**3. Simulation**

The emerging AV will definitely change the travel demand; however, whether this change is positive or negative is still under research. To simulate the current realistic traffic conditions, traffic flow was generated based on the data measured by KIRA (Transportation Information System Database of Hungary). Based on historical information provided by this database, the volume on the main road was set to 1440 vehicles/h, and the volume on the ramp was set to 312 vehicles/h, with eight percent of them Heavy Goods Vehicles (HGV). As mentioned above, the HAV and CAV will be introduced to the road system gradually. Based on this view, 31 scenarios representing different vehicle model combinations were simulated. Scenarios 1–11 contain CAV and Human Drive Vehicles (HDV), the penetration rates of CAV ranging from 0% to 100% with 10% step. HDV are represented by the calibrated Wiedemann 99 model in Vissim. Similarly, scenarios 12–21 contain the HAV model and HDV, the penetration rates of HAV ranging from 0% to 100% with 10% steps. Scenario 22–31 are a mix of three vehicle models with 20% steps. Each simulation involves a 1 h period, from 480 s to 4080 s with the first 480 s as warm-up time.

The test scenario is the upstream part of the M86 motorway. This road is close to the town of Csorna in northwestern Hungary (Gy˝or-Moson-Sopron County, West Transdanubia region), connecting Szombathely with Gy˝or, towards Budapest. The M86 is part of the TEN-T network [31] and also part of Hungarian State Public Road Network. Currently, the M86 is only in service between Szombathely and Csorna, with plans to extend north and south. This road is constructed to support the development and testing of autonomous vehicles. Figure 9 illustrates the overall 3.4 km profile of the M86 where the four sections of the road are marked:


**Figure 9.** The upstream network of the M86 motorway.

In order to make the simulation scenario reproduce the real road conditions and environment, a digital twin-based M86 motorway is generated, including every detail of test environments at high accuracy. Ref. [32] introduced high precision mapping to build an ultra-high definition map based on the road geometry. Meanwhile, the M86 road network has been used in [33] as virtual test scenarios, and its accuracy can reach ±2 cm, this accuracy is enough to ensure the effectiveness of the test and verification. As consequence, our simulation results are more realistic.

#### **4. Result and Discussion**

Through the presented comprehensive simulation system, the operation process of different proportions of autonomous vehicle models were simulated. The simulated data over the whole network are collected every 60 s interval. As the most indicative and intuitive parameters for the network status, the average speed of the network, the total travel time, and the average delay are used to evaluate traffic efficiency. The average speed is calculated by dividing total distance vehicles traveled by total travel time. The average delay is calculated by dividing total delay by the number of vehicles in the network plus the number of vehicles that have arrived. This delay is obtained by subtracting the actual distance traveled in the time step and desired speed from the duration of the time step.

Table 1 shows the simulation results of the mix of three vehicle models. The negative impact of the introduction of SAE level 3+ AV on traffic efficiency is evident observed from the simulated data. Compared to CAV, this negative effect is worse when HAV isintroduced to the network. This can be explained by an over perfect Wiedemann 99 model and as, compared to the human drivers that may take aggressive driving behavior, SAE level 3+ AV will not take any risky behavior when changing lane. Furthermore, SAE level 3+ AV require much larger gaps to perform a lane change than human drivers, which causes congestion at the merging area.


**Table 1.** Traffic performance evaluation of the simulated network.

To intuitively present the relationship between the penetration rate of the three vehicle models and the average speed, Figure 10a was drawn in ternary plots. It graphically depicts the penetration rate of the three vehicle models from 0% to 100% as the three sides in an equilateral triangle. The color inside the triangle indicates the average speeds over the

network. At every point within the triangle, the ratio of each combination is inversely proportional to the distance from the corner. Combining Table 1 and Figure 10a, it can be known that in the mixed flow, 40% CAV and 60% HDV show the best traffic efficiency, with the highest travel speed and the shortest travel time. Although the introduction of SAE level 3+ AV has a negative impact on total travel time and average speed, average delays in mixed traffic flow are significantly reduced. Especially in 100% CAV scenarios, the average delay dropped from 7.32 s to 0.51 s. The standard deviation plot of average speed in Figure 10b demonstrated the huge advantage of mixed flow consisting of CAV and HDV on the traffic stability. A smaller standard deviation means that the speed measurements are closer to the mean speed, which represents that the vehicles on the network can travel at a relatively uniform speed.

