Towards a computational transportation science

This workshop report sets out to define Computational Transportation Science as the science behind intelligent transportation systems. In particular it develops a first research agenda for this science, illustrating its unique challenges and putting them to public debate.


AN EMERGING DISCIPLINE
In the near future, vehicles, travelers, and the infrastructure will collectively have billions of sensors that can communicate with each other.This environment will enable numerous novel applications and order of magnitude improvements in the performance of existing applications.Future transportation systems, due to their distributed/mobile nature, can become the ultimate testbed for a ubiquitous, i.e., embedded, highly-distributed, and sensorladen computing environment of unprecedented scale.
The field is currently subsumed as intelligent transportation systems, or ITS.However, the question arises whether behind intelligent transportation systems we also need a science (for a similar discussion in another discipline see [2]).The witnessed paradigm shifts in technical possibilities-for example, from centralized to distributed or decentralized computing, from carefully managed authoritative data to massive real-time data streams of unknown quality-may require new scientific foundations.More and more aspects of transportation science require deep computational methods to deal with the complexity of dynamic environments.We argue that a better interface between transportation science and computer science, including information science.The communication and exchange between scientific communities would improve, but there are also shared common themes and long-term research questions.Around these core research themes we define a new discipline: computational transportation science.
Computational transportation science (CTS) is studying transport systems where people interact with systems (e.g., interfaces of driver assistance, or integrated transport information), where systems monitor and interpret traffic (e.g., mining for activity patterns, or crowd-sourcing to inform about events), or where systems manage the traffic (e.g., control of traffic flow at traffic lights, or toll management).CTS inherits from computer science the aspects of distributed and decentralized computing and spatiotemporal information processing, and from transportation science the aspects of transportation control and management.The discipline goes beyond vehicular technology, and addresses pedestrian systems on handheld devices, non-real-time issues such as data mining, as well as data management issues above the networking layer.CTS studies how to improve the safety, mobility, efficiency, and sustainability of the transport system by taking advantage of information technologies and ubiquitous computing.In particular it needs scholars and practitioners that maintain and push forward an agenda and a body of knowledge that is rooted deeply in both established disciplines.We are also the first to admit that drawing the lines between the established and the emerging discipline is to some extent arbitrary.The purpose of claiming an emerging discipline is rather to focus work that otherwise is spread all over, and to foster research in this area by recognition.By no means is it intended to become exclusive or divisive.In all the examples above, and in the research agenda below, research is already underway.
This way CTS becomes the science behind ITS.Where academic communities look after ITS, such as in the IEEE ITS Society (founded 2005), these communities actually interpret the "S" as science, not systems, otherwise they would not pass scientific peer review.However, the scientific discipline behind ITS cannot be named intelligent transportation science, in analogy to ITS.First of all there is nothing such as an intelligent science, and then there is also an established discourse in artificial intelligence whether machines can be intelligent, which is unnecessary to open up here.So computational transportation science seems to say it all.With the first workshop appeared a preliminary publication exploring a research agenda in this area [1].Then a Dagstuhl Seminar on Computational Transportation Science was held in 21-26 March 2010 to characterize the discipline and identify its research agenda.The seminar was attended by 25 invited researchers from USA, Australia, Germany, Belgium, and Switzerland, with nationalities also from China, India, Greece and former Yugoslavia.This report presents the highlights of this Dagstuhl Seminar.Major steps at the seminar have been:

HISTORY
• Collaborative definition of CTS, vision of CTS, and core research agenda for CTS • Set up of a Wikipedia entry for the definition and vision • Set up of a webpage as a bulletin board for the growing community • Engagement with funding bodies promoting CTS as a discipline (outreach) • Establishing collaboration by developing some larger joint research project proposals • Publishing the (first) core research agenda via this report

CORE RESEARCH AGENDA OF CTS
A discipline is, among other properties, characterized by a common core research agenda.Computational Transportation Science must have some underlying long-term fundamental research problems to distinguish it from its application area, ITS.ITS, in comparison, has defined several research and development agendas that are typically short-term, such as the US DOTs strategic ITS plan [7].
The following research agenda is the outcome of the discussion of the Dagstuhl Seminar on Computational Transportation Science.As such it is preliminary and biased by the composition of the participants.Nevertheless it demonstrates vision and need for this discipline.This preliminary agenda is structured into five sections: applications, knowledge discovery, decentralized computing, social computing, and societal issues.

