Traffic Simulation and Transportation Engineering

With continuing improvements in computer hardware and software, traffic simulation models have become more useful and popular. One widespread use of such models is for video games, such as Grand Theft Auto with more than 250 million lifetime sales in five versions (Cragg 2016). There is also a research community using traffic simulation for transportation engineering. Numerous software programs exist for such simulations, and more keep appearing (Federal Highway Administration 2016). Dozens of traffic simulation application papers are submitted annually to the Journal of Transportation Engineering (Part A). However, many of these submissions are declined prior to review or during the review process. This editorial is intended to explore the role of traffic simulation in furthering the state of knowledge for transportation engineering.

Many traffic simulation studies involve assessing the effects of design changes of various types. A typical study involves calibration of model parameters to reflect observed traffic behavior and a report of the agreement between model estimates and observed behavior. Some studies also include a validation comparison between model estimates and observed behavior not used for calibration. After calibration, design parameters such as roadway geometry, traffic signal settings, operating rules, vehicle characteristics, and driver characteristics may be varied to assess performance improvements. Optimization may be incorporated in such a search. Researchers then make recommendations for design parameter improvements.

The difficulty editors and reviewers have with such studies is that there is no way of knowing from the submissions whether the traffic simulation would be realistic over the full range of parameters used. Just because model estimates compare well to current conditions does not necessarily mean they will make correct predictions for nonobserved situations. Although examination of the underlying traffic simulation software can be helpful, this is typically beyond the scope of work of a paper reviewer or reader. For relatively simple traffic simulations, the model equations can be presented in a paper for examination by reviewers. More generally, what is needed is empirical assessment of design parameter changes, either on a test track or on actual roadways. For example, a traffic simulation study recommending improved traffic signal parameters could be tested by adopting the recommended parameters and observing the result in the field. Unfortunately, many submitted traffic simulation studies do not carry the work through to such field testing.

Archival peer-reviewed journals such as the Journal of Transportation Engineering are intended to publish work that will contribute to the state of knowledge and have an impact on engineering decisions. With this goal, the readers and users of the work must be confident that decision-making recommendations are valid. Although traffic simulation studies without field testing are useful for indicating useful designs to test, a pure traffic simulation study is more suitable for a conference presentation than an archival journal. In contrast, studies of novel calibration and validation methods for models as well as validated model applications can be significant contributions.

Driving simulation studies using a combination of computer models and human drivers in a simulation environment have similar issues. Of course, many of these studies are intended to assess driver behavior or vehicle design changes, which are not within the scope of the Journal of Transportation Engineering (Part A) and belong in a human factors–oriented journal. But some do involve infrastructure elements, such as driving simulations with different traffic calming measures in place. Once again, an empirical test of such measures on actual roadways is a useful complement to the simulation studies.

The technological transformation of vehicle automation and connectivity is another good example of the role of traffic simulation. Vehicle automation and connectivity currently is a topic of widespread interest and rapid technological development. Although numerous automation simulation studies have appeared, developers and regulators both believe any simulations must be complemented with roadway testing, either on test tracks such as Mcity (Lynch 2017) or on actual roadways (National Highway Traffic Safety Administration 2016).

Safety studies provide another challenge for the use of simulation models. Although the aggregate number of vehicle crashes is large, the number of crashes per kilometer of vehicle travel is small. With 6 million U.S. roadway crashes per year and $5\times {10}^{12}\mathrm{km}$ of vehicle travel, the chance of a crash per kilometer of vehicle travel is only $1\times {10}^{-6}$. With 6.8 million km of roadway, only one crash is expected per year on a kilometer of roadway. Unfortunately, an assessment of crash risks solely from simulation models is uncertain, but the rarity of crashes makes empirical testing difficult. Studies of actual versus simulated safety outcomes could be useful contributions.

In summary, traffic simulation can be a useful tool for a preliminary assessment of alternatives. Papers comparing real and simulated results and suggesting improved calibration methods or models can be very useful. However, a purely simulated traffic study often falls short of the requirements of a peer-reviewed archival journal without a component of field testing.