Simulation Platform to Analyse Future Traffic Regulations for Automated Vehicles in a Mixed Traffic Environment

When starting the simulation platform, the user has the possibility to show the results of already simulated scenarios or to define a new simulation scenario. Based on the defined parameters a unique hash-code is created and allocated to each scenario, which allows to re-open already simulated scenarios and hinders users from simulating the same scenario twice. The scenario configuration is done in a level-based manner (see Fig. 1), where a user can adjust the simulation parameters. First, the user has to define, which highway segment (e.g. on-ramp, off-ramp, weaving segment, construction site, tunnel) to investigate. Depending on the chosen segment, varying infrastructure related parameters can be defined in the first level. For on-ramp segments this includes the length of the acceleration lane and the location of loop detectors. In a next step, the user may defines different AV driving regulations in the second level. These possible future AV driving regulations were elaborated based on an international workshop focusing on legal aspects from the perspective of increasing vehicle automation. The third level allows to parameterise weather circumstances (e.g. sunny, rainy, cloudy weather with bad visibility). In the forth level, the AV penetration rates can be set, the maximum allowed speeds needs to be selected in level five and the truck rate can be defined in level six. For on-ramp segments, the share between vehicles entering via the on-ramp and vehicles entering via the main carriageway can be chosen in level seven. Next, in level eight the traffic volume can be defined for every 5-min interval. As default, traffic volume is getting increased in a 5-min cycle. After 2/3 of the simulation time, traffic volume reaches the maximum level and is decreased again, which allows to investigate varying traffic states including the dissolution of congested traffic. In the last level simulation parameters, like the number of simulation runs and the simulation duration can be selected. All parameters are adjusted automatically within the simulation environment.

Three different levels of AV driving functions were developed and integrated into the simulation platform. A connected environment is assumed, which means that the AV driver models receive vehicle state information (speed, acceleration, etc.) of all surrounding vehicles, including obstructed vehicles. For AVs with Adaptive Cruise Control (ACC) only the longitudinal behaviour is executed by the AV driving function, lane changing behaviour is controlled by Vissims’ internal lane change model, which reflects human driver behaviour. Additionally, we introduce a Connected Highway Chauffeur (C-HC) and a Connected Highway Pilot (C-HP), for which both the longitudinal and lateral control is done by the AV driving function. However, the lateral control of the C-HC driving function is only active on the main road. Therefore, a C-HC equipped AV is not capable of performing mandatory lane changes (e.g. merging at on-ramp). The C-HP is active in all areas and is also able to plan and execute merging and diverging lane changes in on-ramp and off-ramp areas, respectively. For longitudinal control, an ACC-controller was implemented. For lane changing we developed a rule-based gap acceptance framework, which is shown in Hofinger et al. [6].

Since reliable results are crucial, human driver behaviour was calibrated extensively utilising varying data sources. Based on single vehicle data from multiple highway cross-sections we computed various desired speed distributions and time headway distributions to calibrate car-following behaviour. To also calibrate lane changing behaviour, trajectory data from aerial videos of drones was investigated. We developed a 3-step calibration process, that is described in detail in Hofinger et al. [7]. Any adaptions in human driver behaviour due to the existence of AVs were neglected, since there is only little data available.

After each simulation, the results files are processed automatically. Besides single vehicle cross-sectional data, all simulated vehicle trajectories are investigated. In the result section of the simulation platform, we first included overall simulation statistics with the most relevant indicators (e.g. travel time, delay, number of simulated vehicles, etc.). Next, the user can choose a cross-section for which the fundamental diagram will be displayed as shown in Fig. 1a. Given that the capacity was reached, a stochastic capacity estimation is computed based on the work from Geistefeldt et al. [8]. To investigate car-following properties (see Fig. 1b) speed distributions and time headway distributions are presented for a selected cross-section within the platform. Regarding lane changing, a gap acceptance distribution is computed and the reaction of approaching vehicles on the target lane due to lane changing vehicles can be investigated (see Fig. 1c). Last, the spatio-temporal speed profiles (see Fig. 1d) are displayed within the simulation platform. There, the user can select the simulation run for which the results should be shown. The spatio-temporal speed profiles allow the user to assess the overall impact on traffic flow and to investigate the spatio-temporal occurrence of congestion.