Automobile Traffic

Automobile traffic has a very extensive body of literature involving simulation. Most people in the industrialized world deal with automobile traffic on a daily basis, and many studies are funded annually to alleviate existing or potential traffic problems. Several academic journals are dedicated exclusively to automobile traffic dynamics, new textbooks on the subject are published regularly, and the number of articles published each year dealing with automobile traffic number in the hundreds.

Consistent with the paradigms listed in table 2.1, automobile traffic modelers have been faced with the choice of whether to model their traffic macroscopically as continuous flows (like modeling water flowing through a pipe), or model individual vehicles. There are many examples in the literature of both approaches.

Macroscopic simulations have the advantage of being mathematically compact, and can be expressed as a set of differential equations that can be solved with little or no computer coding. The method has much in common with fluid dynamics and information theory, and is best suited to systems that consist of relatively homogeneous units with limited and predictable interactions. One common theme in traffic simulation using this approach is the addition of factors that allow the equations to more closely describe the observed complexity of real-world traffic systems. An example is Leo and Pretty (1990), using a derivation of Roe's flux difference splitting algorithm for a macroscopic description of a model to anticipate changes in density when traffic is braking.

In recent years, much more of the literature in traffic simulations has turned towards discrete event and individual-based models. Initial efforts along this line were generally large scale, large budget projects, due to the considerable investment in hardware and programming ability that was needed to complete even modest sized simulations. It was common practice to simulate smaller systems on available computers and extrapolate the results onto larger computers. Mahmassani et al. (1990) used a supercomputer to simulate larger versions of the smaller traffic models in an environment relatively unconstrained by computing resources, and concluded that such extrapolations are largely justified. Others, like Ho (1991) experimented with networks of smaller computers to simulate the power of a supercomputer. Nagel and Schleicher (1994) reported ``extremely fast'' execution speeds for traffic simulations that were set up to run on several relatively inexpensive computers operating in parallel.

Other issues that were addressed in automobile traffic simulations that had implications later in other traffic models were improvements in how random numbers were used (Rathi, 1992), updating of individuals or entities through dynamic stochastic assignment (Cascetta and Cantarella, 1991; Vythoulkas, 1990), driver information about the state of the system (Bir Akiva et al., 1991; Ho, 1991; Leiser and Stern, 1988; Reiss et al., 1991), and improvements on how driver information is represented (Alvarez et al., 1990).

  As increases in computing power become available, more interest has turned to exploring the dynamic nature of automobile traffic. Importantly, many of the complex dynamics of traffic systems are being seen as emergent properties of decentralized interactions between individual vehicles in the models. Resnick (1996) described how complex traffic patterns emerged in systems where individual vehicles were given extremely simple rules in informal experiments conducted by high school students. A more formal description of complex behavior resulting from emergent phenomena in traffic systems is discussed in Barrett (1996), using only slightly more elaborate rule sets than those used by Resnick's students. Successes in capturing complex behavior in decentralized simulations has led some researchers to consider automobile traffic to be a self-organizing system (Nagel et al., 1996a), and application of those principles has been part of the framework for a major transportation simulation project at Los Alamos National Laboratory (Smith et al., 1993). Research in improvement of individual-based models for automobile traffic has included allowing the (simulated) drivers to make their own decisions as to how they navigate the networks (Nagel, 1997), and how to allow for representation of such a model in a realistic computing environment (Nagel et al., 1996b).

  One notable application of bottom-up modeling of automobile traffic appeared in a collaborative effort of the Robot Auto Racing Simulation (RARS), which was a contest held via the internet in 1994 and 1995 (Gymer, 1995). The RARS simulation was written by Mitchell Timins at Penn State University, and consisted of a racetrack with several simulated autos that would race around the track against each other. The cars were coded as objects that were subject to physical laws of their racetrack (mass of the car, friction with the road surface, etc.), but their perceptions of the world around them, their responses to the racetracks, and thus their controlling mechanisms of how they navigated the racetrack were encapsulated within the car objects themselves. By having the controlling programs for each car segregated from the main program, it was possible for each car to have a distinct set of instructions on how it would go about racing around the track. A contest was then announced over the simulation and computer game design Usenet groups with an open invitation for participants to design software controllers for the simulated racecars. The contributed controllers then were pitted against each other in a race on a specified date, and the results published to the relevant discussion groups. While no scholarly papers have been generated from this exercise to date, innovative modeling and coding techniques were demonstrated that were of equal quality in terms of realistic behavior to simulations that are described in the peer-reviewed literature described in the preceding paragraphs.

The main significance of the RARS project is the illustration of how advances in object-oriented programming tools and sophisticated development environments on cheap computers have brought the arena of complex traffic simulation from the exclusive domain of large organizational efforts to the status of computer games. In the present state of instantaneous communication between researchers via the internet, many innovative ideas about how to simulate diverse systems are communicated throughout the world and be propagated through many institutions long before they are published in peer-reviewed articles. The large community of parallel developers seem to fit the model of a self-organizing system that is discussed in other parts of this paper.