Impacts of Different Types of Automated Vehicles on Traffic Flow Characteristics and Emissions: A Microscopic Traffic Simulation of Different Freeway Segments

Abstract

Different types of automated vehicles (AVs) have emerged promptly in recent years, each of which might have different potential impacts on traffic flow and emissions. In this paper, the impacts of autonomous automated vehicles (AAVs) and cooperative automated vehicles (CAVs) on capacity, average traffic speed, average travel time per vehicle, and average delay per vehicle, as well as traffic emissions such as carbon dioxide (CO₂), nitrogen oxides (NO_x), and particulate matter (PM₁₀) have been investigated through a microsimulation study in VISSIM. Moreover, the moderating effects of different AV market penetration, and different freeway segments on AV’s impacts have been studied. The simulation results show that CAVs have a higher impact on capacity improvement regardless of the type of freeway segment. Compared to other scenarios, CAVs at 100% market penetration in basic freeway segments have a greater capacity improvement than AAVs. Furthermore, merging, diverging, and weaving segments showed a moderating effect on capacity improvements, particularly on CAVs’ impact, with merging and weaving having the highest moderating effect on CAVs’ capacity improvement potential. Taking average delay per vehicle, average traffic speed, and average travel time per vehicle into account, simulation results were diverse across the investigated scenarios. The emission estimation results show that 100% AAV scenarios had the best performance in emission reductions in basic freeway and merging sections, while other scenarios increased emissions in diverging and weaving sections.

1. Introduction

Automated vehicles (AVs) are vehicles proficient in recognizing the road environment and navigating with limited or no involvement from the human driver. The Society of Automotive Engineers (SAE) categorizes AVs into six levels of automation based on the allocation of the driving task between the human driver and the vehicle system. These levels are level 0 (no automation, conventional vehicles (CVs)), level 1 (driver assistance), level 2 (partial automation), level 3 (conditional automation), level 4 (high automation), and level 5 (full automation). In the first three levels, the human driver executes the main driving task. Conversely, for the next three levels, the vehicle system performs the main dynamic driving tasks [1,2]. Fully AVs can be clustered into two: autonomous and cooperative. Autonomous AVs (AAVs) use only their sensors as a source of information, whereas cooperative AVs (CAVs) contain all AAVs functions with vehicle-to-vehicle (V2V) and vehicle-to-infrastructure (V2I) communication functions that intensify the efficiency and reliability of operation [3].

AVs have emerged promptly and are beginning to receive remarkable attention from researchers, policymakers, and car manufacturers due to their anticipated positive impact on the mobility system. These vehicles promisingly bring positive change in terms of roadway performance and provide better mobility for underserved individuals such as the elderly population and people with disabilities, however, there are uncertainties related to AVs’ long-term effects [4]. It is important to note that the positive impact of AVs on traffic will be significant when the market penetration is getting higher. A study by Friedrich shows that, at 100% market share, AVs can improve the capacity of highway segments by 80% [5]. In addition, AVs could provide a more substantial enhancement in traffic performance in a congested traffic state than in low-density traffic [6,7]. A simulation-based study by Wang and Wang on a single-lane road revealed that AVs can improve traffic flow by reducing delay and speed deviation, and this will result in a better level of service [8].

A thought-provoking study by Talebpour and Mahmassani discovered that, in some circumstances, there is a possibility that AAVs could have better traffic flow stability than CAVs at the same market penetration [9]. Likewise, the studies illustrate that when more safety and comfort issues are taken into account, cautious driving behavior in AVs might have a detrimental effect on traffic performance [10,11,12]. Another finding in Germany describes that aggressive driving behavior can lessen delay considerably and improve the capacity of freeway infrastructure by 30%, whereas the cautious driving behavior of AVs has a deleterious effect on capacity [13,14]. Regarding the potential impacts of AVs on emissions, literature reports diverse findings. Some studies report that the introduction of AVs has the potential to reduce road transport emissions [15]. Wadud et al. argue that automation might reduce road transport emissions by nearly half–or nearly double them–depending on the dominant features of AVs [16]. Rafael et al. report that introduction of AVs might increase NO_x and CO₂ emissions [17]. This finding is further confirmed by Manjunatha et al. [18]. Their results show that increasing AAV penetration rates results in emissions reductions, while increasing CAV penetration rates results in higher emissions.

