Traffic Generation for the Vti Driving Simulator

Driving simulators are often used in a type of experiments where the result may depend on traffic conditions. One example is the evaluation of cellular phones and how they affect driving behavior. It is clear that the ability to use phones during driving depends on the amount of traffic, and that realistic experiments in driving simulators therefore must include surrounding traffic. This paper describes an ongoing project on the development of a micro simulation model for generation of stochastic traffic for the driving simulator at the Swedish National Road and Transport Research Institute (VTI). The goal is to generate a traffic stream, corresponding to a given target flow, and simulate realistic interactions in the neighborhood of the driving simulator vehicle. The model is built upon established techniques for time-driven micro simulation of traffic, where driver behavior is described by a set of fundamental sub-models for car-following, speed adaptation, lane changing etc. In this respect the model is similar to ordinary traffic simulation models, like AIMSUN (1), MITSIM (2) and VISSIM (3). A fundamental difference is, however, that we in the new model only have to consider the closest neighborhood of the driving simulator vehicle. The neighborhood can be interpreted as a window that consequently moves with the speed of the simulator vehicle. The basic purpose, and motivation for the model, is traffic generation of vehicles within sight distance of the simulator vehicle. To make this traffic realistic and to allow for speed changes of the simulator, a much wider window must be considered. The size of the window has, on the other hand, an impact on the computational time. A compromise is obtained by dividing the window into one inner and two outer regions. Close to the simulator vehicle (inner region) the “full” micro simulation model is used, while a simplified and less time consuming model is used in the two outer regions. In the latter model, vehicles are simply assumed to travel at their desired speeds. In experiments using driving simulators, a key issue is to minimize statistical variations of the result by giving all the subjects of the experiment the same conditions. With stochastic traffic different drivers will, on a micro level, experience different situations. They will soon come into new situations and interact with other traffic in a unique way. We have to accept these differences and the uncertainty that it implies, and instead say that conditions are comparable, but at a higher level. That is, each driver “feels” the same traffic flow and, in the long run, meets the same type of traffic. The first version is able to simulate traffic on a motorway with two lanes in each direction and without ramps. We are currently working on a rural road model, with possible overtakes and conflicts with oncoming traffic. INTRODUCTION Driving simulators are widely used for experiments concerning driver behavior, vehicle characteristics, road design, etc. Two of the main advantages are that it is possible to accomplish safe experiments with equal conditions for all test drivers. The driving simulator at the Swedish National Road and Transport Research Institute (VTI) utilizes a real vehicle cab with an advanced motion system. The surroundings, the road and the vehicles, are presented for the simulator driver on three screens. A new, more advanced, simulator is under construction. The new simulator will offer, among others, higher accelerations and smoother movements. The old simulator has been used for many different types of experiments, for example experiments concerning road design, the use of cellular phones during driving, vehicle characteristics, and human-machine interfaces. (4) Since the idea with experiments in driving simulators is to create an environment that resembles driving in a real traffic system as close as possible, it is obvious that trustworthy ambient traffic is needed. But is simulation of ambient traffic always desirable? Simulation of ambient traffic increases the statistical variation of the results due to different conditions between the test drivers. Drivers will, on a micro level, experience different situations. If it is important that all drivers experience exactly the same situations before a test event, the experiment should not include stochastic simulation of ambient traffic. This can, for example, be in experiments concerning reaction times, where the reaction time may differ within one driver, depending on the situations that he or she has experienced before the event. If, for example, one driver has been overtaken by several vehicles, before the point in time where the reaction time measurement take place, and another driver has been able to travel unconstrained with respect to ambient traffic, their conditions differs. It can then be hard to observe whether the difference in reaction time depends on the test drivers or on their different test conditions. Simulation of ambient traffic is, however, needed in studies where the amount of traffic has a large influence on the driver’s ability to drive the vehicle. When performing experiments about the use of cellular phones during car driving, the need of ambient traffic is obvious. It is a lot easier to drive a vehicle and use a cellular phone if there is no surrounding traffic. The same holds for applications in the area of Intelligent Transportation Systems (ITS), for example different kinds of route guidance systems. Ambient traffic is also an important ingredient in experiments related to collision warning systems and design and placing of road signs. As mentioned above different drivers will indeed experience different situations but will in the end “feel” the same traffic load and meet the same type of traffic. In these types of experiments we have to accept the increase in statistical variation and instead say that the drivers test conditions are comparable at a more aggregated level. Traffic simulation is a useful and popular tool in studies of traffic systems. On the basis of a road network description, an Origin-Destination matrix (OD-matrix) and information about driver characteristics like desired speed and desired acceleration, a traffic system can be simulated and the user can get information about average speeds and traffic flows on different road sections. The vehicles’ state variables are updated continuously, with short time steps, less than one second between updates. The simulation is done at a microscopic level, which implies detailed modeling of vehicle characteristics and driver behavior. The output is often macroscopic measurements like average flow, speed and density. Calibration and validation are usually done at a macroscopic level, that is against traffic flows, queue lengths, average speeds etc. A simulation model uses different sub-models to model driver behavior in different situations. One essential behavioral model is the car-following model. This model describes a driver’s behavior when following another vehicle. The basic idea is to conclude when a vehicle is free and when it is following another vehicle and what actions the driver applies in each case. Several car-following models have been developed through the years. The research in this area started in the fifties, see Car-following: a historical review (5) for an overview of car-following models. When simulating traffic on multilane roads, a lane-changing model is needed. This model is used to decide whether a vehicle should change lane or not. This is often done in several steps, for example first deciding whether it is desirable to change lane and then concluding if it possible and safe to do so. To be able to model roads with oncoming traffic, an overtaking model is necessary. This model cannot only concern the actual lane change to the oncoming lane. Instead, the whole overtake process must be considered, starting with deciding whether the driver is willing to execute an overtaking at the available sight distance. A simulation model also has to include models for calculation of desired speeds at different road sections and models for calculation of vehicles’ route choices through the network.