Integrating Shared Autonomous Vehicles with Public Transit

This is a school project completed in the Spring 2019 semester of Professor Joseph Chow’s ‘Travel Behavioral Informatics’ course, alongside my classmate Shams Sahar. The subtitle of this project is System Design — Serving First and Last-Mile Trips of Public Transit

Introduction

In the U.S., public transit ridership has been seeing historic lows due to several factors, with declining service quality, rising cost and availability of mobility on-demand options being three of them. In light of this, unreliable, infrequent and poorly used transit networks are impractical for today’s sharing options — such as UberPool — due to high cost and low demand. This is particularly true in low density areas, let alone for transit-rich cities like New York City, which itself saw a 6% reduction in bus ridership in the year 2017.

In recent years, however, the potential of autonomous vehicle (AV) technology has been touted as a solution to many problems. One of the application areas has been in the use of public transit. In general, using autonomous or shared autonomous vehicles (SAV) is more cost effective than sole automobile ownership, and more time effective than relying exclusively on public transit. In order to mitigate some of the negative effects of adding more vehicles, autonomous or otherwise, on the road, would be to integrate these new modes with existing modes, like public transit. This would result in a mode, termed AV+PT, that would bring benefit from at least two directions. When looking at low density areas that are currently poorly served by transportation, AV+PT could encourage greater use of these public services, while making people less dependent on automobile ownership. Combining the convenience of on-demand mobility with the sustainability and resource efficiency of public transit will have, hopefully, a net increase of the quality of life for those using the AV+PT infrastructure.

In this case, we chose to focus our study on the impact of introducing shared autonomous vehicles for first and last mile trips of Long Island Rail Road (LIRR) users, by surveying the revealed and stated preference respectively as it relates to their current mode choice and their prospective mode choice were shared autonomous vehicles introduced to them as an option, respectively. We chose LIRR users because we understood that, unlike New York City residents, those who use LIRR are more likely to live in suburban and rural areas with less access to public transit, making this population an ideal use case for our system design. The ultimate result of our project is a projection of mode share under the introduction of a new mode, and under different conditions.

Literature Review

Increased use of public transportation is associated with the reduction of traffic congestion, fuel consumption, and air pollutants in urban centers. Quality of service and accessibility to public transport are the major factors affecting the demand. However, the access time and distance to transit stations remain a hassle to the commuters. In this case, providing a convenient means to provide first and last-mile to public transit services can significantly increase the use of the transit mode.

According to numerous studies, autonomous vehicles (AVs) are expected to increase vehicles kilometers travelled (VKT), energy use, urban sprawl and decrease transit mode share (Childress et al., 2015). Thus, the deployment of AVs is going to impact the performance of public transport negatively. However, shared autonomous vehicles (SAVs) can meet the demand of travelers with lower cost and greater convenience relative to private car trips or on-demand mobility services used now (AAA, 2019). Deployment of SAVs serving first and last-mile of public transit is considered to be an attractive option that increases ridesharing and supports public transit.

Different studies predict that SAVs can provide services at a significantly lower cost compared to today’s mobility options. The forecasts predict SAVs would cost up to 44 cents per mile (27.5 cents per km) and as low as 8 cents per mile (5 cents per km) for a private vehicle trip and a pooled SAV ride in a two-seat micro-car, respectively (Johnson, 2015). It is significantly lower than Uber, with $3 to $3.50 per mile for private UberX car or $1 to $1.50 per mile for an UberPool vehicle (Naughton, 2015). SAVs is also more cost-effective than riding private vehicles. The composite average cost operating a private automobile is 58 cents per mile for a vehicle traveling 15,000 miles per year (AAA, 2019). Besides, SAVs would also reduce demand for large parking lots as one SAV is predicted to replace 8 to 9 conventional vehicles (Johnson, 2015; Zhang et al., 2015). SAVs can reduce auto dependency and help cities reach a ‘car-free’ vision (Tight et al. 2016). Meanwhile, there are some social benefits associated with the deployment of SAVs, such as reduced crash incidents, increased mobility and accessibility for disabled individuals, low income, elderly and young population.

Public transit services in low-density areas are infrequent, slow, and unreliable due to low demand and less accessibility. It has resulted in heavy dependence on private automobiles. Providing on-demand mobility services can significantly impact travel behavior in low-density areas (Newman and Ken-worthy, 2006). Thus, integration of SAVs with on-demand mobility application with public transit networks can provide better access to public transport and reduce private car ownership. It results in better public transit services due to increased demand along transit routes. It enables transit operators to extend public transit routes to outside urban centers where SAVs are operating as feeders along the routes.

