Effects on Trafﬁc Performance Due to Heterogeneity of Automated

Vehicles on Motorways: A Microscopic Simulation Study

Ivan Postigo

1,2 a

, Johan Olstam

1,2 b

and Clas Rydergren

2 c

Swedish National Road and Transport Research Institute (VTI), Link

oping, Sweden

Department of Science and Technology, Link

oping University, Norrk

oping, Sweden

Keywords:

Automated Vehicles, Automated Driving, Microscopic Simulation, Mixed Trafﬁc.

Abstract:

The introduction of automated vehicles (AVs) is commonly expected to improve different aspects of trans-

portation. A long transition period is expected until AVs become prevalent on roads. During this period,

different types of AVs with different driving logics will coexist along human driven vehicles. Using micro-

scopic trafﬁc simulation, this study investigates the range of potential impacts on trafﬁc performance in terms

of throughput and travel delays for different types of AVs and human driven vehicles on motorways. The

simulation experiment includes scenarios with combinations of three different driving logics for AVs together

with human driven vehicles at increasing penetration rates. The utilized AV driving logics represent the evo-

lution of AVs, they were deﬁned in the microscopic simulation tool Vissim and were created by modifying

and extending the human driver behaviour models. The results of the simulation experiment show a decrease

in vehicle throughput and signiﬁcant effects on delay times when AVs with a more cautious driving logic are

predominant. Overall, results show higher vehicle throughput and lower travel delays as AVs evolve to more

advanced driving logics.

1 INTRODUCTION

The introduction of automated vehicles (AVs) is com-

monly expected to improve different aspects of trans-

portation; reduce operational costs, improve safety,

ease congestion, decrease energy usage, increase driv-

ing comfort, among others. From a trafﬁc perfor-

mance perspective, AVs are usually expected to im-

prove trafﬁc by being able to keep smaller gaps be-

tween vehicles, by always complying with road regu-

lations, and by smaller variations compared to human

drivers both in the way they drive as well as on how

they react to their surroundings.

In order for AVs to be allowed in public roads,

they need to be proven safe. This requirement would

lead ﬁrst generations of AVs to focus on minimizing

risks, and to drive more cautiously than most human

drivers.

Road authorities are interested on how the deploy-

ment of AVs will affect trafﬁc conditions, in order to

avoid possible negative impacts and to take advan-

https://orcid.org/0000-0002-4745-4865

https://orcid.org/0000-0002-0336-6943

https://orcid.org/0000-0001-6405-5914

tage of possible beneﬁts. Motorways arguably present

less challenging conditions for automated driving to

be ﬁrst introduced, mainly due to the one-directional

trafﬁc ﬂow and separation from pedestrians and non-

motorized transport modes (ERTRAC, 2019). On mo-

torways, AVs would primarily have to deal with the

interaction with other vehicles and not with trafﬁc

lights, pedestrian crossings, nor would they have to

anticipate all potential circumstances of urban scenar-

ios.

To estimate the impact of AVs on motorways, mi-

croscopic trafﬁc simulation is a suitable tool since the

movements of all vehicles and the interaction between

them are simulated. However, models used in simu-

lation tools were originally developed to describe hu-

man driving behaviors, thus, in order to used them for

automated driving, the models require modiﬁcations,

extensions or replacement with new models. Sugges-

tions for approaches on how to extend current models

exist, but while large scale ﬁeld data remains unavail-

able to validate them, any approach will include large

uncertainties.

The presence of AVs is expected to gradually in-

crease over the next decades (Calvert et al., 2017; Mi-

lakis et al., 2017; Tillema et al., 2017; Litman, 2015).

142

Postigo, I., Olstam, J. and Rydergren, C.

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study.

DOI: 10.5220/0010450701420151

In Proceedings of the 7th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2021), pages 142-151

ISBN: 978-989-758-513-5

During this transition period, public acceptance and

AV-related technologies will evolve, allowing follow-

ing generations of AVs to gradually have less conser-

vative driving styles. It is reasonable to expect that

different generations of AVs, with different capabili-

ties and driving styles, will be present on roads along-

side human driven vehicles.

This study aims to investigate the range of poten-

tial impacts on motorway trafﬁc performance caused

by the simultaneous presence of different types of

AVs and human driven vehicles. This coexistence is

expected during the transition to a predominant pres-

ence of AVs. The impact on trafﬁc performance is

measured in terms of vehicle throughput and travel

delays.

A simulation experiment using the microscopic

trafﬁc simulator Vissim is presented. The simulation

experiment includes multiple scenarios, with increas-

ing market penetration rates of AVs and with differ-

ent mixes of three driving logics within the AV share.

