Raymond Hoogendoorn

Calibration of microscopic traffic-flow models using multiple data sources.

Parameter identification of microscopic driving models is a difficult task. This is caused by the fact that parameters—such as reaction time, sensitivity to stimuli, etc.—are generally not directly observable from common traffic data, but also due to the lack of reliable statistical estimation techniques. This contribution puts forward a new approach to identifying parameters of car-following models. One of the main contributions of this article is that the proposed approach allows for joint estimation of parameters using different data sources, including prior information on parameter values (or the valid range of values). This is achieved by generalizing the maximum-likelihood estimation approach proposed by the authors in previous work. The approach allows for statistical analysis of the parameter estimates, including the standard error of the parameter estimates and the correlation of the estimates. Using the likelihood-ratio test, models of different complexity (defined by the number of model parameters) can be cross-compared. A nice property of this test is that it takes into account the number of parameters of a model as well as the performance. To illustrate the workings, the approach is applied to two car-following models using vehicle trajectories of a Dutch freeway collected from a helicopter, in combination with data collected with a driving simulator.

Generic Calibration Framework for Joint Estimation of Car-Following Models by Using Microscopic Data

Microscopic simulation models have become widely applied tools in traffic engineering. Nevertheless, parameter identification of these models remains a difficult task. This is partially because parameters are generally not directly observable from common traffic data; also there is a lack of reliable statistical estimation techniques. This study puts forward a new general and structured approach to identifying parameters of car-following models. One of the main contributions of this study is joint estimation of parameters for multiple vehicles. Furthermore, prior information on the parameter values (or the valid range of values) can be estimated. The study also deals with serial correlation in the trajectory data. In doing so, the newly developed approach generalizes the maximum likelihood estimation approach proposed by the authors. The approach allows for statistical analysis of the model estimates, including the standard error of the parameter estimates and the correlation of the estimates. With the likelihood ratio test, models of different complexity (defined by the number of model parameters) can be cross-compared. A useful property of this test is that it takes into account the number of parameters of a model as well as the performance. The approach is applied to car-following behavior by using Dutch freeway vehicle trajectories collected from a helicopter.

Automated Driving, Traffic Flow Efficiency, and Human Factors

Automation may be assumed to have a beneficial impact on traffic flow efficiency. However, the relationship between automation and traffic flow efficiency is complex because behavior of road users influences this efficiency as well. This paper reviews what is known about the influence of automation on traffic flow efficiency and behavior of road users, formulates a theoretical framework, and identifies future research needs. It is concluded that automation can be assumed to have an influence on traffic flow efficiency and on the behavior of road users. The research has shortcomings, and in this context directions are formulated for future scientific research on automation in relation to traffic flow efficiency and human behavior.

Wiedemann Revisited: New Trajectory Filtering Technique and Its Implications for Car-Following Modeling

A new data-driven stochastic car-following model based on the principles of psychospacing or action-point modeling is presented. It uses empirical or experimental trajectory data and mimics the main microscopic behavioral characteristics present in the data. In the action-point model, regions are defined in the relative speed–distance headway plane, in which the follower is likely to perform an action (increase or decrease acceleration) or not. These regions can be established empirically from vehicle trajectory data and thereby yield a joint cumulative probability distribution function of the action points. Furthermore, the conditional distribution of the actions (the size of the acceleration or deceleration given the current distance headway and relative speed or given the acceleration before the action) can be determined from these data as well. To assess the data correctly, a new filtering technique is proposed. The main hypothesis behind this idea is that the speed profile is a continuous piecewise linear function: accelerations are piecewise constant changing values at nonequidistant discrete time instants. The durations of these constant acceleration periods are not fixed but depend on the state of the follower in relation to its leader. The data analysis illustrates that driving behavior shows nonequidistant constant acceleration periods. The distributions of the action points and the conditional accelerations form the core of the presented data-driven stochastic model. The mathematical formalization that describes how these distributions can be used to simulate car-following behavior is presented. Empirical data collected on a Dutch motorway are used to illustrate the workings of the approach and the simulation results.

Mental workload, longitudinal driving behavior, and adequacy of car-following models for incidents in other driving lane.

The values on parameters describing longitudinal driving behavior in car-following models differ substantially between drivers. Different individual interactions with the environment are assumed to play an important role, which might be explained through mental workload. Therefore a driving simulator experiment with a repeated measures design was performed to investigate to what extent perception of an incident in the other driving lane influences physiological indicators as well as subjective estimates of mental workload and longitudinal driving behavior. As almost none of the current models of car-following behavior incorporate mental workload as a determinant of driving behavior, an investigation was conducted by using a calibration approach for joint estimation to determine whether these models, represented by the intelligent driver model and the Helly model, adequately described longitudinal driving behavior in case of incidents in the other driving lane. The results indicated that perception of an incident in the other driving lane influenced mental workload as measured by physiological indicators and longitudinal driving behavior. In addition, the results indicated that current car-following models did not adequately describe driving behavior in case of incidents in the other driving lane.

Modeling Driver, Driver Support, and Cooperative Systems with Dynamic Optimal Control

Previous work proposed an optimal control framework for modeling driver behavior. Drivers were assumed to minimize the predicted subjective effort of their control actions, taking into account the anticipated actions of other drivers. The framework was generic. Several assumptions and simplifications had to be made; this factor hampered the applicability of the framework. One of these assumptions was that the behavior of other vehicles in the flow was stationary during the prediction horizon. Furthermore, the resulting model was computationally complex. A new approach based on the generic optimal control framework is proposed for modeling and computing driving behavior. The model can deal with the dynamics of the vehicles to which a driver reacts. At the same time, the computational complexity is small and does not increase exponentially with the complexity of the prediction model or with the size of the control vector. The mathematical solution approach is presented and illustrated with several examples. Face validity of the model is shown, and an application of the theory in the field of automated vehicle guidance is discussed. In particular for these applications, the proposed optimization approach allows for the computation of cooperative driving strategies that minimize a generic range of objective functions. The improvements in performance made by cooperation are substantial, as illustrated by several examples.