Literature mainly investigated the impact of AVs, in general, on traffic flow parameters [19,20,21,22,23] and traffic emissions [24,25,26]. There are a few studies in the literature that attempt to investigate the effects of different types of automated vehicles, rather than AVs in general, on capacity [27], traffic flow [28], traffic speed, average trip time [29], and traffic emissions [30]. Nevertheless, to the best of the authors’ knowledge, the investigations of diverse types of AVs on several types of freeway segments are scarce. The current study, however, attempts to investigate not just the impacts of different types of automated vehicles, taking different market penetration of them into account, on capacity, traffic speed, travel time, delay, and traffic emissions (CO₂, NO_x, and PM₁₀), but also studies the impact in different freeway segments to see how much this effect could be moderated by different freeway segments. As a result, the purpose of this paper is to investigate the following through an experimental analysis using microscopic simulation with VISSIM software coupled with EnViVer:

• Quantify the impacts of different types of AVs on capacity, traffic speed, travel time, and delay, as well as traffic emissions (CO₂, NO_x, and PM₁₀) using microscopic simulation.
• Investigate the moderating effects of different AVs’ market penetration on their impacts.
• Investigate the moderating effects of different freeway’s segments on AVs’ impacts (basic freeway, merging, diverging, and weaving).

The rest of the paper is structured as follows. In Section 2, the applied methodology containing the investigated scenarios, software calibration, finding the capacity condition in each scenario, traffic flow analysis, and emission estimation is described. The results obtained are analyzed and compared with the base scenario (100% conventional vehicles) in Section 3. Section 4 provides the conclusions of this research and recommendations for further study are outlined in Section 5.

2. Methodology

The study follows the steps shown in Figure 1 to assess the impacts of different types of automated vehicles (AVs) on capacity, speed, travel time, and delay, as well as three traffic emissions such as CO₂, NO_x, and PM₁₀, taking the moderating effects of different AVs’ market penetration and different freeway segments on their impacts into account.

2.1. Investigated Scenarios

An experimental simulation study was conducted using VISSIM to assess the impacts of different types of AVs on capacity, speed, travel time, and delay. To this end, simulations contained two different types of AVs: autonomous AVs (AAVs) and cooperative AVs (CAVs), along with conventional vehicles (CVs). Furthermore, to investigate the moderating effects of different AVs’ market penetration on their impacts, the following market penetrations were studied in simulations: 100% CV (base scenario), 50% CV-50% AAV, 50% CV-50% CAV, 100% AAV, 100% CAV, and 50% AAV-50% CAV. Moreover, to investigate the moderating effects of different freeway segments on AVs’ impacts, the following freeway segments were studied in simulations: the basic freeway segment, merging segment, diverging segment, and weaving segment.

Figure 2 illustrates the geometric configuration of the studied freeway segments. The traffic volume from the on-ramp has been assumed to be 15% of the main lane. Likewise, for the diverging segment, the traffic volume on the off-ramp is 15% of the main lane. In the case of the weaving segment, a symmetrical weaving segment with one lane on-ramp and one lane off-ramp has been considered. The volume of traffic from the on-ramp is considered as 15% of the main lane, and 15% of the traffic volume from the on-ramp leaves the freeway via an off-ramp. Moreover, the percentage of traffic volume that moves from freeway to off-ramp is 15% of the main lane.

2.2. Adapting AVs’ Driving Behavior in VISSIM

It is of immense importance to adapt the driving behavior of AAVs and CAVs in VISSIM to be able to assess the impacts of these types of automated vehicles on road traffic flow parameters since they behave differently in road traffic flow compared to CVs.

There are two main streams of studies in literature that simulated driving behaviors (e.g., car-following, lane-changing, etc.) of AVs/CAVs in VISSIM microsimulation software following either internal or external modeling approaches. The first stream (e.g., see [31,32,33,34,35]), following the internal modeling approach, adjusts the parameters of the VISSIM’s default car-following (i.e., Wiedemann 99 algorithm) and lane-changing (i.e., a rule-based algorithm) algorithms to mimic the anticipated driving behaviors of interest. The second stream, following the external modeling approach, simulates driving behavior of AVs/CAVs through external VISSIM interfaces (e.g., Component Object Model Application Programming Interface (COM API), and External Driver Model (EDM) [36]) and user defined algorithm and code development. The COM API enables changes in vehicle movements and driving behaviors and can be developed in several programming languages (e.g., C# [37], Python [38]). The EDM is compiled in C++ and replaces the default driving behavior model in VISSIM (e.g., see [39,40]) such that the information on all vehicles in the network will be collected and sent to a dynamic link library to determine the behavior of the vehicles based on specifications of the custom/user-defined algorithm.