The three AV driving logics represent different gener-

ations of AVs and the evolution of their driving capa-

bilities. They were developed within the H2020 Co-

EXist project (Coexist, 2020) and are implemented in

Vissim (Sukennik, 2018b).

The remaining of this article is structured as fol-

lows. Section 2 presents a background on which this

study is base and limitations of microscopic models

for modeling automated driving. The simulation ex-

periment is described in Section 3, a description of the

simulated motorway segment, a conceptual descrip-

tion of the different driving logics used for AVs and

details of the included scenarios. Section 4 presents

the results and the impact of AVs on trafﬁc perfor-

mance in terms of change in capacity by measuring

throughput and travel delays. Finally, Section 5 ends

the article with conclusions and need for future re-

search.

2 BACKGROUND

Some commercially available vehicles are currently

implemented with advanced driving assistance sys-

tems like adaptive cruise control (ACC) or lane cen-

tering systems which automate the driving experience

to a certain degree. Top global automakers claim to

have fully automated driving vehicles available for

consumers by the early 2020s. However, current

available automated driving systems are not yet ca-

pable of operating fully autonomously.

The society of automotive engineers (SAE), which

develops standards for various transport industries,

proposes six levels to describe driving automation

(SAE, 2018). SAE levels 0 to 2 describe driver sup-

port features that assist drivers to different extents

but require the driver to steer, brake or accelerate as

needed to maintain safety. The more advanced levels

(SAE levels 3 to 5) describe automated driving fea-

tures, from optional automated driving under speciﬁc

conditions to full automated driving on any condition.

First generations of AVs (SAE level 3) are ex-

pected to have automated driving as an option under

speciﬁc conditions with the driver still responsible for

the driving. While ﬁrst SAE level 3-4 vehicles could

be expected by early 2020s, some studies give a time

estimation of decades until AVs become a consider-

able share of the vehicles and automated driving be-

comes the norm.

In (Tillema et al., 2017) the penetration rate of

AVs is estimated by the different levels of automa-

tion, taking around 15 years for AVs to have percep-

tible effects on trafﬁc ﬂows and up to 50 years for

AVs to be a prevalent presence on roads. In (Milakis

et al., 2017) the transition path to a prevalent presence

of AVs depends on different factors from local gov-

ernmental policies to technology development. While

the path remains unclear, (Milakis et al., 2017) sets

horizons for the years 2030 and 2050 for transport

implications in the Netherlands caused by the wider

deployment of AVs. In (Calvert et al., 2017), based

on different sources, it is estimated that it will take

at least 15 years until 20% of the vehicles become

automated. Based on patterns it took previous vehi-

cle technologies to be deployed, cost of the technol-

ogy and sales projections, (Litman, 2015) estimates

a period of three to ﬁve decades until the majority of

trips are made by AVs. (Litman, 2015) presents es-

timations from 2013, and predict a small number of

AVs by the time this study is presented (2020). How-

ever, most advanced commercially available vehicles

are not yet capable of fully automated driving though

there are claims from vehicle manufacturers that it is

just around the corner. Overall, it is estimated that it

is a matter of decades until AVs become predominant

in roads.

Different trafﬁc simulation studies aiming to in-

vestigate the impacts caused by AVs can be found

in the literature. Effects caused by different vehi-

cle technologies on different road environments have

been investigated. The impact of AVs on urban en-

vironments has been investigated (Lu et al., 2019), in

which AVs were modelled by modifying parameters

in the car following model. Analysis on the effects of

ACC and cooperative adaptive cruise control (CACC)

are commonly found in the literature (Liu et al., 2018;

Yuan et al., 2009; Arem et al., 2007; Minderhoud and

Bovy, 1999), but analysis are limited to longitudinal

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study

143

control of the ACC or CACC and not to vehicles capa-

ble of overtaking or lane changes. The impact of AVs

in merging roadways is investigated in (Rios-Torres

and Malikopoulos, 2017). The inﬂuence of AVs on

ﬂow stability and throughput is studied in (Talebpour

and Mahmassani, 2016). The studies have shown that

AVs will affect trafﬁc ﬂows in different ways, both

improving or degrading trafﬁc performance. Usu-

ally, it is assumed that all AVs will drive according to

the same driving logic (e.g. (Olia et al., 2018)), and

their impacts are commonly estimated by including

scenarios with increasing market penetration rates of

AVs. Moreover, the approaches taken to model AVs

or automated driving in simulation investigations vary

from study to study.

