U.S. Department of Transportation
Federal Highway Administration
1200 New Jersey Avenue, SE
Washington, DC 20590
202-366-4000
Federal Highway Administration Research and Technology
Coordinating, Developing, and Delivering Highway Transportation Innovations
| SUMMARY REPORT |
| This summary report is an archived publication and may contain dated technical, contact, and link information |
|
 |
| Publication Number: FHWA-HRT-15-067 Date: August 2015 |
Publication Number: FHWA-HRT-15-067 Date: August 2015 |
This chapter discusses the current and future applications of connected driving and traffic simulators and transportation policy and planning models, as well as technical and resource challenges that may need to be addressed to advance the progress of these federations.
Driving simulators are used throughout academia, Government, and the automotive industry to address fundamental and applied research questions. There is a range of floor-mounted driving simulators available, with some having no motion, some having minimal motion, and some being able to simulate almost all aspects of motion. An entirely new category of simulator has recently emerged, which includes both hand-held and head-mounted simulators.
The U.S. Department of Transportation’s Office of Safety uses the following three simulators at the Turner-Fairbank Highway Research Center (TFHRC): (1) an advanced driving simulator, (2) a lower fidelity simulator, and (3) a sign simulator. A total of 95 percent of the research funded by the human factors group at TFHRC makes some use of the driving simulators. Examples of research questions addressed on the high-fidelity driving simulator include the following: What happens to drivers approaching an intersection who are caught in a dilemma zone? Does the eye glance behavior of drivers viewing variable message signs, traffic information signs (displaying travel time), or electronic billboards indicate a distraction created by any one or more of these display types? How do different ways of displaying information on active traffic management signs (variable speed signs and lane closure signs) affect drivers’ comprehension? An example of a research question addressed on the low-fidelity simulator included an investigation of how well left-turn warning systems functioned when V2V communications were possible.
Connecting driving simulators would enable new research. With current driving simulators, it is possible to study the interaction of a driver with another scripted vehicle. However, researchers do not know well enough how drivers interact with one another and other road users. Thus, almost any situation where safety is at issue in an incident involving two interacting drivers requires connected driving simulators communicating with each other in real time. This allows the two or more drivers, bicyclists, or pedestrians to see the same world and the other drivers in that world. Examples of problems that could be studied include the following:
Many of the technical challenges at the most basic level have been overcome for distributed asynchronous simulation at different sites and distributed synchronous simulation at the same site. However, there have been relatively few studies using either of these types of connected simulators. Because the engineering advances required have been available for well over a decade, why has the use of connected simulators for research been so limited? One explanation, discussed at the TRB workshop, is a lack of both available resources and compelling use cases. The next sections discuss asynchronous and synchronous federation at different sites, technical implementation issues, and what advances in technology could reduce these issues.
Distributed Asynchronous Simulation
Distributed asynchronous simulation at different sites has long been possible with driving simulators that provide a medium level of flexibility in programming scenes and scenarios.(2) However, with more powerful simulators, that success has been harder to achieve even though not everything needs to be shared for distributed asynchronous simulation to work. It is enough that two sites have identical visual databases, terrains, scenarios (behaviors), logical road networks, and entity positions. In theory, it should be easy enough to create a model with these shared elements and then run one and the same virtual world (model) on two simulators of the same make at different locations and different times. However, this has turned out to be surprisingly difficult. Software is constantly being updated, and version compatibility issues arise; some simulators use one screen, some three, and some five, creating differences that lead to unexpected problems; and differences exist in the movable objects (e.g., vehicles and pedestrians) across sites that make a motorcycle appear in one simulator and a truck in another for example. All these issues can and have been resolved for limited cases. However, the virtual world built on one simulator cannot simply be copied to another simulator made by the same company and expected to run. Improved interoperability would clearly alleviate many of these problems.
Advances in hardware are helping increase opportunities to connect floor-mounted driving simulators asynchronously and lowering cost while keeping functionality. High-performance mounted simulators that once cost hundreds of thousands of dollars can now cost tens of thousands of dollars.