Furthermore, the external modeling approach provides the opportunity of testing various control strategies (e.g., freeway ramp control [41], intersection traffic control [42], variable speed limits control [43], etc.) in the AV environment through setting a specific control problem (e.g., minimizing travel time [43], delay [42], etc.) embedded in a relevant control system architecture (e.g., a model predictive control architecture [44]). Thus, the most efficient input signals (e.g., vehicle acceleration [41]) will be computed for the AVs in each time step through the control problem and sent back to VISSIM via the external modules.

Nevertheless, for AVs’ impact assessment, both internal and external modelling approaches are used in the literature to adapt AVs’ driving behavior in microsimulation environments. For instance, studies by Virdi et al. [40] and Tibljaš et al. [35] both use VISSIM microsimulation and a surrogate safety assessment module for AV’s safety impact assessment. However, the former followed the external—through an external control protocol called VCCP algorithm—while the latter followed the internal—through adjusting VISSIM’s default driving behavior model—modelling approaches to adapt AVs’ driving behavior in a microsimulation environment.

In the current study, we followed the internal modeling approach by adapting the VISSIM’s default car-following (i.e., Wiedemann 99 algorithm), lane-changing and look-setting parameters, considering a comprehensive review of the past studies that followed the internal modeling approach to simulate the driving behavior of AVs/CAVs in VISSIM.

Table 1. Calibrated simulation parameters in VISSIM software.

Calibrated Parameters	Basic Freeway/Merging/Diverging/Weaving			References
	CV	AAV	CAV
CC0: Standstill distance (m)	1.50/2.50/2.50/3.00	1.00/2.50/2.50/3.00	1.00/2.50/2.50/3.00	[31,32,45,46,47]
CC1: Headway time (s)	0.90/1.25/1.25/1.65	0.50	0.30	[6,45,46,47]
CC2: ‘Following’ variation (m)	4.00	2.00	1.00	[33,46,47]
CC3: Threshold for entering ‘following’ (-)	−8.00	−12.00	−16.00	[33,34,46,47]
CC4: Negative ‘following’ threshold (-)	−0.35	−0.35	−0.35	[34,46,47]
CC5: Positive ‘following’ threshold (-)	0.35	0.35	0.35	[34,46,47]
CC6: Speed dependency of oscillation (-)	11.44	0.00	0.00	[35,46,47]
CC7: Oscillation acceleration (m/s2)	0.25	0.40	0.40	[10,46,48]
CC8: Standstill acceleration (m/s2)	3.50	3.50	4.00/3.50/3.50/4.00	[31,34,45,47]
CC9: Acceleration at 80 km/h (m/s2)	1.50	2.00/1.50/1.50/2.00	2.00/1.50/1.50/2.00	[31,46,47,48]
Number of Observed vehicles (-)	2.00/2.00/2.00/4.00	8.00/8.00/8.00/10.00	10.00	[45,46,47]
Minimum look ahead distance (m)	0.00	150.00	5000.00	[46,47]
Maximum look ahead distance (m)	250.00	250.00	5000.00	[46,47]
Minimum look back distance (m)	0.00	150.00	5000.00	[46,47]
Maximum look back distance (m)	150.00	250.00	5000.00	[46,47]
Min. headway (front/rear) (m)	0.50	0.38	0.20	[10,31,34]
Safety distance reduction factor (-)	0.60/0.80/0.80/0.30	0.45/0.45/0.45/0.30	0.30	[10,34,45,46]
Maximum deceleration for cooperative braking (m/s2)	−3.00/−6.00/−9.00/−9.00	−4.00/−6.00/−9.00/−9.00	−4.00/−6.00/−9.00/−9.00	[31,34,45,46]
Overtake reduced speed area (-)	Unchecked	Unchecked	Checked	[46] and own suggestion
Desired speed distribution (deviation from the speed limit (km/h))	The default ‘S’ shaped distribution	±2.00	±2.00	[6,34]
Advanced merging (-)	Unchecked	Checked	Checked	[6]
Cooperative lane change: maximum speed difference (km/h) and maximum collision time (s)	Unchecked	Checked:10.80 and 10.00	Checked: 3.00 and 10.00	[6,34,46]

