A Self-organising System Combining Self-adaptive Trafﬁc Control and

Urban Platooning: A Concept for Autonomous Driving

Heiko Hamann

, Julian Schwarzat

, Ingo Thomsen

and Sven Tomforde

Intelligent Systems, Christian-Albrechts-Universit

at zu Kiel, 24118 Kiel, Germany

Institute of Computer Engineering, University of L

ubeck, Ratzeburger Allee 160, 23562 L

ubeck, Germany

Keywords:

Platooning, Progressive Signal Systems, Self-organisation, Swarm Coordination, Trafﬁc Control.

Abstract:

Platooning is an approach to coordinate the driving behaviour of vehicles on major roads such as motorways.

The aim is to take advantage of, e.g., slipstream effects to reduce cost. We present an approach to transfer

the platooning concept to urban road networks of cities. The reduced slipstream effect is compensated by

integration with the signalisation infrastructure to dynamically allow for prioritisation of platoons using pro-

gressive signal systems (i.e., “green waves”). We deﬁne the scenario and derive a research road map towards

fully self-organised platoon operations and integrated coordination with self-adaptive and self-organising ur-

ban trafﬁc control systems. Starting from both directions, that is, self-organised urban platooning as well

as self-organised progressive signal systems in urban road networks, we deﬁne the scenario, identify main

challenges, and present ﬁrst results to demonstrate the feasibility of our research agenda.

1 INTRODUCTION

The ﬁeld of self-adaptive and self-organising (SASO)

systems, as an umbrella for initiatives such as Au-

tonomic Computing (Kephart and Chess, 2003),

Organic Computing (M

uller-Schloer and Tomforde,

2017) or Interwoven Systems (Tomforde et al., 2014),

aims at developing mechanisms to counter complex-

ity in control problems of technical systems. The vi-

sion is to distribute decision-making about appropri-

ate behaviour to a set of autonomous entities that co-

operate with each other based on current goals and

demands. Compared to standard system design and

operation, the advantage is a more robust and more ef-

ﬁcient solution while simultaneously countering chal-

lenges such as single-point-of-failure, limited scala-

bility, or decreased administrability.

Trafﬁc operation is an ideal environment for

SASO technology. Individual cars can already oper-

ate fully autonomously, the control problem of inter-

sections has typically a limited scope, and the urban

road network can be tremendously large, which turns

centrally optimised dynamic planning infeasible. As

a result, a variety of SASO-based trafﬁc control and

management systems have been proposed in the last

decades (Rehena and Janssen, 2018). These systems

mostly consider the current state-of-the-art in trafﬁc

operation and neglect trends towards green and au-

tonomous driving. Alternatives to the combustion en-

gine (electric, hydrogen) may pave the path towards

more efﬁciency-focused, green solutions. However,

whether our growing cities will be able to withstand

and whether society will continue to tolerate individ-

ual trafﬁc in cities is questionable. Platooning and its

consideration in urban trafﬁc control may serve as a

compromise between individual and public transport

with good user acceptance.

Until now, swarm behaviour (Hamann, 2018) to

enable platooning and (centralised or decentralised)

infrastructure-based control have been considered

only in isolation. We propose a concept for an inte-

grated solution that tightly couples self-organised pla-

tooning with maintenance and decentralised SASO-

based trafﬁc control. Such an integrated approach has

to balance the potentially conﬂicting goals of differ-

ent stakeholders. For instance, a classic urban trafﬁc

planning perspective focuses on reducing travel times,

number of stops, emissions, as well as on strong ca-

pacity utilisation. On the other hand, platooning is

advantageous for individual cars when being coor-

dinated and accordingly prioritised to decrease their

expected travel times. This allows, in turn, trafﬁc

authorities to increase the efﬁciency of public trans-

portation through urban platooning.

Based on an initial discussion of platooning con-

cepts in Section 2, this paper deﬁnes in Section 3 such

Hamann, H., Schwarzat, J., Thomsen, I. and Tomforde, S.

A Self-organising System Combining Self-adaptive Trafﬁc Control and Urban Platooning: A Concept for Autonomous Driving.

DOI: 10.5220/0010436304290437

In Proceedings of the 7th International Conference on Vehicle Technology and Intelligent Transport Systems (VEHITS 2021), pages 429-437

ISBN: 978-989-758-513-5

429

Figure 1: Example of a platoon on a highway simulated in

PLEXE/SUMO.

an integrated platoon-based urban trafﬁc control ap-

proach. We then outline a research road map in Sec-

tion 4 and describe preliminary results in Section 5.