Microscopic simulation models have been devel-

oped aiming to describe human driving behavior. To

model the driving logic of AVs, these models need to

be modiﬁed, extended or replaced. Modifying param-

eters of existing car following models or lane chang-

ing models is a fast and simple approach to model

AV driving logics. This could for example be modi-

ﬁcations on distances kept between vehicles, reaction

times, accepted distances for lane changing, acceler-

ation and speed parameters. If a speciﬁc feature of

AVs can not be modeled by changing parameter val-

ues, then the models could be extended or replaced.

New or extended models could aim to simulate vehi-

cles sensors, control algorithms or safety features.

Nonetheless, regardless of the approach taken to

model AVs, the calibration of the models is based on

the available data, which currently is limited to exist-

ing partly automated vehicles, and not based on future

automated driving. Therefore, investigations based

on trafﬁc simulation of AVs should consider a range

of possible driving logics for AVs (e.g (Olstam et al.,

2020; Mintsis et al., 2019)).

3 EXPERIMENT SETUP

The purpose of the simulation experiment done in this

study is to enhance the understanding of potential im-

pacts caused by AVs. The simulation experiment has

been delimited to a motorway environment due to the

less challenging conditions they present for ﬁrst auto-

mated driving systems to consider them within their

operational design domain. The results present the

range of effects on trafﬁc performance caused by dif-

ferent AV mixes, showing the impact of the coex-

istence of different types of AVs alongside human

driven vehicles.

The uncertainties related to the transition to full

vehicle automation and to the evolution of the capa-

bilities of AVs are addressed by including multiple

scenarios in the simulation experiment. The transi-

tion period to full vehicle automation is considered by

creating scenarios with increasing penetration rates

of AVs, that is, the proportion of AVs present in the

trafﬁc ﬂow. This represents the expected gradual in-

crease in the of number of AVs present in roads. Three

AV driving logics are used to represent the evolution

of AVs. These driving logics assume different lev-

els of cautiousness in their driving styles, recognizing

that as the technology that allows automated driving

advances, AVs will become more reliable in regards

to safety. However, the uncertainty does not relate

only on how the technologies will evolve but also on

how fast they will be implemented in vehicles. Thus,

scenarios with all possible combinations of the three

AV driving logics are included, each combination is

a speciﬁc AV mix. They represent both a faster and

a slower evolution of AVs as well as the coexistence

of different types of AVs. The results obtained regard-

ing the effects caused by this coexistence are the main

contribution of this simulation experiment.

The simulation experiment is conducted for a Ger-

man representative motorway stretch (Sonnleitner,

2018). The AVs in the simulation are assumed to have

perfect knowledge of the geometry of the motorway.

Some elements of the motorway, like the position of

lanes or trafﬁc signs, are perceived in reality through

sensors on the vehicle, which is assumed to be per-

fect in the simulation. As seen in Table 1, AVs with

different driving logics interact with different number

of vehicles or objects. The accuracy, latency and re-

liability of the perception of other vehicles or objects

depend on whether the perception is done through on

board sensors or through connectivity features. This

perception is assumed to be perfect in the simulation

and with zero latency, and how it is accomplished is

implicitly included in the driving logics.

Future motorways might support automated driv-

ing by including digital infrastructure which will pro-

vide static and dynamic information to the AVs. The

infrastructure support for automated driving (ISAD)

(Carreras et al., 2018) considered in the simulation

experiment is ISAD-level D or E, which corresponds

to a conventional infrastructure with little support for

automated driving. The vehicles in the simulation per-

ceive different number of leading or trailing vehicles,

as well as immediate vehicles in target lanes for lane

changing maneuvers, they don’t have additional in-

formation about the trafﬁc conditions elsewhere along

their routes. Since there is no support from the infras-

tructure included in the motorway model, all dynamic

driving tasks are handled by the driving logics.

VEHITS 2021 - 7th International Conference on Vehicle Technology and Intelligent Transport Systems

144

3.1 Modeling of Automated Vehicles

The three driving logics used to model AVs are called:

cautious, normal and all-knowing. These driving log-

ics were developed within the H2020 CoEXist project

(Coexist, 2020) and are implemented in Vissim. They

are not exact driving algorithms, instead they rep-

resent different levels of cautiousness based on the

possible behaviour of AVs and how they would re-

solve conﬂicts, e.g. reaction times, gap thresholds,

etc (Sukennik, 2018b). The logics are based on the

Wiedemann 99 car following model (PTV, 2020) with

adjusted parameters.