Finally, cross-platform driving simulation—or crosses between single-purpose, high-fidelity simulators and virtual reality—may be the holy grail of advances in driving simulation, although it is very challenging. To solve this problem, at the very minimum, one needs to ensure that the two platforms render the same world with the same objects. Assuming this can be achieved, there are four options: (1) make both simulators high-level architecture (HLA) compliant and run a common run-time infrastructure, (2) make both simulators distributed interactive simulation (DIS) compliant and pass the scene state back and forth, (3) use the same scenario software in both platforms, or (4) develop a custom interoperability solution.
Distributed Synchronous Simulation
The first study using distributed synchronous driving simulation at a single site was reported in 2003.(3) Very few studies have been reported since, although that number is growing.1 However, most universities (or other sites where driving simulators are operational) cannot currently afford multiple simulators. Thus, distributed simulation with the current cost structure is typically only going to occur at geographically distant sites. In fact, no studies have been reported at different sites of distributed asynchronous simulation. Currently, across different sites, attempts are being made to implement distributed simulation as a peer-to-peer model. Here, the peers have their own simulation software and compute the state information locally based on the inputs and then pass information through a server about the state (e.g., the position of the vehicles in the local environment), and the server distributes that state information to the other peers. The peers then populate the updated world with the positions of each of the other vehicles and the driver’s vehicle. The following three problems unique to distributed synchronous simulation using the peer-to-peer model dominated the workshop discussion: (1) network latency can be too long (greater than 50 ms), (2) network drops can and do occur, and (3)security can be an issue.
There are examples outside of the transportation research community where this geographically distant, synchronous driving simulation occurs. For example, in the gaming industry, long latencies (upward of 500 ms) and network drops are typically dealt with by predicting the behavior of an agent. However, in an experiment designed to uncover the behavior of a driver, this fix is clearly contraindicated. Thus, it is by no means clear that the gaming industry has a solution for the fundamental behavioral problems that face researchers interested in distributed synchronous simulation at distant sites.
Faster internet speeds (lags less than 50 ms) would have a strong impact on researchers’ ability to connect driving simulators in real time with one another at different sites. These developments are on the horizon for both copper wire and optic fiber.(4,5)
Perhaps the most disruptive technologies to affect the development of connected simulators across different sites running in real time are head-mounted devices. These devices allow the participant to navigate through an immersive three-dimensional (3D) world (e.g., using virtual or augmented reality), making it possible for the participant not only to navigate through the 3D world but to also see his or her hands on the steering wheel and reaching for the controls, all while receiving feedback.(6) Connecting tens, hundreds, or thousands of drivers in the same virtual world simultaneously may soon be technically feasible. For drivers using connected simulators to see another driver’s head, eye, and hand movements, there will need to be the development of avatars in the simulated vehicles of the other drivers that will capture said movements. No current examples were found.
Similar to driving simulators, there are also several different types of traffic simulators. These include nanoscopic (individual vehicles and driver behaviors), microscopic (vehicles), mesoscopic (groups of vehicles), and macroscopic (aggregate flows only). Traffic simulators are also classified according to whether they are continuous or discrete in the time, space, and state domains.
FHWA’s Office of Operations frequently uses traffic simulation in areas such as congestion management, ITS deployment, traffic operations, emergency management, and freight management and operations. Through the Second Strategic Highway Research Program (SHRP2), the Office of Operations has developed a multiresolution modeling framework that ties together macroscopic models of traffic in distinct transportation subnetworks such as freeways, corridors (including freeways and parallel arterials), surface street grid networks, and rural highways; microscopic models that predict individual vehicle trajectories and diversions, which are especially useful for congested conditions, complex geometric configurations, and system-level impacts of proposed transportation improvements; and mesoscopic models that tie together the macroscopic and microscopic models. The traffic analysis tools are now being supplemented with those needed for integrated corridor management (the reduction of underused capacity in the form of parallel roadways, single-occupant vehicles, and transit services that could be better leveraged to improve person throughput and reduce congestion) and active traffic demand management (the dynamic management of recurrent and non-recurrent congestion based on prevailing and predicted traffic conditions).
Distributed asynchronous traffic simulation has been available for some time; FHWA pioneered the development of guidelines over a decade ago.(7) Unlike the software for driving simulators, the software for traffic simulators is not typically configured for a particular site. Moreover, there are generally no connections to hardware that further complicate the possibility of distributed, asynchronous traffic simulation. Therefore, researchers can and do replicate each other’s simulations at different sites.