shows the adapted parameters from the literature. It is worth highlighting that CC0 refers to the desired distance between two stationary vehicles, CC1 refers to the time the driver wants to maintain a safe distance to the preceding vehicle (higher values represent more cautious drivers), CC2 restrains the longitudinal oscillation of a vehicle in relation to the vehicle in front, CC3 defines at what time the deceleration process will begin in terms of seconds before reaching the safety distance, CC4 and CC5 regulate the speed differences during the following process, CC6 refers to the impact of distance on speed oscillation within the following process, CC7 defines the actual acceleration during the oscillation process, CC8 refers to the desired acceleration when starting from a stationary state, and CC9 refers to the desired acceleration at a speed of 80 km/h. In addition, look ahead/back distances define the distances that a driver can see ahead/behind of his vehicle and still be able to react to actions made by surrounding drivers, observed vehicles controls the driver’s ability to predict other vehicle’s actions and respond to them (the higher the value the more vehicles can be observed), Min. headway (front/rear) refers to the minimum remaining distance required between two vehicles after a lane change, safety distance reduction factor determines how much the safety distance between vehicles should be reduced during the lane change, Maximum deceleration for cooperative braking defines the maximum deceleration that the trailing vehicle driver will accept for cooperation to help the leading vehicle to perform its maneuver, overtake reduced speed area specifies a direct free lane change upstream of a reduced speed area; thus if unchecked, vehicles will not perform a lane change upstream a reduced speed area, advance merging specifies any necessary lane change towards the next connector along the route; thus if checked, more vehicles can change lane at an earlier point, and cooperative lane change makes it possible for a vehicle to observe if a vehicle on an adjacent lane intends to change to its own lane, and hence will try to change lane itself to accommodate the lane change.

Furthermore, the driving behavior for CVs corresponding to each freeway segment has been modified accordingly [45]. A speed limit of 130 km/h is used on most of the freeways in Europe in general and Hungary in particular. Consequently, a 130 km/h speed limit is assumed in this study. In fully AAV and CAV driving modes, the vehicle system will not override the specified speed limit. Therefore, a speed deviation of ±2 km/h from the limit is assumed for both types of AVs [6], while for CVs the VISSIM default speed distribution has been considered.

2.3. Capacity Condition in Each Scenario

For the current study, the capacity condition in each of the specified scenarios has been estimated for the following reasons: on the one hand, to assess the impacts of different types of AVs on capacity; and on the other hand, to study the moderating effects of different AVs’ market penetration as well as different freeway segments on AVs’ impacts on capacity.

To assess the impacts of AVs on speed, travel time, and delay as well as traffic emissions in each of the specified scenarios, it was essential to have a comparison platform with a similar scale throughout the scenarios to give the opportunity to compare the values in each scenario with the ones in a base scenario, that is, 100% CV. To do so, the changes in speed, travel time, and delay, along with CO₂, NO_x, and PM₁₀ in the capacity condition of each of the specified scenarios were compared with the ones in the capacity condition of the base scenario to reveal the maximum impacts of different types of AVs on traffic flow parameters and traffic emissions.

To estimate the capacity condition in each scenario, the inflow was increased from near-zero volume to higher volumes during the successive simulation runs, and the throughput of the system (outflow) was carefully studied. At lower inflows, the outflow volume increases as vehicle input (inflow) increases. However, when the maximum throughput is attained, a higher vehicle input will not result in the same increment in the outflow volume. A decrement in outflow, despite the increase in vehicle input in consecutive simulations, shows that the freeway segment reaches its capacity, which is measured in vehicles per hour (vph).

Figure 3 illustrates the obtained capacity condition in each scenario for different road segments. At capacity condition, for each scenario, the average traffic flow characteristics (average delay per vehicle (s/veh), average traffic speed (km/h), and average travel time per vehicle (s/veh)) have been estimated. Correspondingly, the average traffic emissions in grams per hour per vehicle (g/h/veh) for CO₂, NO_x, and PM₁₀ have been thoroughly investigated.

It is worth highlighting that the simulation period in VISSIM for each inflow level was adjusted to 4500 s, including the 900 s warming time and 3600 s during which simulation results are collected for analysis. Moreover, five simulation runs (trials) with a random seed increment of one were applied, and the average value of the results (outflows) in these five runs was considered for each inflow level. In total, for all of the specified scenarios, 1474 inflow (vehicle input) levels were studied, which resulted in 7370 (1474 × 5) simulation runs.

2.4. Traffic Emission Estimation Using EnViVer

EnViVer is a VERSIT+ based add-on module that can estimate traffic emissions from VISSIM traffic simulation models. Using EnViVer, three types of regulated emissions can be determined: particulate matter (PM₁₀), nitrogen oxides (NO_x), and carbon dioxide (CO₂) for different vehicle types and circumstances [49]. Figure 4 illustrates vehicle category, fuel type, and vehicle age distribution according to the Hungarian vehicle fleet that is considered in this research. The results of the traffic simulation from VISSIM were imported into EnViVer software to calculate the average vehicle exhaust emissions per hour in the form of CO₂, NO_x, and PM₁₀. The batch calculation tool was used and the average value of five simulation trials at the capacity condition of each freeway segment was reported.