The cautious driving logic is the most conserva-

tive and aims to ensure not only that AVs don’t cause

any accidents, but also to establish conﬁdence on the

public about the safe operation of AVs. This forces

the vehicle to adopt larger distances to surrounding

vehicles. An enforced absolute braking distance fea-

ture is enabled (see Table 4). The absolute braking

distance is the distance a vehicle must keep in order

to brake safely and avoid a collision if a vehicle in

front comes to a sudden full stop. The calculation

of the absolute braking distance considers a vehicle’s

own speed, braking capacity and relative position to

the vehicle in front, it neglects the leading vehicle’s

speed and braking capacity, as it assumes that the ve-

hicle in front could suddenly ”turn into a brick wall”.

This distance is also required for the lane change ma-

neuver, which means that it will have less chance of

taking place, the distance must exist both to the lead-

ing and trailing vehicles in the target lane.

The normal driving logic was developed with the

intention of being similar to human drivers. In con-

trast to human driven vehicles however, AVs with

this driving logic are capable of shorter reaction times

and have more accurate measurements of distances to

other vehicles as well as relative speeds. They are re-

stricted by the range of their sensors to only perceive

vehicles in their proximity and surroundings, unlike

human drivers who are often aware of vehicles be-

yond their proximity.

The all-knowing driving logic intends to model the

most advanced AVs capable of keeping smaller gaps

for all maneuvers and also have shorter reaction times.

AVs with this driving logic are assumed to be con-

nected to the infrastructure and thus receive informa-

tion about vehicles and objects beyond their surround-

ings. This connectivity feature is implicitly included

in this driving logic, as shown in Table 1, all-knowing

AVs interact with 10 objects and 8 vehicles, compared

to only 2 objects and 1 vehicles for the AVs with the

other driving logics.

All implicit stochastics have been disabled for all

AVs, which makes them show less variations on their

driving compared to human drivers. Human driven

vehicles are modeled with a desired speed distribution

which allows them to travel at a much wider range

of speeds, even surpassing the speed limits. On their

side, AVs always comply with the speciﬁed speed

limits and are modeled with a desired speed distri-

bution with a range of only ±2 km/h from the speed

limit. The acceleration and deceleration functions are

also different for AVs, showing less variation than hu-

man driven vehicles. The values of the used parame-

ters are presented in Tables 1, 2, 3 and 4.

Table 1 shows the difference in perception mod-

eled for each driving logic. The values show that the

cautious driving logic has a shorter range for percep-

tion. And, as previously mentioned, only the more

advanced all-knowing driving logic is able to perceive

several more vehicles and objects. In Table 2, the ex-

act parameter values used for the Wiedemann 99 car

following model (PTV, 2020) are presented as they

appear in Vissim. These values show the difference

on cautiousness or aggressiveness between the driv-

ing logics. Table 3 shows the values for the param-

eters used to perform necessary lane changes. They

show that acceleration and deceleration is more re-

stricted in the cautious driving logic, as well as the

required distances to perform the maneuver. Lastly,

Table 4 shows the additional functionalities of Vis-

sim used for each driving logic. All values are based

on the recommendations from (Sukennik, 2018a) and

(Olstam et al., 2020).

3.2 Scenarios Setup

Figure 1 shows the model of the motorway used in

the simulation experiment. It is a two lane motorway

going in one direction from point A to point C with

a speed limit of 130 km/h, covering 1 km in length.

The model includes an off-ramp with one lane with a

length of 200 m, and a similar on-ramp. A ’warm-up’

section of 1 km in length was included before point

A which is not shown in the Figure. The main ﬂow

of vehicles will enter the network through point A.

A secondary ﬂow, entering the network through point

B is set to be 25% of the number of vehicles gener-

ated for the main ﬂow. Lastly, the number of vehicles

taking the off-ramp is 20% of the number of vehicles

passing through point A.

The heterogeneity of AV that may exist in fu-

ture motorways was included by creating different AV

mixes. Each unique AV mix, where each AV driving

logic has 0, 20, 40, 60 or 80 or 100% share of the total

number of AVs were created, leading to 21 possible

combinations. Three of these unique mixes contain a

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study

145

Table 1: Driving behaviour parameters for following in Vissim.

Driving Logic

Cautious Normal All-

knowing

Human

Driven

Look ahead distance

Minimum (m) 0 0 0 0

Maximum (m) 150 250 300 250

Number of interaction objects 2 2 10 2

Number of interaction vehicles 1 1 8 99

Look back distance

Minimum (m) 0 0 0 0

Maximum (m) 150 150 150 150

Table 2: Driving behaviour parameters for car following model in Vissim.