**Figure 10.** (**a**) Average speed over the network with mixed traffic; (**b**) Standard deviation of average speed over the network with mixed traffic.

Figures 11 and 12 present the changes of average speed with simulation time for various penetration rates of HAV and CAV, respectively. Overall, the introduction of HAV or CAV individually will cause the speed drop. For mixed traffic flow of HAV and HDV, 30% HAV with 70% HDV can generally keep the average speed at 100 km/h. Congestion can be observed at the end of simulation on the 100% HAV scenario; the average speed drops down to 40 km/h. It is foreseeable that, as the simulation time increases, the network will be fully blocked. This phenomenon can be explained by the much larger gap required by the HAV than human drivers when changing lanes. In addition, due to comfort considerations, the maximum acceleration of HAV is smaller than that of HDV, which results in the fact that when the network is full of HAV, the traffic downstream of the bottleneck decreases, and the upstream situation deteriorates.

**Figure 11.** Average speed over the network for the various HAV penetration rates.

**Figure 12.** Average speed over the network for the various CAV penetration rates.

Although the introduction of CAV individually also has a negative impact on the road network performance, due to the connection and cooperation functionality of CAV, the distribution of average speed is more concentrated than that of HAV. In the later stages of the simulation, as the penetration rate of CAV in the network increases, they can make full use of their connectivity to travel with small gaps, change lane faster, and absorb shock wave. Focusing on speed change, we see that under a high CAV penetration rate, the speed of the network shows a continuous growth tendency.

## **5. Conclusions**

This paper demonstrates the potential effects of the introduction of HAV and CAV on a real-world network. A microscopic traffic simulation framework that integrates vehicle models with different automated driving functions was constructed. These functions were implemented as an external driver model in the microscopic traffic simulator PTV Vissim. The framework was tested in a detailed digital twin based on the M86 motorway located in the southwest of Hungary. A case study consisting of different scenarios was performed to declare the effects of various combinations of HDV, HAV, and CAV. The traffic demand was obtained from real traffic counts. The possible combinations in 10% and 20% steps of the variable penetration rates per vehicle model formed 31 simulations. Each simulation was performed within a 1 h time period. Simulation results indicate the introduction of HAV and CAV deteriorating network performance. HDV outperformed HAV and CAV because HDV may take aggressive driving behaviors and is able to function over the speed limit. This characteristic is magnified by the presence of the ramp in the network. Among multitude scenarios with mixed traffic flow, the combination of 60% CAV and 40% HDV possess the optimal traffic performance in terms of average speed, total travel time, and average delay.

Due to the connectivity between CAV, the uniformity of speed was better in scenarios with high CAV penetration rates, which led to the excellent driving stability and the inhibition of the formation of traffic oscillations. In addition, the high CAV penetration rates in the network result in a significant reduction in traffic delays.

**Author Contributions:** Conceptualization, X.F. and H.L.; methodology, H.L.; simulation, X.F.; validation, X.F. and H.L.; writing—original draft preparation, X.F. and H.L.; writing—review and editing, X.F., H.L., T.T. and A.E.; supervision, T.T. and A.E; project administration, T.T., A.E. and M.F. All authors have read and agreed to the published version of the manuscript.

**Funding:** The research was supported by the Hungarian Government and co-financed by the European Social Fund through the project "Talent managemen<sup>t</sup> in autonomous vehicle control technologies" (EFOP-3.6.3-VEKOP-16-2017-00001).

**Institutional Review Board Statement:** Not applicable. **Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.