According to a Zmud et al. survey in 2016, some participants perceived the experience of automated driving similar to riding public transit, but and relatively more convenient and desirable. Thus, combining SAVs and public transit will reach user expectation and will be an attractive option. However, the service quality and travel time offered by this new mode must be competitive with private automobiles. The pricing policy also plays a critical role in this aspect. Smart integration of these modes will result in a more competitive travel option against the private automobile.

Proposed System Design

In the most basic formulation of our proposed system, shared autonomous vehicles are serving the beginning and end of any user’s trip that involves transit. This is otherwise termed as the first and last mile of a trip. This way, SAVs are integrated with existing transit infrastructure, with the intended outcome being an increase in transit ridership. This would take the form of having access to shared AVs with an on-demand mobility application. To serve first and last mile trips using public transit, SAV’s could, for example, be stationed and parking lots near public transit stops and stations.

From this design there are a number of anticipated benefits or changes. One that was already mentioned is the increase in transit ridership, however, the benefit from this scenario could also be seen as increasing transportation access in low density areas that are currently not well served by public transit. Another benefit is the improvement of mobility equity among different demographics that are also currently not well served by public transit, e.g. elderly individuals. Decreased traffic congestion and carbon emission in urban areas by increasing use and greater access of public transport can be seen as another benefit. Finally, having transportation options that decrease the use of single occupancy automobile trips will contribute to an affordable, equitable, sustainable and efficient urban mobility system in the future.

In the current, or benchmark system, public transit is the core or sole element of the trip, even factoring in the time and effort needed to get to and from a public transit stop. The benchmark use case then, looks like the following:

This is fairly standard and familiar to anyone who is a regular user of public transit. There is a certain time it takes to get to a destination from your origin, and once you get to the stop, there is a certain waiting time. Ultimately, the transportation arrives, and you pay your fare, getting off at a stop at or near your destination, where you travel the rest of the way there, usually by foot.

For our proposed design, as mentioned before, we are turning the access time to public transit to a time where a user is driven and connected to it through an SAV. The following are diagrams that illustrate the design for the proposed system.

First Mile (Use Case)

As is illustrated in the diagram above, users can access a public transit stop by connecting with a shared autonomous vehicle. This will allow people who are otherwise far away from public transit by foot to access it more readily. Once connected to transit, the trip has the same structure that it would have in the benchmark case.

Last Mile (Use Case)

The last mile use case is effectively a mirrored process of the first mile. In this case, the time it takes to get from public transit to your destination is changed due to the introduction of SAVs. What would previously have taken a certain amount of time by foot is shortened to access to a vehicle.

The following are also activity diagrams for our proposed system, for first and last mile trips:

First Mile (Activity Diagram)

In the first mile, the user is given a choice to use an SAV or another mode to get to their PT station. If they use an SAV, then they are put through a process where they are connected with one through a location and route optimization algorithm, similar to what is currently found in applications like Uber and Lyft. At this stage, SAV’s will function similarly to what can be seen with UberPool, in this case specifically connecting people to public transit.

Once that segment of the entire trip is completed, users are then dropped off at public transit stations, where they will have access to the bus, train, or other mode that they use.

Last Mile (Activity Diagram)

In the last mile of the trip, the user will still have the choice between using other modes to get to their destination, and using public transit mixed with SAV’s. In this scenario, if they do not choose other modes, they will choose public transit, where the activity flow is similar to what we experience already as transit users (wait, get on, pay fare, ride, exit). A parallel process also takes place during the transit section of this activity, where the SAV is found and the route starts to be optimized. Once the transit trip is over, the last mile, or SAV portion of this trip, can commence. This results in payment and passengers being dropped off.

Data Collection

In order to collect data, we decided to survey LIRR users in-person at two locations. The two in-person locations were LIRR stations within the Greater New York City Area, located at Atlantic Terminal in Brooklyn and Pennsylvania Station in Manhattan. We carried out our survey in part at two physical locations where people who ride LIRR would not be in a hurry. Specifically, we surveyed people in the waiting areas at the respective LIRR terminals. Another reason for surveying people in these settings is that we would ideally get a large portion of respondents who do not live in transit-rich environments like New York City, but live in settings where there are less robust transit options. For example, in places where the only available public transit stop is an LIRR stop, not a bus or subway stop.