A promising direction in distributed traffic simulation is the development of nanoscopic and microscopic models that can be used to influence operations, planning, and policy at much larger scales than previously possible.(8) By connecting or federating traffic simulators, much larger regions could be modeled at a microscopic level. There is also a real need for connecting different categories of simulators and models. For example, through the SHRP2 C10A project, researchers have developed an integrated traffic modeling framework with true interoperability. It starts with an activity-based travel demand model. The outputs (origin-destination tables and activity matrices) of the travel demand model are then fed into a mesoscopic model to get traffic flows and capacity constraints, which are then input into a microsimulation model to evaluate the effects of a strategy (e.g., ramp metering) in terms of impact on traffic flows. The outputs of the microsimulation model are then fed back into the mesoscopic model to get new diversions that will also affect demand (e.g., time of departure or mode choice), which are then fed back to the activity-based travel demand model to get new demands until one reaches convergence. Each time a transition between levels (simulation or models) is made (from macro- to meso- to micro- and back), the fidelities are different and critical information is lost. Improvements in the interoperability of different categories of connected simulators and models would improve the situation considerably.
Distributed synchronous simulation has also been around for over a decade.(9) Its use makes it possible to employ multiple microscopic simulators in the analysis of a transportation network in real time as opposed to the analysis of a single location. The computational load is distributed across perhaps hundreds of different computers. This can be done in several ways. This report focuses only on spatial distribution or decomposition; the traffic network is decomposed into separate traffic sub-networks, and communication is needed only when one vehicle must travel to an adjacent subnetwork.(9) A distributed database must be maintained to hold the results, and a distributed synchronization mechanism is used to coordinate the flow of information.(10) While there are challenges, they are generally less difficult than those that occur in the world of driving simulation, where latency on the order of milliseconds affects performance. With distributed traffic simulators solving problems in real time, the level of analysis may be as large as a second or longer.
Another future direction might combine driving and traffic simulators, something that has already been done with a single driving and traffic simulator.(11) However, as previously noted, with the advances in technology, there will soon be an opportunity to bring together tens of thousands of participants making route choices while driving simulators in real time. Those choices would be embedded within a distributed nanoscopic traffic simulators that update their models based on the driver choices.
There are several technologies that have enabled the technical advances in connected traffic simulation. Perhaps of most general importance are increased computer capabilities (central processing unit speed and random access memory) and advances in the tools and techniques used to distribute the computational load across multiple processors and computers. Also important is the development of methods to maintain interoperability across distributed simulations. In fact, many articles in the literature review address issues of interoperability of distributed simulations. The majority of this research (i.e., 9 out of 10 articles) was published after 2010. (See references 12–20.) Finally, within the traffic simulation community, there has been a large advance in the efficiency of various algorithms central to performance. Given the previous discussion, there is much to be done in improving existing technologies.
Travel demand modeling and prediction is a complex process that requires input from more general models of land use and more detailed simulations of traffic.(21) The transportation models and traffic simulators are now connected but only loosely. Land use—where houses will be built and where jobs are located—is something that transportation planners have to consider because it drives transportation needs of individuals. Land use changes happen slowly, with visible changes happening 2–30 years after they are discussed. Therefore, transportation planners start by looking 2–30 years in the future from today’s conditions at what the land use models predict will happen.
The traditional four-step approach to travel demand modeling dates back some 50 years, with the following four steps:
Metropolitan planning organizations (MPOs) and State transportation departments use the four-step process for comprehensive multimodal transportation planning, air quality conformity analysis, and major project evaluation.
The four-step process has received some criticism that the steps are not well integrated and that trips do not occur in isolation. In reality, travelers engage in tours with multiple stops, and the mode chosen for the first leg of a tour may affect which modes are available for subsequent legs. For example, if a traveler left his or her car at home and took transit to work, then he or she is unlikely to use that car for a lunch outing.
As computers have become more capable and transportation planning problems have become more complex, several trends have emerged. First, the use of activity-based models has become more common, particularly at large MPOs.(22) An activity-based modeling framework (figure 1) considers household activities and tours, deriving individual trips from the tours.