Driving Logic

Wiedemann 99 model Cautious Normal All-

knowing

Human

Driven

CC0 - stanstill distance (m) 1.5 1.5 1.0 1.5

CC1 - headway time (s) 1.5 0.9 0.7 1.05

CC2 - ’following’ variation (m) 0.0 0.0 0.0 4.0

CC3 - threshold for entering ’follow-

ing’ (s)

-10.0 -8.0 -6.0 -8.0

CC4 - negative ’following’ threshold

(m/s)

-0.1 -0.1 -0.1 -0.3

CC5 - positive ’following’ threshold

(m/s)

0.1 0.1 0.1 0.35

CC6 - speed dependency of oscillation

(10

−4

rad/s)

0.0 0.0 0.0 11.44

CC7 - oscillation acceleration (m/s

) 0.1 0.1 0.1 0.25

CC8 - standstill acceleration (m/s

) 3.0 3.5 4.0 3.5

CC9 - acceleration with 80 km/h (m/s

) 1.2 1.5 2.0 1.5

Table 3: Driving behaviour parameters for lane change in Vissim.

Driving Logic

Cautious Normal All-

knowing

Human

Driven

Maximum deceleration (m/s

)

Own -3.5 -4.0 -4.0 -4.0

Trailing vehicle -2.5 -3.0 -4.0 -3.0

-1 m/s

per distance (m)

Own 80 100 100 300

Trailing vehicle 80 100 100 200

Accepted deceleration (m/s

)

Own -1.0 -1.0 -1.0 -1.0

Trailing vehicle -1.0 -1.0 -1.5 -0.75

Minimum headway (front/rear) (m) 1.0 0.5 0.5 0.5

Safety distance reduction factor 1.0 0.6 0.75 0.6

Maximum deceleration for cooperative

breaking (m/s

)

-2.5 -3.0 -6.0 -3.0

VEHITS 2021 - 7th International Conference on Vehicle Technology and Intelligent Transport Systems

146

Table 4: Driving behaviour functionalities in Vissim.

Driving Logic

Functionality Cautious Normal All-knowing Human

Driven

Enforce absolute braking distance On Off Off Off

Advanced merging On On On On

Cooperative lane change Off On On On

Figure 1: Motorway model. Flows are generated at points

A and B.

single type of AV. Twelve mixes contain two types

of AV, with one type being 60 or 80% the total num-

ber of AVs. The remaining 6 mixes include all three

types of AVs, where each type is 40 or 60% the total

number of AVs.

The gradual increase of AVs in roads is consid-

ered by including six different AV penetration rates

of 0, 20, 40, 60, 80 and 100%. At 0% penetration rate

there are only human driven vehicles, while at 100%

penetration rate, there are only AVs on the network.

Seven different demand levels for the main ﬂow of

1000, 2000, 3000, 4000, 5000, 6000 and 7000 vehi-

cles per hour (veh/h) were generated before point A.

For the secondary ﬂow, generated at point B, the cor-

responding demands are 250, 500, 750, 1000, 1500

and 1750 veh/h.

Each scenario consists of a speciﬁc combination

of demand level, AV penetration rate and AV mix,

giving a total of 742 scenarios, seven of which don’t

include any AVs (0% penetration rate). A simulation

time of one hour plus 30 minutes of ’warm up’ time

until the network reached a stable state was run for

each scenario. The results for are based on 10 repli-

cations with varying random seeds.

Additional scenarios were added to see the impact

of the enforced absolute braking distance feature in

the cautious driving logic. The feature was disabled

and the results compared to the scenarios with a single

AV driving logic.

4 RESULTS AND ANALYSIS

The results of the simulation experiment are presented

in terms of vehicle throughput and travel delay. The

lines shown in the ﬁgures are color-coded for each AV

mix, the legend shows thick solid lines indicating the

AV mixes that include a single type of AV. The col-

ors used are cyan for cautious AVs, orange for nor-

mal AVs, green for all-knowing AVs and gray for hu-

man driven vehicles. The remaining 18 lines are not

shown in the legend to avoid cluttering. The color of

each line represents the type of AV is most present in

the mix, while the color of the marker represents the

second most present type of AV. The line is shown

without a maker if there is an equal share between the

remaining two types of AVs. The 95% conﬁdence in-

terval shadows each line in Figures 5 to 8.

Figure 2: Throughput measured at point C for different de-

mands levels generated on the main ﬂow at point A for sce-

narios with a single type of vehicle.

Figure 2 shows the measured throughput at point

C. The Figure shows that each case reaches a max-

imum throughput at different demand levels. The

highest vehicle throughput occurs for the 100% all-

knowing AV mix, almost doubling the maximum

throughput measured for the 100% cautious AV mix.

Every scenario reaches a maximum throughput when

demands on the main ﬂow are higher than 5000 veh/h.

Thus, the effects on vehicle throughput caused by

the different AV mixes at different penetration rates

are more interesting for scenarios with high demands.