The survey itself was split up roughly into two parts. The first part, which corresponds to revealed preference and is split into two sections, asks users about their current primary mode choice (between public transit, private car and taxi/Uber/Lyft), their primary purpose of travel, as well as socioeconomic information such as their household income, the kind of area that they live in (urban, suburban or rural), and whether or not they are employed, in the first section. The later section asks respondents to report how much a typical trip for them costs, and how long it takes. It also asks them for their perceived costs and trip times of other modes that they do not primarily use. The second part, which corresponds to the stated preference part of our survey, asks respondents to state their willingness to use the AV+PT service under four conditions, listed below:

1. Using AV+PT if waiting time and cost associated with Shared Autonomous Vehicles are 5 min, and $0.08 per mile respectively
2. Using AV+PT if waiting time and cost associated with Shared Autonomous Vehicles are 5 min, and $0.44 per mile respectively
3. Using AV+PT if waiting time and cost associated with Shared Autonomous Vehicles are 10 min, and $0.08 per mile respectively
4. Using AV+PT if waiting time and cost associated with Shared Autonomous Vehicles are 10 min, and $0.44 per mile respectively

For the last part, we chose to follow the stated preference model for the survey in order to capture respondent data that would more closely align with choices that they would make in a hypothetical situation concerning SAVs, or a mode that currently does not exist. Looking forward to our mode share elicitation, we structured many of our questions as a Yes/No because the dichotomous choice formation made it easier to collect the data. The summary of data collection is presented in the table at the left.

Data Analysis

The collected data are used to predict demand for the new mode of AV+PT. The purpose is to predict mode shares after deployment of a new transit mode by the integration of shared autonomous vehicles with public transit. Almost 140 survey responses were collected. However, some of the responses were incomplete and were omitted.

A nested logit model is used to predict mode shares. There are two levels in our nested logit model, including Private Cars, Taxi/Uber/Lyft, and Transit in the upper nest with Public Transit and AV+PT in the lower nest. The Public Transit branch combines subway, bus and LIRR riders. The difference between AV+PT and Public transit mode is that the first one is accessed by SAVs while later with walking. The sum of mode shares of AV+PT and Public transit represents the share of transit mode.

The two major attributes of utility functions are travel time and travel cost. The utility functions and their attributes are presented below.

Where;

i stands for the modes (Car, Taxi, Public Transit and AV+PT)
Xm is the vector of attributes of the mode (travel time and travel cost)
Bm is the vector of corresponding Coefficients

The reader is encouraged to refer to the R-Script uploaded with the paper for detailed estimation of the utility functions and mode shares. The summary of mode share under different conditions is presented below.

Revealed Preference

An MNL model is used to estimate the mode share of the travelers at the current state. As noted, public transit is the primary mode of transport for most travelers with 65% mode share. Since AV+PT is not into the market yet, its share is zero at the revealed preference state.

The proposed system of AV+PT is considered to operate under a different number of conditions. Operation performance and pricing policy of autonomous vehicles significantly affect demand for AV-PT system. Waiting time for SAV represents system performance, while cost per mile represents system pricing policy. As mentioned earlier, the forecasts predict SAVs would cost up to 44 cents per mile and as low as 8 cents per mile for a private vehicle trip and a pooled SAV ride in a two-seat micro-car, respectively (Johnson, 2015). Therefore, we have combined two system performance levels (5 vs. 10 min waiting time) and two pricing policy ($0.08 and $0.44 per mile) to predict the demand under different conditions. Condition 1 represents the best scenario with 5 min wait time and $0.08 per mile cost of SAVs while Condition 4 represents the worst scenario with 10 min wait time and 44 cents per mile cost of SAVs. The summary of mode share under condition 1 and condition 4 is presented below.

Condition 1 (SAVs’ waiting time and cost are 5 min and 0.08$ per mile respectively)

The deployment of AV+PT under condition 1 is going to reduce the share of Car and increase the share of transit significantly. Meanwhile, most travelers are willing to use shared autonomous vehicles to connect with public transport stations compared to walking by to the station.

Condition 4 (SAVs’ waiting time and cost are 10 min and 0.44$ per mile respectively)

However, the demand for AV+PT system is at the lowest while operating under condition 4. The car share almost remains the same as prior to deployment of AV+PT system with the majority of public transit users preferring to walk to the station or stops rather use SAVs.

The share of AV+PT system under four different conditions has been presented in the figure below. It suggests that demand for the proposed system is the highest at the first two conditions where the SAVs waiting time is 5 min. There is a sudden reduction in demand when the waiting time increases to 10 min in condition 3 and 4. We can conclude that travelers are more sensitive to time relative to cost, and quality of service plays a critical role in maintaining demand for the AV+PT mode. It is worth mentioning that the majority of derived demand for AV+PT comes from public transit users who are willing to use shared autonomous vehicles for the first mile or last-mile instead of walking. This segment of the population is more sensitive to time relative to the cost of AV+PT.

Transit share under different AV+PT operating conditions is presented below. The sum of public transit and AV+PT mode shares represents the transit share. As noted, under condition 1 and 2, AV+PT majorly contributes to the share of transit. As the SAVs wait time increase to 10 minutes in condition 3 and 4, the AV+PT does not remain an attractive option for travelers, and they would prefer not to use SAVs for the first or last-mile of their trip. Of all the respondents, 43% cited some reason for not using AV+PT related to time, either in waiting or travel time. Additionally, 31% cited cost as the reason for not using AV+PT. Finally, only 15% stated safety related concerns for not using AV+PT.

