicles to 12 taxis and 6 vans. The ratio of customers to total users was set to 28.8%, 33%, 40%, 50%, and 60%.

From the operators' point of view, there is not a linear pattern to the VKTs in the shop-specific and collaborative scenarios. As parcels increase in number, this results in a reduction of trips (as passengers take two trips) but an increase in dwell time. The VKTs reach a peak at a ratio around 50%, before decreasing again.

Both passengers and parcels benefit from a co-modal approach. Ignoring dwell time, the total travel time (i.e., wait time plus travel time) is consistently around 25 minutes for parcels and 17 minutes for passengers for all ratios in the co-modal scenario. Comparing this to the collaborative scenario shows a clear increase in the parcel travel time as the parcels increase (from 25 minutes to 107 minutes) and a decrease in the passenger travel times (from 22 to 12 minutes). Note that it is only at the final ratio (60% parcels) where the passengers do receive better service from the collaborative scenario compared to the co-modal scenario.

Looking at the efficiency of the system, the co-modal scenario is reasonably constant across all values, however the shop-specific and collaborative scenarios show a decrease as the proportion of parcels increases. This again confirms the balancing nature of the co-modal scenario, which is more resilient to changing patterns of demand.

5. Conclusion

The efficient movement of goods and passengers is an important issue in today's society. This paper has used simulation to explore an on-demand shared passenger/parcel transport service, using travel demand data from Melbourne, Australia as input. Three scenarios were explored: each shop has their own delivery vehicles, all shops share delivery vehicles, and a co-modal scheme where passengers and parcels can share vehicles.

The simulation shows that a co-modal scheme does provide benefits, however these depend on the demand pattern of both passengers and parcels. In the idealised scenario we presented, it could be seen that passengers received an advantage from the co-modality, as the vans were underutilised with parcel deliveries. In the experiments with different ratios of passengers and parcels, it could be seen that a co-modal scheme provided benefits to all.

This model is a proof-of-concept only. A key limitation of the model is the simplistic optimisation procedure. Firstly, the optimisation can be improved; secondly, there is scope to develop specific algorithms for the co-modality problem. Only certain variables were explored in this preliminary paper: future work could consider varying the number of vehicles amongst others. Costing could be produced, however this requires estimation of running costs, specification of vehicle types, and development of pricing schemes for both passengers and parcels. Finally, preferences and attitudes need to be included: it could be that passengers do not wish to share their ride with parcels and/or other passengers. Given some of the results in the third experiment where customers waited almost 2 hours for their orders, this could lead to shops not using the service or customers going to pick up their deliveries.

As a preliminary study, this shows that on-demand co-modality has advantages. It is more resilient to uneven or unexpected demands, and provides more options for travel. This could be seen as a solution to making more effective use of vehicles on the road network.


The first author was supported by a grant from the Australian Research Council (LP120200130). The second and third authors were supported by the Volvo Research and Educational Foundations' Center of Excellence for Sustainable Urban Freight Systems (CoE-SUFS).


Commission of the European Community (2006). Keep Europe moving -- Sustainable mobility for our continent.

Erath, A., Fourie, P., Eggermond, M. V., Ordonez, S., Chakirov, A. and Axhausen, K. (2012). Large-scale agent-based transport demand model for Singapore. ETH Zurich.

Joubert, J., Fourie, P. and Axhausen, K. (2010). Large-Scale Agent-Based Combined Traffic Simulation of Private Cars and Commercial Vehicles.

Transportation Research Record: Journal of the Transportation Research Board, 2168, 24-32. Li, B., Krushinsky, D., Reijers, H. and Van Woensel, T. (2014). The Share-a-Ride Problem: People and parcels sharing taxis. European Journal oof Operational Research, 238, 31-40.

Maciejewski, M. and Nagel, K. (2012). Towards multi-agent simulation of the dynamic vehicle routing problem in MATSim. In: Wyrzykowski, R., Dongarra, J., Karczewski, K. and Wasniewski, J. (eds.) Parallel Processing and Applied Mathematics (Lecture Notes in Computer Science 7204). Berlin, Germany: Springer-Verlag Berlin Heidelberg. Maciejewski, M. and Nagel, K. (2013). Simulation and optimization of taxi services in MATSim. Verkehrssystemplanung und Verkehrstelematik, TU Berlin.

Matsim. (2014). Multi-Agent Transportation Simulation Toolkit [Online]. [Accessed 16 July 2014].

Padgham, L., Nagel, K., Singh, D. and Chen, Q. (2014). Integrating BDI Agents into a MATSim Simulation. ECAI2014, 681-686.

Public Transport Victoria. (2013). Mooroolbark Station - Chirnside Park (Telebus Area 2) [Online]. Available:

around/maps/telebuses/view/1183 [Accessed 28 February 2015]. Rissanen, K. (2014). -- a Novel, Automated Real-time Demand Responsive Public Transport Service. Proceedings oof the 10th ITS

European Congress. Helsinki, Finland. Ronald, N., Thompson, R. G. and Winter, S. (2015a). A Comparison Of Constrained And Ad-hoc Demand-responsive Transportation Systems.

Transportation Research Board 2015 Annual Meeting, Washington, DC, USA. Ronald, N., Thompson, R. G. and Winter, S. (2015b). Simulating demand-responsive transportation: a review of agent-based approaches. Transport Reviews (online first).

The Urban Transport Institute (2011). Victorian Integrated Survey of Travel & Activity 2009-10: Survey Procedures and Documentation. Thompson, R. G. and Taniguchi, E. (2014). Future Directions. City Logistics - Mapping the Future. CRC Press.

Trentini, A. and Malhene, N. (2010). Toward a shared urban transport system ensuring passengers & goods cohabitation. Tema Journal of Land

Use, Mobility and Environment, 3, 37-44. Trentini, A., Masson, R., Lehuede, F., Malhene, N., Peton, O. and Tlahig, H. (2012). A shared "passengers & goods" city logistics system. 4th

International Conference on Information Systems, Logistics and Supply Chain, Quebec, Canada. Uber. (2014a). The everyday items you need in 10 minutes or less [Online]. Available: [Accessed 28 April 2015]. Uber. (2014b). A reliable ride for your deliveries [Online]. Available: [Accessed 21 Jan 2015]. Uber. (2015). Uber [Online]. Available: [Accessed 28 April 2015].

White, K. (2014). Is that a pizza in your Uber? [Online]. Available:

uber/2014050612230.html [Accessed 21 Jan 2015]. Zilske, M., Schroder, S., Nagel, K. and Liedtke, G. (2012). Adding freight traffic to matsim. VSP Working Papers, . TU Berlin, Transport Systems Planning and Transport Telematics.