The results presented in Figure 3 correspond to a de-

mand on the main ﬂow of 6000 veh/h. Figure 3

shows that the highest vehicle throughput always oc-

curs for the ’100% all-knowing’ AV mix, while the

lowest vehicle throughput occurs for the ’100% cau-

tious’ AV mix. The measured throughput of every

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study

147

Figure 3: Throughput measured at point C for all 21 AV

mixes at different AV penetration rates for a demand on the

main ﬂow of 6000 veh/h.

other AV mix is found within this range. By closer

inspection of each AV mix, the throughput increases

when the normal, and all-knowing driving logics are

more present in the ﬂow. Similarly, when more AVs

with the cautious driving logic are present, the mea-

sured vehicle throughput decreases even compared to

the case when there are only human driven vehicles

(i.e. 0% AV penetration rate).

Back in Figure 2, it is shown that in the ab-

sence of AVs (100% human driven vehicles), the max-

imum vehicle throughput is observed when the de-

mand on the main ﬂow is above 4000 veh/h. Figure

4 shows the measured throughput for a lower demand

of 3000 veh/h. At this lower demand, effects on ve-

hicle throughput is noticed only at high AV penetra-

tion rates, and occurs for the AV mixes composed by

mostly or by only cautious AVs. Though little im-

pact occurs on vehicle throughput, Figures 5 and 6

show more noticeable impacts on average travel de-

Figure 4: Throughput measured at point C for all 21 AV

mixes at different AV penetration rates for a demand on the

main ﬂow of 3000 veh/h. The lines are mostly overlapping.

lays even at lower demands, both on the main ﬂow, as

well as on the secondary ﬂow.

Figures 5 and 6 show results in terms of aver-

age travel delays for vehicles travelling from point

A to C (the main ﬂow) and from point B to point C

(the secondary ﬂow) respectively. Different demand

levels for scenarios with a single type of vehicle, i.e.

only human driven vehicles and only each type of AV

are presented. The Figures show that compared to the

main ﬂow, the secondary ﬂow shows higher delays for

the AV mix of 100% cautious as demand increases.

The larger gaps required by cautious AVs might be

preventing them to merge into the main ﬂow, causing

the larger increase on delays seen on the secondary

ﬂow.

A more detailed investigation on queue forma-

tion would be required to understand the large in-

Figure 5: Average travel delays for vehicles going from

point A to point C for different demands levels generated

on the main ﬂow at point A for scenarios with a single type

of driving logic.

Figure 6: Average travel delays for vehicles going from

point B to point C for different demands levels generated

on the secondary ﬂow at point B for scenarios with a single

type of driving logic.

VEHITS 2021 - 7th International Conference on Vehicle Technology and Intelligent Transport Systems

148

Figure 7: Average travel delays for vehicles going from

point A to point C for all 21 AV mixes at different AV pen-

etration rates for a demand on the main ﬂow of 3000 veh/h.

Figure 8: Average travel delays for vehicles going from

point B to point C for all 21 AV mixes at different AV pen-

etration rates for a demand on the secondary ﬂow of 750

veh/h.

crease in delays seen in the secondary ﬂow as de-

mand increases for the AV mix of 100% cautious.

Based on observations, our hypothesis is that once the

main ﬂow reaches its maximum throughput demands

higher than 2000 veh/h, the required absolute braking

distance prevents most cautious AVs coming from the

secondary ﬂow to merge into the main ﬂow. Simi-

larly, if a cautious AVs comes from the left lane on

the main ﬂow, it might experience problems to switch

to the right lane to reach the off ramp if its route is to

leave the motorway. This could explain the observed

ﬂattening on the average delay curve of the main ﬂow

(A to C) for the AV mix of 100% cautious as demand

increases, shown in Figure 5, as well as the rapid in-

crease of average delays for the same AV mix on the

secondary ﬂow (B to C), shown in Figure 6.

Figures 7 and 8 show the impact of average travel

delays for a demand of 3000 veh/h on the main ﬂow

Figure 9: Throughput measured at point C for AV mixes

with a single AV logic at different AV penetration rates for a

demand on the main ﬂow of 6000 veh/h. An additional AV

cautious driving logic with the enforced absolute braking

distance feature disabled is included.

(A to C, Figure 7) and a corresponding demand of 750

veh/h on the secondary ﬂow (B to C, Figure 8) for dif-

ferent AV penetration rates and for all AV mixes. Sim-

ilar to the vehicle throughput, the minimum and max-

imum average travel delays occur for the AV mixes of

100% cautious and 100% all-knowing. The observed

travel delays of every other AV mix are found within

this range. The average travel delays are higher when

the more cautious AVs are more present in the ﬂow.