3. Results and Analysis

The simulation results provide an important insight into how AAVs and CAVs could influence the freeway traffic flow parameters as well as traffic emissions. Each of the presented results has been estimated by averaging five simulation runs for the corresponding scenarios. The simulation results are presented separately for the basic freeway, merging, diverging, and weaving segments. More importantly, it should be noted that the traffic flow and traffic emission impacts of AVs on the considered freeway segments are due to the change in the driving logic of vehicle automation.

3.1. Basic Freeway Segment

The simulation results for the basic freeway segment are shown in

Table 2. Traffic flow characteristics and traffic emissions for basic freeway segment.

Scenario	Market Penetration	Average Road Traffic Flow Characteristics				Average Road Traffic Emissions
		Capacity (vph)	Delay (s/veh)	Speed (km/h)	Travel Time (s/veh)	CO2 (g/h/veh)	NOx (g/h/veh)	PM10 (g/h/veh)
1	100% CV (base scenario)	7293	7.51	98.39	36.86	234.204	0.719	0.047
2	50% CV-50% AAV	8283	4.50	109.07	33.12	226.240	0.690	0.046
3	50% CV-50% CAV	8464	10.04	93.26	38.69	317.214	1.067	0.058
4	100% AAV	11,312	0.08	129.06	27.99	216.841	0.627	0.044
5	100% CAV	13,797	3.93	115.57	31.76	272.297	0.879	0.051
6	50% AAV-50% CAV	12,614	0.60	126.97	28.48	221.978	0.648	0.044

. Taking the maximum throughput (capacity in vph) of the road segment into account, one can truly find that the introduction of automated vehicles has the potential to increase the capacity of the basic freeway segment. CAVs have a greater impact on capacity increment compared to AAVs, since V2V and V2I communications in CAVs have the potential to create a smooth traffic flow, less reaction time, and short headway. However, the average delay per vehicle (s/veh), average traffic speed (km/h), and average travel time per vehicle (s/veh) along with traffic emissions do not follow a similar pattern. The results revealed the fact that the impact of AAVs on the above-mentioned traffic flow parameters is greater than that of CAVs. Moreover, CAVs have the potential to improve the above-mentioned parameters in the condition that either they are mixed with other types of automated vehicles, here AAVs, or the vehicles in traffic flow are all CAVs. Nonetheless, the introduction of CAVs results in a higher impact on traffic emissions compared with AAVs.

To quantify the impacts of different types of AVs, the results are compared with the base scenario (100% CV), as shown in Figure 5. Taking a wide look at Figure 5 reveals the fact that a considerable improvement in capacity is achieved when the traffic flow is entirely automated (+89%, +73%, and +55% with 100% CAV, 50% AAV-50% CAV, and 100% AAV, respectively). Moreover, the condition in which conventional vehicles constitute 50% of the traffic flow, types of AVs, here AAVs and CAVs, will not have a significant effect on capacity improvement (2% increase from 50% CV-50% AAV to 50% CV-50% CAV scenario). This indicates that the benefits from the cooperative driving features of CAVs in the traffic flow are insignificant when their market penetration is low. Furthermore, taking the average traffic speed into account, the results showed the potential of an improvement of 31% with 100% AAVs, which would result in 99% and 24% decrement in average delay per vehicle and average travel time per vehicle, respectively.

Figure 6 displays the impacts of AAVs and CAVs on traffic emissions compared to the base scenario, which is 100% CV. The results show that 50% CV-50% CAV, and 100% CAV market penetrations increase all the considered traffic emissions. However, other market penetrations result in a decrement in traffic emissions. The highest reduction in emissions is realized in 100% AAV market penetration, which is a 7% reduction in CO₂ and PM₁₀ emissions and a 13% reduction in NO_x emissions. On the contrary, the highest rise in emissions was obtained at the 50% CV-50% CAV market penetration level. This resulted in a 35%, 48%, and 22% rise in CO₂, NOx, and PM₁₀ emissions, respectively.

3.2. Merging Segment

The simulation results for the merging segment are shown in

Table 3. Traffic flow characteristics and traffic emissions for merging segment.