Longitudinal driving behavior under adverse weather conditions: adaptation effects, model performance and freeway capacity in case of fog

Adverse weather conditions have been shown to have a substantial impact on traffic flow operations. It is however unclear which adaptation effects in actual longitudinal driving behavior underlie this impact, how these adaptation effects relate to freeway capacity as well as to what extent current mathematical models of car-following behavior are adequate in incorporating these adaptation effects. In this regard a driving simulator experiment with a repeated measures design was performed in order to examine the influence of fog on adaptation effects, freeway capacity and parameter value changes and model performance of the Helly model and Intelligent Driver Model. From the results followed that fog led to a decrease in speed as well as in acceleration. Furthermore a substantial increase in distance to the lead vehicle was observed. These effects were implemented and simulated in a traffic simulation model. A substantial reduction in freeway capacity was found. This stresses the need to possess models of driving behavior, which are adequate in describing and predicting these adaptation effects. From the estimation results of the Helly model and IDM using a calibration approach for joint estimation followed that sensitivity factors, maximum acceleration and deceleration decreased substantially after the start of the adverse weather condition. Parameters representing headway increased significantly. Furthermore it followed from the results that the estimated models decreased in performance after the start of the adverse weather conditions

Dynamic Maximum Speed Limits: Perception, Mental Workload, and Compliance

Application of dynamic maximum speed limits may lead to positive effects for the environment, safety, and traffic flows. However, the efficacy of this dynamic traffic management measure depends largely on the behavior of drivers (i.e., compliance). In this paper, it is conjectured that compliance does not depend solely on attitudes of drivers but also depends on drivers’ perceptions of the dynamic maximum speed limit signs and mental workload. It is assumed that characteristics of the dynamic maximum speed limit signs influence the perceptions of drivers as well as their mental workload. It is, however, not yet clear to what extent characteristics of the signs influence perception, mental workload, and compliance of drivers. Therefore, two driving simulator experiments were performed to investigate the influence of four factors on drivers’ perception, mental workload, and compliance. The factors studied were the signs’ content, implementation, location, and frequency. From the results, it followed that different effects of these factors could be observed. For example, it was observed that the frequency with which dynamic maximum speed limit signs were provided to the drivers had a significant influence on perception and compliance, although a significant effect on mental workload could not be established. The paper concludes with a discussion of results and recommendations for future research.

Assessment of Dynamic Speed Limits on Freeway A20 near Rotterdam, Netherlands

On Freeway A20 near Rotterdam, Netherlands, a trial with dynamic speed limits began on June 28, 2011. On a 4.2-km stretch, the speed limit increased from 80 to 100 km/h as soon as congestion appeared to set in and at night. The aim of dynamic speed limits was to improve traffic operations and to avoid deterioration in the local air quality. This paper presents an assessment of this trial with respect to traffic operations, air quality, noise level, and traffic safety. Traffic operations on A20 appeared to have improved significantly as a consequence of the dynamic speed limits, which produced a reduction in the number of lost vehicle hours by 600 (20%). This improvement was the result of a 4% increase in the free-flow capacity at the main bottleneck on the freeway stretch. The dynamic speed limits caused a change in driver behavior: the median lane was better occupied when flow increased. Air quality deteriorated slightly. The effects varied along the stretch with a maximum increase in nitrogen oxides and particulate matter10 emissions of 3.7% and 3.6%, respectively. However, the effects on the average concentration of nitrogen oxides per year were limited. The noise level appeared to increase slightly with 0.2 dB. This increase occurred mainly during the two peak periods. The indicators for traffic safety showed sometimes a (possibly) positive and sometimes a (possibly) negative effect. However, it was not likely that dynamic speed limits had a significant negative effect on traffic safety.

Adaptation Longitudinal Driving Behavior, Mental Workload, and Psycho-Spacing Models in Fog

Adverse weather conditions have a substantial effect on traffic flow. However, the adaptation effects in longitudinal driving behavior that underlie this impact are unclear, as are the determinants. A driving simulator experiment was performed with a repeated-measures design and 25 participants. The adaptation effects in actual longitudinal driving behavior and the physiological indicators of mental workload (i.e., heart rate and heart rate variability) were measured under two conditions: normal visibility and fog. Significant adaptation effects in longitudinal driving behavior and a significant increase in mental workload were observed. A new estimation method was used to investigate the extent to which fog influenced the position of so-called action points in the (Δv, s) plane of a psycho-spacing model, where Δv was relative speed and s was spacing. In addition, multivariate regression analysis was applied to investigate the extent to which an influence could be observed on acceleration and on jumps in acceleration at the action points. Large differences in the positions of action points in the (Δv, s) plane, acceleration, and jumps in acceleration were observed between conditions; therefore, car-following patterns closely resemble those predicted by psycho-spacing theory. However, a large degree of inter- and intradriver heterogeneity was observed, possibly caused by differences in mental workload within and between drivers. This heterogeneity indicates that the assumption of deterministic perceptual thresholds is unrealistic and necessitates the development of a data-driven stochastic model based on the principles of psycho-spacing models.