The share of all modes including private car, taxi, and transit (public transport and AV+PT system) at different AV+PT operation conditions has been presented in the figure below. As mentioned earlier, the primary objective of AV+PT system is to increase the share of transit use and reduce automobile dependency. The bar chart presented below can provide us some insight on how we can reach this goal under different AV+PT operating conditions. As noted, at condition 1 and 3, the share of transit is at the highest with the lowest shares of private cars. The cost of SAVs is the same during both conditions with $0.08 per mile. Despite the change in waiting time from 5 min in condition 1, to 10 min in condition 3, the majority of travelers prefer to use transit than driving their private cars. It emphasizes that the cost of AV+PT is a critical factor for travelers to prefer transit over car trips. It must be noted that in condition 1, AV+PT represents the majority of transit share, while public transit represents a significant portion of transit share in condition 3 and 4.

There are certainly some significant limitations to this study that hinders the reliability of the predicted mode shares. The coefficients estimated for nested logit models were relatively insignificant. The likely reasons are the insufficient number of survey responses as well as inconsistency in the survey responses. As it can be noted in the survey questionnaire, the inconsistency of the responses is basically due to the perception of travelers regarding travel cost and time, which are usually very inaccurate. The reader is encouraged to refer to the R-Script uploaded with the paper for detailed estimation of the utility functions and mode shares. These limitations make the predicted mode shares unreliable. Despite this fact, we have made efforts to develop a framework for studying the integration of autonomous vehicles with public transit. Allocating sufficient time and resources for surveying and data collection can address these challenges.

Conclusion

There are two major aspects of the results of this study. When it comes to the ultimate adoption of AV+PT as a mode, SAV pricing and system performance characteristics are the most influential factors. System performance characteristics include waiting time or access time to AV+PT. In order to increase share of AV+PT, we need to have a high-quality and reliable system performance alongside affordability.

The second major aspect is the defined objective in operating an AV+PT system. If the operator wants to maximize the share of AV+PT, they should offer services that are aligned with what is seen in Condition 1 or Condition 2. Otherwise, the demand for AV+PT is going to be very low while operating in Condition 3 or Condition 4. Conditions 1 and 2 represent the lowest waiting time for SAVs (5 minutes) while Conditions 3 and 4 represent the longest waiting time (10 minutes). What this suggests is that waiting time plays a critical role in predicting demand for the proposed system design of AV+PT for this segment of population. Therefore, we propose that similar studies should be done in different localities in order to understand their travel behavior and provide them services that maximizes profit or revenue for the operators of an AV+PT system.

According to our study, we see that AV+PT is going to significantly reduce the mode share of private car, while increasing the use of public transit. This would have great effects in the sustainability and equity of the transportation network under study, in terms of making users less reliant on private cars, reduce traffic congestion, increase the use of more efficient forms of transportation like public transit, and provide transportation access to those populations which are currently underserved.

References

Ohnemus, M., & Perl, A. (2016). Shared autonomous vehicles: Catalyst of new mobility for the last mile?. Built Environment, 42(4), 589–602.

Childress, S., Nichols, B., Charlton, B. and Coe, S. (2015) Using an Activity-Based Model to Explore Possible Impacts of Automated Vehicles. Paper presented at the Transportation Research Board 94th Annual Meeting. Available at: http://trid.trb.org/view.aspx?id=1339044.

AAA (2019) Your Driving Costs: How much are you really paying to drive? Heathrow, FL: American Automobile Association. Available at: http://exchange.aaa.com/automobiles-travel/ automobiles/driving-costs/.

Johnson, B. (2015) Disruptive Mobility: AV Deploy- ment Risks and Possibilities. New York: Barclays Capital Inc. Available at: http://orfe.princeton. edu/~alaink/SmartDrivingCars/PDFs/Brian_ Johnson_DisruptiveMobility.072015.pdf.

Tight M., Rajé F. and Timms P. (2016) Car-free urban areas — a radical solution to the last mile problem or a step too far? Built Environment, this issue.

Newman, P. and Kenworthy, J. (2006) Urban design to reduce automobile dependence. Opolis, 2(1), pp. 35–52.

Zmud, J., Sener, I. and Wagner, J. (2016) Self- Driving Vehicles: Determinants of Adoption and Conditions of Usage. Paper presented at the Transportation Research Board 95th Annual Meeting, Washington DC.

Ben-Akiva, M. E., Lerman, S. R., & Lerman, S. R. (1985). Discrete choice analysis: theory and application to travel demand (Vol. 9). MIT press.