Second, dynamic traffic assignment models have emerged for route choice assignment.(23) These models perform the route choice function at a greater level of temporal and spatial detail than older traffic assignment models. In some cases, traffic microsimulation has been used on a regional level.
Third, models have begun to address greater spatial and temporal detail. A traditional four-step process using the computers of several decades ago might have divided an entire metropolitan area into a few thousand zones while considering traffic assignment for four broad periods of the day (morning peak, midday, afternoon peak, and night). More recently, some cities have built models based on individual land parcels, while others have created models with a time-of-day resolution measured in minutes rather than hours. With finer-grained time resolution, it becomes possible to analyze short-lived congestion issues and real-time traffic management strategies.
Fourth, there is an interest in model integration. For example, the Rapid Policy Assessment Tool is one of several strategic planning tools that deal with aspects of both transportation and land use. Several SHRP2 capacity pilots are underway to integrate activity-based models and dynamic traffic assignment.
Connected interoperable models will become increasingly important in the very near future because of what one cannot do in the modern policy environment, especially for planning models, which is to build new policy contingencies into a model. For example, suppose that someone has a regional demand model and is attempting to take into account automated vehicles and how they are deployed (e.g., in fleets, with individual owners, and/or as shared use cars). There are many permutations in the automated vehicle space (e.g., synchronizing signals and how many cars are equipped with automated driving suites (ADSs)). Trying to code all of these permutations into a detailed regional demand model is a next to impossible task. Instead, one could construct simplified planning models (e.g., on the effects of automated vehicles that may be reflected in a traffic model via car following) that can include conceptual findings from other models and adjust accordingly. Simply put, what is required is an educated guess based on the outputs from another model and the ability to integrate that educated guess seamlessly into a heavy-duty traffic model. This level of interoperability would provide not only labor savings but also the flexibility to alter assumptions efficiently and interrogate assumptions and compare them with data from implementation on the roads.
There are many different types of models used in the formulation of transportation policy. Among the various economic models used, one can undertake the following four types of analyses: cost effectiveness analyses, benefit-cost and net benefit analyses, lifecycle cost analyses, and multiple accounts evaluation. The economic models are used to determine value of a policy, project, or program.
For example, FHWA’s Office of Transportation Policy Studies has sponsored the development of the Highway Economics Systems Requirements (HERS), an engineering and economic analysis tool. The tool uses engineering standards to identify highway deficiencies and then applies economic criteria to select the most cost-effective mix of improvements for system-wide implementation. HERS is designed to evaluate the implications of alternative programs and policies on the conditions, performance, and user cost levels associated with highway systems. The model provides cost estimates for achieving economically optimal program structures as well as predicting system condition and user cost levels resulting from a given level of investment. The cost equations are always being updated to reflect new realities. Many other tools are available.
There are several questions that could be answered if the research questions addressed with traffic and driving simulators were formulated in the context of the issues critical to the HERS model and then could easily be input to the HERS model. First, consider traffic simulators. In the HERS model, there are equations that predict the vehicle operating costs per mile traveled. These costs include fuel consumption, tire wear, oil consumption, maintenance and repairs, and mileage-related depreciation. There are also equations that relate those costs to characteristics of a roadway section, which are needed because the HERS model evaluates improvements that would add to highway capacity by adding lanes or improving pavement condition through resurfacing and reconstruction. Both types of improvements will affect operating costs, and traffic simulators could provide the required information. Next, consider driving simulators. Behavioral studies on driving simulators that would be valuable for models like HERS include research, which leads to a better understanding of the impact of system-wide congestion pricing, the discomfort costs of driving on rough pavements on an aging highway system, the influence of curvature on speed, the effect of different levels of automation on the cost of the time spent in a vehicle, the effect of variable speed limits, and the impact of real-time traveler information on drivers’ route choices. In terms of safety, most of the things that could be studied with driving simulators would be of some value to the HERS model. For example, if research through connected driving simulators or other means could provide better information about how highway conditions like congestion would affect the rates of crashes of different severities and how the radical improvements in vehicle technologies in the future would affect the crash rates, then all of this information would be useful for the HERS model.
1 Thomas Kerwin, personal communication of Donald Fisher with Thomas Kerwin, Ohio State University (OSU), January 8, 2016.