Finally, in Section 6 we summarise the paper and give

an outlook to our future work.

2 PLATOONING

2.1 Term Deﬁnition

Shladover (Shladover, 2007) deﬁned platooning as

“spontaneous and dynamic forming of convoys of ve-

hicles, so-called platoons.” This is based on the idea

that each participating vehicle drives within short dis-

tance to its neighbours (see Fig. 1). The concept has

gained increasing attraction due to the recent develop-

ments in autonomous driving since manual steering

is typically not an option for optimising the beneﬁts

of platoons (Bergenhem et al., 2012). In turn, vehi-

cles need to act autonomously or at least support the

driver. Conceptually, even an individual vehicle may

be considered as a platoon but the beneﬁts (such as

reduced energy consumption due to slipstream effects

or better utilisation of the infrastructure due to min-

imised distances) increase with the platoon size.

Traditional platooning scenarios are situated at

highways and motorways: Halle and Chaibdraa

deﬁne the concept of platooning as “[. . . ] ve-

hicles travel[ling] on highways in closely spaced

groups.” (Hall

e and Chaib-draa, 2005). Here, ded-

icated access and exit roads (typically at large dis-

tances in kilometre-scale), sometimes combined with

several lanes, allows for static composition of pla-

toons. The shorter the coordination time of vehi-

cles and the higher the ﬂuctuation (e.g., due to re-

organisation of the platoon), the lower is the possi-

ble beneﬁt for participants (Shladover, 2007). The

challenge in urban trafﬁc would be to quickly and

dynamically form and re-conﬁgure platoons while

still achieving beneﬁts in terms of reduced overall

organisational effort, increased safety, and possibly

saved fuel. Table 1 offers more details about the-

ses challenges while comparing traditional highway-

based platooning to novel urban platooning systems.

2.2 State of the Art in Platooning

The platooning technology dates back to the 1980s.

Current approaches mainly consider the coordination

of truck platooning due to, for example, the poten-

tially longer duration of the coordination and possible

savings in salaries. Additionally, researchers recently

focus on the efﬁcient assignment of vehicles to pla-

toons, while mostly neglecting individual constraints

and behaviour of participating drivers. For instance,

group and individual beneﬁts need to be balanced as

a vehicle leading a platoon does not save fuel. So,

from an individual perspective, drivers might try to

avoid this position. Consequently, in contrast to exist-

ing approaches (see (Bhoopalam et al., 2018) for an

overview), personal preferences of drivers and indi-

vidual decisions about which platoon to join must be

considered. To demonstrate the feasibility, a platoon-

ing coordination test environment exists (Krupitzer

et al., 2019), which is based on the PLEXE platooning

simulation (Segata et al., 2014). A recent overview of

platooning is given by Kalbitz (Kalbitz, 2017).

One of the ﬁrst platooning projects was the PATH

programme (Shladover, 2007): All vehicles are self-

driven and have the same role. Platoons make use of

dedicated lanes and longitudinal control is achieved

by following magnetic nails in the street. In contrast,

the SARTRE project (Bergenhem et al., 2012) con-

siders platooning on existing public roads without al-

tering the infrastructure. A leading truck or bus is

steered by a trained driver, which is then followed

by autonomously driven vehicles. An additional re-

mote system guides novel drivers to the nearest pla-

toon with a suitable destination.

Another ‘Intelligent Transportation System’ (ITS)

that forgoes a modiﬁed infrastructure is En-

ergy ITS (Tsugawa et al., 2011). It employs on-

board equipment: dedicated short-range communica-

tions between vehicles as well as lidar for gap mea-

surement, which is also used in combination with

radar for obstacle detection. Only fully automated

trucks were upgraded, resulting in 14% fuel reduc-

tion.

Also focusing only on trucks, the EU project

COMPANION (Eilers et al., 2015) aims at dynamic

platooning. It uses on-board systems for coordinated

platooning along with an off-board platform for the

coordination of established platoons. The project is

supported by large car manufacturers.