Scenario	Market Penetration		Average Road Traffic Flow Characteristics			Average Road Traffic Emissions
		Capacity (vph)	Delay (s/veh)	Speed (km/h)	Travel Time (s/veh)	CO2 (g/h/veh)	NOx (g/h/veh)	PM10 (g/h/veh)
1	100% CV (base scenario)	4134	61.50	46.36	102.26	391.598	1.275	0.074
2	50% CV-50% AAV	4836	31.70	67.84	69.17	411.884	1.352	0.074
3	50% CV-50% CAV	4911	30.05	70.89	67.92	413.779	1.356	0.074
4	100% AAV	6208	1.87	123.49	38.09	287.055	0.842	0.058
5	100% CAV	6959	16.94	92.37	53.09	361.116	1.132	0.067
6	50% AAV-50% CAV	6758	22.91	81.82	58.77	441.528	1.474	0.077

. Taking the capacity of the road segment into account, the results illustrate an increasing pattern with the introduction of automated vehicles, with CAVs having a greater impact on this increment compared to AAVs due to their cooperative capability. However, simulation results clearly demonstrate the moderating effect of merging segments on capacity improvements, particularly on CAV impact (compare 100% CAV capacity changes in Figure 5 and Figure 7).

Furthermore, unlike the basic freeway segment, the introduction of AVs (both AAVs and CAVs) will decrease average delay and average travel time per vehicle. It can be pointed out that CAVs have a slightly greater impact in lower market penetration (50% CV-50% CAV compared to 50% CV-50% AAV) and AAVs have a greater impact in higher market penetration (100% AAV compared to 100% CAV). Moreover, potential improvements in average traffic speed due to the introduction of AVs are considerably higher compared to the basic freeway, with 100% AAV scenario achieving 166% increment (see Figure 7).

On the other hand, as demonstrated in Figure 8, the findings outline that 100% AAV and 100% CAV market penetration result in a decrement for all traffic emissions. The highest reduction in emissions results from 100% AAV market penetration that is 27%, 34%, 22% reductions for CO₂, NO_x and PM₁₀ emissions, respectively. On the contrary, 50% AAV-50% CAV market penetration results in an increment for all the three traffic emissions i.e., 13% in CO₂, 16% in NO_x, and 5% in PM₁₀. Furthermore, 50% CV-50% AAV and 50% CV-50% CAV scenarios have nearly the same effect on traffic emissions. Both scenarios result in a 6% rise in NO_x while a 5% and 6% rise in CO₂ have been recorded for the respective market shares. Regarding PM₁₀ emissions, 50% CV-50% AAV has no impact and 50% CV-50% CAV resulted in a 1% increment.

3.3. Diverging Segment

The simulation results for the diverging segment are shown in

Table 4. Traffic flow characteristics and traffic emissions for a diverging segment.

Scenario	Market Penetration	Average Traffic Flow Characteristics				Average Road Traffic Emissions
		Capacity (vph)	Delay (s/veh)	Speed (km/h)	Travel Time (s/veh)	CO2 (g/h/veh)	NOx (g/h/veh)	PM10 (g/h/veh)
1	100% CV (base scenario)	3500	86.25	37.02	127.46	447.735	1.331	0.081
2	50% CV-50% AAV	4311	72.99	41.82	112.37	471.771	1.451	0.082
3	50% CV-50% CAV	4519	73.10	41.75	112.45	472.037	1.426	0.081
4	100% AAV	5166	61.08	47.62	98.98	534.633	1.722	0.088
5	100% CAV	6050	56.05	50.18	93.60	506.472	1.625	0.085
6	50% AAV-50% CAV	5539	63.05	46.60	100.74	533.283	1.680	0.087

. The capacity improvement in the diverging segment follows the previous increasing pattern already discussed in the basic freeway and merging segment, with a 100% CAV scenario reaching the maximum capacity increment. Furthermore, it is worth highlighting the moderating effect of diverging segments on capacity improvements, particularly on CAVs’ impact (compare 100% CAV capacity changes in Figure 5 and Figure 7). Taking the average traffic speed, average travel time per vehicle, and average delay per vehicle into account, the introduction of AVs shows a potential for improving the aforementioned parameters, with CAVs having slightly better performance, unlike basic freeway and merging segments (compare 100% CAV and 100% AAV scenarios in Figure 5, Figure 7 and Figure 9). On the other hand, in contrast to the scenarios in basic freeway and merging segments, the introduction of AVs results in an increase in CO₂, NO_x, and PM₁₀ at all levels of market share compared to the base scenario.