The results obtained for the cautious driving logic

could relate to the enforced absolute braking distance

feature. To investigate the impact the feature has in

terms of vehicle throughput scenarios were run with

the enforced absolute braking distance feature dis-

abled, the results are shown in Figure 9. Even though

the measured throughput still decreases as the AV

penetration rate increases, the decrease is not as large

as when the feature is enabled.

The results obtained for the different mixes of

AVs show that effects are intuitive, meaning that the

higher share of less cautious AVs (i.e normal and all-

knowing driving logics), the higher throughput and

the lower average travel delays. Only as the share

of more cautious AVs increases (i.e cautious driv-

ing logic), there is a decrease in measured vehicle

throughput and an increase in average travel delays.

Lastly, the results show that impacts on vehi-

cle throughput are noticeable only at high demands,

while effects on travel delays are noticeable even at

lower demands.

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study

149

5 CONCLUSIONS

The simulation experiment shows that as automated

driving moves away from cautious driving styles, or

in other words, as the capabilities of AVs allow them

to keep shorter gaps and have faster reaction times,

an increase in the capacity of motorways can be ex-

pected. In the same way, if automated driving does

not evolve and stay as the cautious driving logic,

capacity on motorways could decrease. Moreover,

while the effect on vehicle throughput is unnoticeable

at lower demands, the effect on average travel delays

are noticeable at lower demands.

The results for scenarios with heterogeneous AV

mixes are within the ranges of AV mixes with a single

type of AVs. Thus, the conclusion is that there are no

unexpected effects caused by the interaction between

the different AVs types.

A desirable road-map for the introduction of AV

can be drawn from the results obtained, knowing that

an automated driving style as the cautious driving

logic could have negative impacts and that an evolu-

tion on the driving style of AVs should be encouraged.

The results found from the scenarios with a large

share of cautions AVs indicate that a deeper un-

derstanding of what occurs in the merging zones is

needed. Moreover, since ﬁrst AVs will most likely

keep large gaps as the cautious driving logic, a deeper

investigation is needed on how to prevent the disrup-

tion of the trafﬁc ﬂow that this cautious driving logic

is causing.

Disabling the enforced absolute braking distance

in the cautious driving logic shows that the high safety

requirements that the feature has is preventing vehi-

cles to merge and might be over cautious. Though

we don’t know the exact details on how the feature

is implemented in Vissim, a relative braking distance

feature would be more ﬂexible, and might accomplish

the same safety requirements without being over cau-

tious.

The simulations assumed AVs to have a perfect

perception, not affected by inaccuracies or latency in-

troduced by sensors or connectivity features. More-

over, the infrastructure considered corresponds to

ISAD-Levels D and E, which provides little support

for automated driving. Changes in the modeling of

perception and/or including digital support from the

infrastructure might allow to change the modeling ap-

proach of automated driving, causing to observe dif-

ferent effects in the trafﬁc ﬂow due to the presence of

AVs.

It is debated how the presence of AVs will af-

fect the driving behavior of human drivers (HF Auto,

2017). This discussion has been left out of the scope

of this study and the assumption made was that the

driving behavior of human driven vehicles will not

change. Future investigations could include changes

on driving behavior of human drivers caused by the

presence of AVs.

The focus of this study has been the effects of

AVs on motorways using Vissim. Further research

could focus on other road environments, using differ-

ent microscopic simulators with different approaches

on how to model automated driving.

Finally, vehicle throughput and average travel de-

lays, show larger differences in results between the

cautious and normal driving logics than between the

normal and all-knowing driving logics. This might

indicate that there is a large beneﬁt of evolving from

the cautious to the normal driving logic. If AVs are

not expected to have such a big leap in their driving

styles, an additional driving logic representing an in-

termediate evolution stage should be included in fu-

ture investigations.

ACKNOWLEDGEMENTS

This research was part of the SMART project funded

by the Swedish Transport Administration (Traﬁkver-

ket dnr. TRV 2016/20608 and TRV 2019/27044)

and the CoEXist project funded by the European

Union H2020-ART-2016-2017 (grant agreement no.

723201). The authors would like to thank Peter

Sukennik and Jochen Lohmiller (PTV) and Iman

Pereira (VTI) for their support regarding the simula-

tions.

REFERENCES

Arem, B., van Driel, C., and Visser, R. (2007). The im-

pact of cooperative adaptive cruise control on trafﬁc-

ﬂow characteristics. Intelligent Transportation Sys-

tems, IEEE Transactions on, 7:429 – 436.

Calvert, S., Schakel, W., and Lint, J. (2017). Will automated

vehicles negatively impact trafﬁc ﬂow? Journal of

advanced transportation, 2017.

