Journal of Transportation Engineering, Part A: Systems

The geometric design of a roundabout follows an iterative process that aims to achieve an optimal balance among safety provisions, operational performance, and large vehicle accommodation. The evaluation of safety involves checking the speed consistency along the vehicle paths, which requires identifying the trajectories of the fastest vehicles and the speeds at which they travel. A systematic procedure to identify the fastest vehicle paths for all possible movement types between the approaches of a roundabout using as input data the roundabout geometry and the minimum distance between the vehicle and the roundabout edge is proposed. The determination of the fastest path between an entry and an exit is based on the generation of multiple feasible paths and the final selection of the fastest one. The proposed procedure supports the safety assessment of a roundabout geometric design according to the most relevant international guidelines and can be easily implemented in a heuristic roundabout design procedure. The application of the method to a trial case is presented to demonstrate its performance and applicability.

Pedestrian crossing at undesignated midblock locations is a prevalent phenomenon in developing countries like India that adversely affects the traffic performance of such sections. The present study aims to determine the effect of such pedestrian crossings on traffic characteristics like vehicular speed and capacity of the section. Data was collected on six-lane urban arterial midblock sections in six different metropolitan cities in India. Fifteen such sections were selected, with and without the presence of crossing pedestrians, and named base and friction sections, respectively. Speed models were developed for the sections, and from the speed models, the capacity of base and friction sections was estimated with varying pedestrian crossflow. The model capacity values were further validated by collecting field data at six more sections—three base sections and three friction sections. The model capacity values were in good agreement with the field capacity values, with mean absolute percentage error (MAPE) values less than 6% for both data sets, that is, for base and friction sections, respectively.

A traffic volume which can trigger a breakdown event at one point in time may not trigger it at another time. The critical question is why a roadway under the same loading is stable in one instance and unstable in another. This paper explains this behavior by using a linear first-order stochastic differential equation (SDE) model, a geometric Brownian motion (gBm) model. This simple stochastic (time-dependent) model of diffusion treats traffic volume, the load on the roadway system, as a random process. The model response variables are (1) the breakdown probability, which is the transition from a free-flow to a congested state for a given traffic loading; and (2) traffic delay. There are two major challenges. The first is formulating an approach, i.e., a mathematical model, that can reliably forecast traffic breakdown at a data collection site located upstream of a bottleneck where no data are collected. The second is selecting and calibrating an appropriate gBm model with extremely volatile data. The approach was assessed by performing match tests, assessing the field data summaries against model forecasts of traffic volume, breakdown probability, and delay. The potential for using the gBm modeling approach as an operational analysis tool was discussed.

The location of a U-turn median opening has an important effect on traffic flow operations for a shared left-turn and U-turn (SLU) lane with traffic signals. It is the first thing to analyze the influences of the arrival modes of multitype vehicles on the traffic delay for an SLU lane. Two models are then formulated to calculate the loss time for the phase used by the concurrent left-turns and U-turns and that used by the concurrent through and U-turns. To seek the optimal location of a U-turn median opening, a single-objective optimization model is developed on the basis of the SLU lane capacity. Six experiments that consider three levels of traffic demand and whether the U-turning volume has an effect on the green splits and the optimal U-turn location are carried out. The results reveal that the optimal location of a U-turn median opening depends on traffic demand and composition, and such a location will approach the stop line if the arrival rate of multitype vehicles decreases or the proportion of U-turn vehicles increases.

This study develops a method of calibrating and validating the Wiedemann car-following model using vehicle trajectory data. Unlike sensitivity analysis and optimization, this method conforms to the assumptions of the original Wiedemann 99 model related to drivers’ car-following behavior. Eight calibration constants (CCs) of the model were estimated using the vehicle trajectory data from a section of the US-101 freeway in Los Angeles, California. $CC1$ (desired time gap from lead vehicle) and $CC2$ (maximum change in spacing) were determined from the observed maximum and minimum spacing between the lead and following vehicles with similar speeds. $CC4$ and $CC5$ (minimum relative velocity at which the driver starts decelerating and accelerating, respectively, with short spacing of the lead vehicle or so-called action points) and $CC6$ (effect of spacing on these action points) were determined using a segmented linear regression model. This model provided the estimated relative velocities at which the speed of a following vehicle changed in response to a lead vehicle using constant acceleration/deceleration. It was found that the absolute values of $CC4$ and $CC5$ were not the same, which indicates that drivers are more sensitive to lead vehicles in the closing process than the opening process. $CC7$ was calculated as the mean difference in constant accelerations of lead and following vehicles. $CC8$ was calculated as the mean acceleration of all vehicles 1 s after the vehicles increased from slow speeds ($<5.5\mathrm{km}/\mathrm{h}$). Moreover, $CC9$ was calculated as the mean acceleration for speeds between 79.5 and $80.5\mathrm{km}/\mathrm{h}$. The traffic simulation with the estimated CCs in this study better reflected the observed speed distributions and action points than simulations with CCs estimated in previous studies using the same trajectory data.