To quantify the impacts of different types of AVs in the diverging segment, the results are compared to the base scenario (100% CV), as shown in Figure 9. Taking a wide look at Figure 9 reveals the fact that full automation in vehicles (CAVs) with 100% market penetration has the best performance with a potential to improve capacity, average traffic speed, average travel time per vehicle, and average delay per vehicle by +73%, +36%, −27%, and −35%, respectively.

Figure 10 presents the traffic emission results for each diverging segment. Compared to the base scenario, all the traffic emissions increased for all market penetration levels except for the 50% CV-50% CAV scenario, which has narrowly the same PM₁₀ emissions. Both the 50% CV-50% AAV and 50% CV-50% CAV scenarios have the same effect on CO₂ emissions (5% rise). In addition, these scenarios result in a 9% and 7% rise in NO_x emissions, respectively. Whenever the market penetration gets to 100% AV, it results in higher emissions than 50% AV scenarios. That is, 100% AAV results in an increment of CO₂, NOx, and PM₁₀ emissions of 19%, 29%, and 10%, respectively, and 100% CAV results in 13% for CO₂, 22% for NO_x, and 5% for PM₁₀ increment, while 50% AAV-50% CAV results in a rise of 19% for CO₂, 26% for NO_x, and 8% for PM₁₀, which is different from the basic freeway and merging freeway segments.

3.4. Weaving Segment

The simulation results for the weaving segment are shown in

Table 5. Traffic flow characteristics and traffic emissions for the weaving segment.

Scenario	Market Penetration	Traffic Flow Characteristics (Average)				Traffic Emissions (Average)
		Capacity (vph)	Delay (s/veh)	Speed (km/h)	Travel Time (s/veh)	CO2 (g/h/veh)	NOx (g/h/veh)	PM10 (g/h/veh)
1	100% CV (base scenario)	3558	73.04	34.59	104.36	296.242	0.825	0.055
2	50% CV-50% AAV	4387	43.85	49.68	73.21	338.506	1.073	0.059
3	50% CV-50% CAV	4499	51.70	45.44	81.62	337.657	1.020	0.059
4	100% AAV	5579	60.40	40.26	89.81	422.151	1.277	0.068
5	100% CAV	6013	67.53	37.21	97.03	433.380	1.277	0.070
6	50%AAV-50% CAV	5829	67.18	37.34	96.69	438.652	1.291	0.070

. Regarding the maximum throughput (capacity in vph) of the road segment, similar to the other investigated road segments, there is an increasing pattern associated with the introduction of automation in traffic flow, with a 100% CAV scenario reaching the maximum capacity (69% improvement in capacity, see Figure 11).

Moreover, the moderating effect of the weaving segment on CAVs’ capacity improvement potentials are observable (compare 100% CAV capacity changes in Figure 5 and Figure 11). However, unlike the merging and diverging segments, the moderating effect of the weaving segment on AAV capacity improvement potential is negligible (compare 100% AAV capacity changes in Figure 5 and Figure 11).

Furthermore, similar to the diverging segment and unlike the basic freeway segment and merging segment, half-automated market penetration scenarios had considerable impacts on capacity improvements (compare 50% CV-50% AAV and 50%CV-50% CAV scenarios in Figure 11, Figure 9, Figure 7, Figure 5, respectively). However, taking the average traffic speed, average travel time per vehicle, and average delay per vehicle into account, the weaving segment shows a totally different pattern compared to the previously investigated road segments, with half-automated market penetration scenarios (50% CV-50% AAV and 50% CV-50% CAV) having considerably better performance compared to fully automated traffic flow scenarios (100% AAV, 100% CAV, and 50% AAV-50% CAV).

Considering the traffic emissions, similarly, the impacts of AAVs and CAVs follow a different pattern from the other investigated freeway configurations (see Figure 6, Figure 8, Figure 10 and Figure 12).

It is worth mentioning that the heterogeneity of traffic (mixing AAVs with CAVs) results in an increase in CO₂, NO_x, and PM₁₀ emissions. Furthermore, 50% CV-50% AAV and 50% CV-50% CAV scenarios have the same impact on CO₂ emissions, i.e., both result in a 14% increment. However, 50% CV-50% AAV results in a higher increment of NO_x emissions than 50% CV-50% CAV, in which the first results in a 30% whilst the latter results in a 24% rise.