Applications
As the name Computational Transportation Science indicates, an important aspect are computational and algorithmic aspects in CTS.The challenges lie in the diversity of sensors and thus data gathered in different spatial, temporal and thematic resolution.The high volume demands for adequate information reduction for processing.One way to solve it is to exploit the principle of locality, i.e. the fact that information is mainly relevant locally and thus can also be processed locally and need not be communicated and processed on a central server.This leads to concepts of decentralized and distributed processing.
The applications described in the following do not only rely on the fact that travelers are provided with information; on the contrary, as travelers are equipped with sensors capable of acquiring information of the local environment, they also act as data providers.This leads to a highly dynamic map of the environment which can be exploited in numerous ways.On the one hand, it provides real-time data and thus can be used for dynamic traffic assignment; on the other hand, it also enhances the perception range of individuals and allows them to "look around the corner", or to "look through the cars in front of them".An additional important benefit is the possibility to augment the environmental information with virtual information about the infrastructure.In this way, virtual traffic lights or virtual lane assignments can be realized to allow for a flexible traffic management.
Not only the data can be shared-but also the transportation resources can be shared.This has been the case ever since for the road network and for public transportation -however it can also be envisaged for sharing vehicles like private cars.
The applications are driven by different factors: • Ever increasing traffic demand leading to congestions with dramatic effects on public safety and on the environment, but also on the economy due to time spent in traffic jams • Real infrastructure is expensive and laborious to maintain; furthermore, it is ageing and has to be replaced by modern, new concepts and systems • Cars and travelers are more and more equipped with sensors which can-among others-also capture information about themselves and about the local environment.This rich data source can be exploited.
In the following, some future applications are described: 1. Shared transportation resources: if all traffic modes are included (also private traffic), a better exploitation of the resources is achieved, with several benefits both for the users (reduced prices), the infrastructure (less congestion) and as a consequence also the environment (less pollution).
2. Collaborative travelling, for example by platooning, i.e. the virtual coupling of vehicles to form larger units like virtual trains.These structures can get priorities e.g. when crossing junctions.Within a platoon, autonomous driving is possible.Also by intersection negotiation and intelligent traffic lights, i.e., a more adaptive giving right of way depending on the current traffic situation instead of fixed schedules.
3. Infrastructure is replaced by virtual infrastructure: in this way, the real infrastructure, which has several disadvantages like ageing and expensive maintenance, can be replaced.Examples are virtual lanes to compensate different traffic volume during a day/a week; virtual traffic lights, virtual signs; it is also relevant for highly temporal and ad-hoc warnings like construction sites or aquaplaning or slippery roads.
4. Driver assistance: drivers can be warned of risks in their local environment or when risking to leave their lane.Furthermore, their visibility range can be expanded by providing highly up-to date information from areas that are currently invisible.
5. Evacuation planning: highly temporal information is provided to support and calibrate simulations.
6. Autonomous driving: as a long-term goal, highly dynamic maps of the environment have the potential to support autonomous driving.
7. Dynamic road pricing: the knowledge about the current usage of roads can be used to manage traffic, e.g. by reducing prices for collaboratively used cars or platoons.
8. Smart grid, electric cars: sharing resources opens the way to extend the flexibility of using and sharing electric cars, e.g. by dynamic planning of the electric grid resources, and of routes by considering charging facilities.9. Road and traffic planning can be greatly enhanced by precise, high resolution travel information, which leads to adaptive traffic systems.For example, the road weather-up-tothe-minute visibility, precipitation, and pavement condition information-can be provided at high spatial resolution.
In general, the major benefits and expected properties are robustness (due to high redundancy of information), resilience (ability to recover after failure), reliability and timeliness, which is relevant both for offline and online applications describe above.