Carreras, A., Daura, X., Erhart, J., and Ruehrup, S. (2018).

Road infrastructure support levels for automated driv-

ing. 25th ITS World Congress, (September):12–20.

Coexist (2020). H2020 CoExist website:

https://www.h2020-coexist.eu/. Jan 2020.

ERTRAC (2019). Connected Automated Driving Roadmap

website: https://www.ertrac.org/. March 2019.

HF Auto (2017). Human factors of automated driving. Final

Report - Publishable Summary.

VEHITS 2021 - 7th International Conference on Vehicle Technology and Intelligent Transport Systems

150

Litman, T. A. (2015). Autonomous vehicle implementation

predictions: Implications for transport planning. Vic-

toria Transport Policy Institute.

Liu, H., Kan, X., Shladover, S., Lu, X.-Y., and Ferlis, R.

(2018). Impact of cooperative adaptive cruise control

on multilane freeway merge capacity. Journal of In-

telligent Transportation Systems, pages 1–13.

Lu, Q., Tettamanti, T., H

orcher, D., and Varga, I. (2019).

The impact of autonomous vehicles on urban trafﬁc

network capacity: an experimental analysis by micro-

scopic trafﬁc simulation. Transportation Letters The

International Journal of Transportation Research.

Milakis, D., Snelder, M., Arem, B., Wee, B., and Homem de

Almeida Correia, G. (2017). Development and trans-

port implications of automated vehicles in the nether-

lands: Scenarios for 2030 and 2050. European Jour-

nal of Transport and Infrastructure Research, 17:63–

85.

Minderhoud, M. and Bovy, P. (1999). Impact of intelligent

cruise control on motorway capacity. Transportation

Research Record, 1679:1–9.

Mintsis, E., Koutras, D., Porfyri, K., Mitsakis, E., L

ucken,

L., Erdmann, J., Fl

otter

od, Y.-P., Alms, R., Rondi-

none, M., Maerivoet, S., Carlier, K., Zhang, X.,

Blokpoel, R., Harmenzon, M., and Boerma, S. (2019).

Deliverable 3.1: Modelling, simulation and assess-

ment of vehicle automations and automated vehicles’

driver behaviour in mixed trafﬁc. Deliverable 3.1 of

TransAID project.

Olia, A., Razavi, S., Abdulhai, B., and Abdelgawad, H.

(2018). Trafﬁc capacity implications of automated ve-

hicles mixed with regular vehicles. Journal of Intelli-

gent Transportation Systems, 22:244–262.

Olstam, J., Johansson, F., Alessandrini, A., Sukennik, P.,

Lohmiller, J., and Friedrich, M. (2020). An approach

for handling uncertainties related to behaviour and ve-

hicle mixes in trafﬁc simulation experiments with au-

tomated vehicles. Journal of Advanced Transporta-

tion, 2020.

PTV (2020). Ptv vissim 2020 user manual. AG Karlsruhe.

Rios-Torres, J. and Malikopoulos, A. (2017). Impact of con-

nected and automated vehicles on trafﬁc ﬂow. pages

1–6.

SAE (2018). Taxonomy and deﬁnitions for terms related

to driving automation systems for on-road motor ve-

hicles, j3016. SAE International.

Sonnleitner, J. (2018). Deliverable 2.7: Av ready macro-

scopic modelng tool. Deliverable 2.7 of CoExist

project.

Sukennik, P. (2018a). Deliverable 2.4: Ptv vissim extension

- new features and improvements. Deliverable 2.4 of

CoExist project.

Sukennik, P. (2018b). Deliverable 2.5: Micro-simulation

guide for automated vehicles. Deliverable 2.5 of Co-

Exist project.

Talebpour, A. and Mahmassani, H. (2016). Inﬂuence of

connected and autonomous vehicles on trafﬁc ﬂow

stability and throughput. Transportation Research

Part C Emerging Technologies, 71:143–163.

Tillema, T., Gelauff, G., Waard, J., Berveling, J., and Moor-

man, S. (2017). Paths to a self-driving future - ﬁve

transition steps identiﬁed. KiM Netherlands Institute

for Transport Policy Analysis.

Yuan, Y.-M., Jiang, R., Hu, M.-B., Wu, Q., and Wang, R.

(2009). Trafﬁc ﬂow characteristics in a mixed traf-

ﬁc system consisting of acc vehicles and manual ve-

hicles: A hybrid modelling approach. Physica A,

388:2483–2491.

Effects on Trafﬁc Performance Due to Heterogeneity of Automated Vehicles on Motorways: A Microscopic Simulation Study

151