Heinovski and Dressler formulate platooning as

an optimisation problem (Heinovski and Dressler,

2018). In simulations, the decentralised approach

(with less knowledge) yields on average larger pla-

toons (2.47 cars) than their centralised variant. How-

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

430

Table 1: Challenges in platooning; comparison between highways and cities (urban platooning).

challenge on highways in cities

saving gas major advantage due to slipstream minor inﬂuence of slipstream due to

low speeds but platoon-wise synchro-

nised trafﬁc lights

safety reduced lane switching reduced dangers for non-car trafﬁc

trafﬁc diversity diverse due to less signiﬁcant speed

limits

reduced due to strict speed limits

beneﬁts of switching depends on trafﬁc diversity more potential beneﬁts due to waiting

times at trafﬁc lights

beneﬁts for infrastructure increased throughput in number of ve-

hicles

prioritisation of throughput trafﬁc to

increase capacity utilisation

ever, both approaches need signiﬁcant time to form a

platoon. Concerning time-efﬁciency, the distributed

solution is slightly worse than the centralised variant.

2.3 State of the Art in Urban Platooning

As shown by Gershenson in the context of self-

organising trafﬁc lights (SOTL) (Gershenson and

Rosenblueth, 2012), platoons may emerge automat-

ically in urban road networks. In multi-agent simu-

lations based on a toroidal trafﬁc grid, applying sim-

ple rules without direct communication can reduce the

average waiting times at red lights and the number of

stopped cars. The request control of the SOTL holds

a counter for the waiting cars. When a sufﬁcient num-

ber is reached, the lights turn green, which creates

platoons of cars. However, this result is limited to

toroidal trafﬁc grids with regular topologies, such as

Manhattan-style networks. A transfer to a real-world

road network topology is pending. We propose to

study such emerging properties as part of our research

roadmap using appropriate experiment platforms.

2.4 Experimental Platforms

Many platforms for simulation aspects of platoon-

ing exist in both forms, commercial as well as open-

source, and are used in the context of road trafﬁc sim-

ulation. However, none of them currently allows for

an integrated investigation of platooning and urban

trafﬁc control behaviour. A fairly recent open-source

development with a scalable client-server architecture

is the autonomous driving simulator CARLA.

It has

a feature for cooperation (“co-simulations”) with the

Simulation of Urban MObility (SUMO),

which is an-

other open-source project, developed by the German

Aerospace Centre. The software is well maintained as

indicated by frequent updates. An active community

http://carla.org/

https://www.eclipse.org/sumo/

exists, providing well-documented projects and tuto-

rials. Vehicles in Network Simulation (Veins)

is an

open-source framework for Inter-Vehicular Commu-

nication (IVC). It is composed of SUMO and the pow-

erful event-based network simulator (OMNeT++).

For a realistic simulation of platooning, it can be

combined with the Platooning Extension for Veins

(PLEXE),

presented by Segata et al. (Segata et al.,

2014). As for commercial modelling and simula-

tion software, Simulink (The MathWorks, 2019) can

also be used in conjunction with PLEXE. Finally, the

commercial solution, Aimsun Next (Aimsun SLU,

2020),

can simulate various vehicles and pedestrians

in models of varying granularity, ranging from single

intersections up to entire urban regions.

3 SCENARIO AND VISION

In this section, we specify the envisioned scenario we

want to study. For certain aspects (e.g., centralised vs.

decentralised), we allow spectra of possibilities while

we exclude other aspects (e.g., privacy) that we keep

aside to focus on despite their importance.

Communication. We assume that all cars establish

a car2car network with scalable local communica-

tion. A (large) fraction of autonomous vehicles is con-

nected to the smart city’s infrastructure via communi-

cation links to receive information and commands.

Autonomy and Cooperation. Trafﬁc light con-

troller (TLC) determine the green-light duration at

intersections and their distributed coordination. The

coordination in progressive signal systems (PSS, so-

called ‘green waves’) is done by identifying the

https://veins.car2x.org/

https://omnetpp.org/

http://plexe.car2x.org/

https://www.aimsun.com/

A Self-organising System Combining Self-adaptive Trafﬁc Control and Urban Platooning: A Concept for Autonomous Driving

431

<platoon>)

<ID)=)x>)

<Destination)=)z>)

<Size)=)a>)

<Speed)=)b>)

<Status)=)c>)

<priority)=)d>)

</platoon>)

)

<ID)=)x>

)

<Offset)=)y>

)

<Begin)=)z>

)

synchron

isedPhase)

=)a>

)

<platoon)=)a>

)

</coordination>

)

_)_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

_)_)_)_)_)_)_)_)_)_)_)_)_)_)

_)_)_)_)_)_)_))

Figure 2: Platoon-responsive coordination of trafﬁc lights

to establish progressive signal systems.

strongest streams and/or platoons, followed by self-

organised coordination of the underlying phases at

distributed intersections (Tomforde et al., 2008). If

an autonomous car (independent of whether part of

a platoon) receives a command from the infrastruc-

ture (e.g., ‘use Lane A’ or ‘leave Lane B’), it directly

follows these commands (see Fig. 2). So we neglect

the possible intervention of users and the possibil-

ity to ignore commands at this preliminary stage of

the concept. At later stages, we plan to investigate

the impact of non-cooperative behaviours on system

performance. Initially, all cars are considered to be

fully cooperative. They voluntarily participate in the

platooning system, and they compromise to optimise

common beneﬁts. However, they are individually mo-

tivated allowing them to switch between platoons or

leave them if the beneﬁt is reduced.