Knowledge Discovery, Filtering, and Visualization
In order to be efficient, safe and environmentally friendly a traveler must be cognizant of their inherently dynamic surroundings both through their own sensing systems and by communicating with other travelers and systems.At present travelers gain most of their situational awareness from their innate sensors (eyes, ears, etc.) perhaps augmented by delayed reports from the radio or the web.
Thus it is important to discover in a timely fashion additional information that can augment the innate sensors.Consider for example the query: what will be the expected traffic conditions at 8pm on I298 at Ontario?This query can be answered by a server that stores historical information; but additional information may be available on the web, e.g., the weather and special events such as a ball game that starts at the time.In this case it is not even clear what data and web sites are relevant to the query.
It will not be long, however, before the traveler will be inundated with real-time information coming from all distance scales over soon-to-be ubiqui-tous always-on wireless networks.Prioritization of messages will be critical.Hence, knowledge discovery, filtering and visualization form research challenges to devise mechanisms that makes sense of the huge amounts of heterogeneous and distributed data in particular decision making contexts.Some challenges are: 1.For the car driver the vehicle itself will not only be aware of the vehicles around it due to a plethora of on-board sensors (such as radar and computer vision systems) but also of their intentions through constant DSRC communication exchanges.Also road signs will be made redundant as the data is sent directly to the vehicle.Speed limits will be mandatorily controlled to increase the safety and efficiency of roads.Such a detailed and consolidated picture of the local environment around the vehicle has the potential to reduce the number and severity of collisions and so increase traveler safety.Other participants in traffic, such as pedestrians and cyclists, will be similarly equipped and communicating constantly with the surrounding vehicles and travelers, thus the safety of even the most vulnerable road users will be enhanced.
2. In large cities and on congested roads the data density will be vast.For the individual traveler (now taken to mean the devices and systems that are assisting a person making a journey) only a small fraction of the received data will be relevant (and even less will be useful) and some form of stream processing will be necessary just to prioritize the messages in order of immediacy and relevance let alone acting on their content.That is not to say that any of the data is useless.Indeed, it contains trends and anomalies that are useful for planning not just the next trip but also the transportation capacities required in the future.Extraction of these trends and anomalies must be automated and conveyed to the relevant user in an easily understandable form.
3. Since there is no guarantee that the data available to a traveler are of useable quality or even available when needed, filling the spatial and temporal data gap is a challenging issue.Is it meaningful to fill the gaps with data from yesterday or even a minute ago?Can statistical machine learning techniques such as Support Vector Regression help?The answers are not clear and must depend on what the data are to be used for.After all a bus timetable is simply a prediction of often dubious reliability.
4. Visualization of the huge, multi-dimensional data sets generated will not be easy.Many users will have their own requirements and will probably want to construct queries and visualize the results using some sort of OLAP cube.
It is unlikely that the mobile device of an individual user will have the computational power or storage for such a task.Will cloud computing come to the rescue?Will peer-to-peer systems help with data storage and download?
The physical presentation of the data is also an issue.An in-vehicle display cannot be too obtrusive and certainly cannot interfere with the driver's ability to control the vehicle (at least until the vehicle is fully autonomous).Questions of relevance, urgency and safety need to be addressed.
There are numerous unanswered questions raised in the paragraphs above.They are potentially solvable in isolation but all the possibilities will only be leveraged through unified study.CTS is aimed fairly at this broad field.

Decentralized Computing
While applications are concerned with what data is relevant to answer a particular query, decentralized computing is concerned with where this data resides and how to access this data.First we will explain what these questions mean in the CTS environment, and then why these questions are different than the ones answered by traditional DBMSs.
Consider for example the query: what is the average speed of traffic a mile ahead of me?Sometimes the query can be posed to a central server, but sometimes a server with this information is unavailable (e.g., because the query pertains to a congested side street that is not instrumented with speed sensors), and the query needs to be answered by polling the vehicles ahead.However, the network IDs of these nodes (the vehicles) are not known.Thus, for this query it is not known where the data resides, and how to get to it.The answer in this case may be to use short-range wireless communication such as Wi-Fi or DSRC to disseminate the query to neighboring nodes transitively.In other words, the limited transmission range of the network is used to compensate for the lack of ID knowledge.
These questions are not addressed by traditional DBMSs.The data integration problem studied by the database research community assumes that the data is always available, but the integration part is the problematic part.In distributed databases it is assumed that there are directories that map data to in the information processing chain.These individuals act and interact in a larger societal context, involving also government (transportation authorities), stakeholders, and transportation providers.These groups have different roles and responsibilities, but also different values and interests.Research questions in this area concern the complex decision making processes, economic models including novel fare models (e.g., e-tolling, e-ticketing, ride sharing, virtual fencing), and also the demands of the community for privacy.

OUTLOOK
A discipline is only as good as its academic community.If this paper finds your support or meets your interests you are cordially invited to participate and engage.The infrastructure set up so far is a beginning but requires your collaboration, be it the Wikipedia entry, the CTS webpage, or the CTS workshop series.These are all small seeds that-if they grow-can lead to conferences and journals on CTS, not only in the content but also in name.Finally, the community should shape its own academic programs or introduce core subjects on computational transportation science into the programs on transport engineering, electrical engineering, software engineering, and geographic information engineering.The spread demonstrates the inter-disciplinarity of computational transportation science, illustrates that engineering problems do not present themselves any longer wholly contained in one traditional discipline, and supports the fundamental concern that engineering disciplines have grown to be too narrow [3].