4. Conclusions

In this paper, the impacts of different types of automated vehicles (AVs) (autonomous automated vehicles (AAVs) and cooperative automated vehicles (CAVs)) on capacity, average traffic speed, average travel time per vehicle, and average delay per vehicle have been investigated through an experimental analysis using microscopic simulation. Moreover, the moderating effects of different AVs’ market penetration, as well as different freeway segments on AV impacts, have been studied.

The simulation results revealed the fact that the introduction of AVs has the potential to increase the capacity of the freeways, with CAVs having a greater impact on this increment compared to AAVs, regardless of the types of freeway segment. Furthermore, merging, diverging, and weaving segments showed a moderating effect on capacity improvements, particularly on CAVs’ impact, with merging and weaving having the highest moderating effect on CAVs’ capacity improvement potential. It is worth highlighting that, unlike the merging and diverging segments, the moderating effect of the weaving segment on AAVs’ capacity improvement potential was negligible. Furthermore, similarly to the diverging segment, and unlike the basic freeway segment and merging segment, half-automated market penetration scenarios (50% CV-50% AAV and 50% CV-50% CAV) had considerable impacts on capacity improvements in the weaving freeway segment.

Unlike the results in capacity, the results in average delay per vehicle, average traffic speed, and average travel time per vehicle were fairly diverse and did not follow a similar pattern in all of the investigated freeway segments. In the basic freeway segment, the impact of AAVs on the above-mentioned parameters was greater than that of CAVs, and CAVs had the potential to improve the above-mentioned parameters in the condition that either they are mixed with AAVs or vehicles in traffic flow are all CAV. In the merging segment, unlike the basic freeway segment, the introduction of AVs (both AAVs and CAVs) will decrease average delay and average travel time per vehicles, with CAVs having a slightly greater impact in lower market penetration (50% CV-50% CAV compared to 50% CV-50% AAV) and AAVs having a greater impact in higher market penetration (100% AAV compared to 100% CAV). Moreover, the potential improvement in average traffic speed due to the introduction of AVs was considerably higher in merging segment, compared to the basic freeway segment. In the diverging segment, the introduction of AVs shows potential in improving the aforementioned parameters, with CAVs having slightly better performance, unlike basic freeway and merging segments. In the weaving segment, unlike the other freeway segments, half-automated market penetration scenarios (50% CV-50% AAV and 50%CV-50% CAV) had considerably better performance compared to fully automated traffic flow scenarios (100% AAV, 100% CAV, and 50% AAV–50% CAV) considering the average delay per vehicle, average traffic speed, and average travel time per vehicle.

Considering the traffic emission impacts of AVs, higher market penetrations of AAVs and CAVs result in higher emissions for diverging and weaving segments compared with basic freeway and merging segments. Furthermore, at a 100% market rate, both AAVs and CAVs result in comparatively similar impacts on emissions in the weaving segment, but this is not the case in the other roadway segments. For instance, both the 100% AAV and 100% CAV scenarios result in a 55% rise in NO_x, while the corresponding PM₁₀ emissions are 25% and 28%, respectively, compared with the base scenario (100% CV). The highest reduction in emissions is achieved at 100% AAV market penetration for basic freeway and merging. On the contrary, for diverging and weaving segments, the traffic emissions increase for all market penetration of AVs compared with the base scenario; whenever the market penetration is 100% AV, it results in higher emissions. For the diverging segment, the 100% AAV scenario results in the highest rise in CO₂, NO_x, and PM₁₀ emissions. For the weaving segment, mixing AAVs with CAV (50%AAV-50%CAV scenario) results in the highest increment for both CO₂ and NO_x emissions, while 100% CAV and 50% AAV-50%CAV scenarios result in the highest PM₁₀ emissions.

5. Recommendations for Further Study

The present study is a first step in analyzing the potential impacts of AVs on traffic flow and traffic emissions in different freeway segments, and it has limitations that will be considered in future studies. The current study’s limitation is that only 0%, 50%, and 100% market penetration levels were considered in the simulation; however, further research into additional penetration levels will provide a more representative image of the real-world scenario. In addition, the results of the current research showed a considerable impact of different types of AVs on traffic speed, travel time, and delays, which would recommend extending the scope of the current study to environmental external costs associated with the presence of different types of automated vehicles, taking the moderating effects of different AVs’ market penetration and different freeway segments on AVs’ impacts into account. Beyond that, the focus of this study is confined to a simulation experiment on the presumptive car-following and lane-changing behavior of AVs, and the analysis did not consider the sensitivity of the results to each of the parameters. Future studies might thus include examining how each parameter affects emissions and traffic flow. More importantly, the driving behaviors of AVs should be updated in future studies with the advancement of automation technology.