Control Level. The infrastructure-based trafﬁc con-

trol is done using a centralised or a decentralised ap-

proach. We may also allow the combination of both,

therefore forming a hybrid approach. Following the

centralised approach, the system generates trafﬁc or-

ganisation plans for whole sub-networks. Following

the decentralised approach, the system is composed

of autonomous intersection controllers. Similarly, we

study a centralised and a decentralised approach to

platooning.

Urban Platooning. By default, we assume that in-

dividual driving cars try to join either an existing pla-

toon or another single car (see Fig. 3). Once a car

has joined a platoon, it continues to monitor potential

beneﬁts based on current alternative platoons includ-

ing the option of driving alone. Within a platoon, the

participants need to negotiate the trajectory (destina-

tion and route), their preferred speed, and positions

within the platoon (e.g., alternating leaders). We as-

sume a particularly increased need to reorganise pla-

toons dynamically due to the urban trafﬁc environ-

ment in comparison to classical platooning on high-

Destination: a

Speed: 2

Status: y

Destination: a

Speed: 3

Status: x

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

_ _ _

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

_ _ _ _

_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

_ _ _ _

Figure 3: Possible communication of an ad-hoc network to

establish urban platoons.

request'

to'join'

<platoon>'

<platoon'size>'

<destination>'

<average'speed>'

</platoon>'

_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'

_'_'_''

_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'

_'_'_'_''

_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'

_'_'_'_''

_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'_'

_'_'_'_''

request'

to'join'

Figure 4: Display of the increased diversity of situations in

urban platooning, the black car may consider to slow down

or even stop to join the platoon, the left-turning light-blue

car may choose to join coming from a different rout.

ways. Human drivers in an urban setting are more

likely to switch to manual driving and to change the

destination or other user-deﬁned constraints. Further-

more, we assume an increase of diversity and uncer-

tainties compared to platooning in highway scenar-

ios (see Fig. 4) because of the more complex road

network and the more diverse trafﬁc (e.g., vulnerable

road users).

Prioritisation. The SOTL has awareness of the pla-

toons and may allow them priority over other road

users (i.e., synchronisation of corresponding phases

using PSS). Even dedicated lanes for platoons can be

considered (or shared bus lanes). Prioritisation is re-

quired in the case of conﬂicting platoons and/or the

trajectory planning in the network with an impact on

both the centralised and decentralised approach.

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

432

Trafﬁc Generation. The destinations of cars are

known to the infrastructure and other cars, and

they can be modelled dynamically (i.e., as random-

waypoint model for inner-city trafﬁc) and as an

origin-destination matrix for passing trafﬁc (i.e., de-

ﬁned as streams). Route choice is modelled stochas-

tically (i.e., as shortest path, fast route, main street, or

individual route).

Privacy. We exclude privacy aspects and allow the

sharing of information about destination, speed, etc.

4 CHALLENGES AND

OBJECTIVES

The scenario speciﬁed above introduces several chal-

lenges to the operation of urban trafﬁc systems and

the coordination of autonomous vehicles. In this sec-

tion, we derive the most urgent challenges from the

overall problem statement to provide a basis for the

subsequent investigations.

4.1 Problem Statement

We propose a research track that studies the poten-

tial beneﬁt of urban platooning supported by and in-

tegrated with SOTL. Methods that need to be devel-

oped can be adjusted in two main qualitative dimen-

sions: (1) Are platoons formed and coordinated by

SOTL or by the cars themselves with SOTL inﬂuenc-

ing them almost only as a side effect? (2) Is the whole

system organised centrally or are platooning and/or

SOTL coordinated in a decentralised approach? For

dimension (1), a mixture of responsibilities between

individual cars and SOTL is possible, too. Similar, for

dimension (2), a hybrid approach could allow some

aspects being managed by a decentralised system in

alternation with situation-aware, temporary interven-

tions by centralised coordination.

4.2 Main Research Challenges

Almost all published methods about platooning fo-

cus on the standard application of platoons on high-

ways. Organising platoons within cities comes with

different requirements, new methods need to be de-

veloped, and even more so for a decentralised ap-

proach. Similarly, SOTL is usually focused on coor-

dinating individual cars or only loosely coupled ﬂows

of cars (see Table 1 for a comparison). Hence, also

methods of SOTL require changes to operate on pla-

toons or even on mixtures of platoons and individ-

ual cars. Therefore, the main research challenge is

about adapting methods from both ﬁelds and integrat-

ing them to achieve a clear beneﬁt in terms of reduced

travel times, improved safety, and reduced emissions.

We can formulate a main research question: “How to

create a self-organising system that combines SOTL

with the decentral organisation of platooning respect-

ing emerging mutual effects?” Given the state of the

art, we do not know all effects that SOTL has on pla-

tooning and how we could provoke constructive coor-

dination between SOTL and platooning to gain added

efﬁciency. This will most certainly require to adapt

classical SOTL methods to appropriately deal with

platoons instead of individual cars. Furthermore, as-

pects of privacy and user acceptance need to be con-

sidered to assess the feasibility of such an approach.

Speciﬁc for the SOTL approach is whether it can

be done centralised or needs to be done decentralised.

A centralised approach would require to plan (opti-

mised) trajectories for platoons through the network

and a platoon-aware progressive signal system. This

may become less feasible as soon as the autonomous

cars can ignore the centralised plan. A new chal-

lenge would be to integrate inner-city platoons with

prioritised public transport and similarly with special-

purpose infrastructure (e.g., bus lanes). A decen-

tralised approach would require to plan trajectories

of homogeneous platoons (e.g., same destination). In

the case of heterogeneous platoons, the split of pla-

toons would be required in certain situations. Over-

all, a key question is whether platoons can be com-

posed dynamically by external self-organisation ro-

bustly and with a clear beneﬁt. For the decentralised

approach, a complementary cornerstone is the ques-

tion of whether it requires scalability for large-scale

city networks with many intersections.

Speciﬁc for the platooning approach is whether in-

dividual cars are allowed to switch between platoons

dynamically and whether that is explicitly supported

by SOTL. Appropriate methods need to be developed

for efﬁcient platoon organisation and re-assignments.

However, we should ask whether dynamic platooning

should be substituted by static platoon assignments

in certain situations. We can even ask whether there

are situations when urban platooning is of no good

use. A unique challenge of platooning in cities could

be non-trivial interplays of phases with bursts of car-

to-platoon assignments, followed by periods of re-

duced re-assignments. Appropriate modelling tech-

niques need to be developed to reﬂect such temporal

asymmetries. Overall, a key question is a balanced

trade-off between centralised control of all platoons

with global information and scalability while achiev-

ing only sub-optimal assignments to platoons.

A Self-organising System Combining Self-adaptive Trafﬁc Control and Urban Platooning: A Concept for Autonomous Driving

433

5 CURRENT STATE OF THE

INVESTIGATIONS

Based on the overall challenges presented above,

we started the investigations from both underly-

ing perspectives: (a) the self-organised adaptation

of infrastructure-based signalisation and coordina-

tion (of trafﬁc lights) behaviour and (b) the self-

organisation of platoons in heterogeneous environ-

ments. In the following sections 5.1 and 5.2, we

present ﬁrst results of these two directions of research.

5.1 Infrastructure-supported Urban

Platooning

In (Krupitzer et al., 2018), a ﬁrst vision of how to in-

tegrate urban infrastructure and platoon coordination

at highways has been presented. The main focus is

on the connection between a centralised highway pla-

toon management and intelligent handling of platoons

leaving the highway and entering the city. Compared

to this paper, the autonomy of the individual partic-

ipants is restricted, the coordination problem is han-

dled by centralised planning, and the trafﬁc light co-

ordination is not fully integrated.

The Organic Trafﬁc Control (OTC) system (Proth-

mann et al., 2011) and its extensions serve as a basis

for investigations towards urban platooning. OTC is

a self-organised trafﬁc control system that decides lo-

cally at each intersection about the behaviour of the

underlying intersection controller. Based on the Ob-

server/Controller paradigm (Tomforde et al., 2011),

it is able to adapt the signalisation of trafﬁc lights

to changing trafﬁc demands, improve this adapta-

tion over time based on reinforcement learning (Stein

et al., 2016), to establish progressive signal systems in

a fully self-organised manner (Tomforde et al., 2010;

Tomforde et al., 2008), and to provide route recom-

mendations to drivers which reﬂect the current state

of the trafﬁc network (Sommer et al., 2016a). Based

on OTC, further contributions investigated are robust

trafﬁc demand prediction (Sommer et al., 2016b), in-

tegration of these predictions in the control strate-

gies (Sommer et al., 2015), and infrastructure-based

anticipatory route guidance (Sommer et al., 2016a;

Sommer et al., 2016b).

The basis for platoon-responsive PSS is the decen-

tralised PSS (DPSS) algorithm as originally proposed

in (Tomforde et al., 2008). This three-step process

is synchronously performed in cycles by all nodes to

establish the sequence of trafﬁc nodes for a PSS:

1. Every controller determines the pair of incoming

(‘upstream”) and outgoing (“downstream”) sec-

tions that exhibit the strongest trafﬁc ﬂow. It then

notiﬁes the upstream node to be its desired prede-

cessor. After this is done by all controllers, and a

node is elected by its downstream node, a partner-

ship is conﬁrmed. All collaborating nodes know

their partners and if they are head or tail in a PSS.

2. A common, agreed cycle time (ACT) is deter-

mined using an “echo algorithm” (Chang, 1982).

Every node i has its own desired cycle time

(DCT

), already selected by OTC. The ACT is the

longest of those, so the most heavily used node is

not restricted: ACT = max {DCT

}

The head node starts with its own DCT by setting

ACT

= DCT

and propagating it downstream.

Every node i with a higher DCT updates ACT

max{ACT, DCT

} until the tail node is reached.

ACT is now determined and propagated back so

every node can store it as ACT

= ACT

3. The time offsets o

of the nodes are calculated by

using another echo algorithm: For every node,

start times of the synchronised phase as well as

offsets, queuing and vehicle travelling times from

upstream nodes are required. When the tail is

reached, every intersection controller knows when

to activate a trafﬁc light controller (TLC) setting

that respects the ACT and establishes the PSS.

The last step requires synchronised clocks. Also,

once-established PSS are updated.

To illustrate the effect of DPSS, we consider its in-

tegration into the OTC system and compare its perfor-

mance to uncoordinated OTC nodes. The trafﬁc in the

Manhattan network in Figure 5 was simulated accord-

ing to a 3 hour, two-part trafﬁc demand: In the ﬁrst

half, 2900 cars are simulated with the most heavily

used routes A to B and D to C, followed by a change

to 3200 cars with the routes F to E, H to G, and J to I

mostly being chosen by the AIMSUN simulator.

Figure 5: Manhattan network with 6 intersections, each

with a 4-phased FTC (ﬁxed-time controller) and sections

of 250m length each.

Two experiments (uncoordinated OTC nodes vs.

OTC-DPSS) were conducted. In the case of the OTC-

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

434

DPSS simulation, every ten minutes the PSSs were

established or updated, depending on the current traf-

ﬁc situation. For the comparison, average travel times

and the average number of stops for the complete net-

work were taken into account (see Figure 6): The

number of dropped by 7%, while the average network

travel time is kept mostly constant during the simu-

lation. For both parts of trafﬁc demand, PPSs were

established along the two or three most heavily used

routes.

This preliminary experiment already shows that

there is a beneﬁt in establishing PSS in urban road

networks in response to changing trafﬁc demands.

When further compared to a centralised variant as,

e.g., discussed in (Tomforde et al., 2010), we empha-

sise the advantages of this self-organised approach:

fast reaction time, robustness against individual node

failures or message loss, and low complexity of the

control algorithm. A drawback is that it requires syn-

chronised clocks. The next steps in this line of re-

search include a) the consideration of existing pla-

toons in the simulations and to adapt the PSS strategy

towards platoon-responsive behaviour, again in a fully

decentralised manner, and b) to transfer the results to

simulations of real-world typologies.

5.2 Platooning with Increased

Uncertainties

As a ﬁrst step towards the self-organisation of pla-

toons in heterogeneous urban environments, we stud-

ied the challenges of platooning with increased uncer-

tainties. Cars not part of platoons or cars that consider

switching platoons may only have unreliable informa-

tion about other platoons. In an urban setting, it may

be proﬁtable for cars to voluntarily stay halted or to

slow down considerably and wait until a desired pla-

toon passes by. However, the relevant properties of

the platoon may change within timescales of a few

minutes. Even on shorter timescales of a few sec-

onds, arrival times of platoons may change consid-

erably due to, e.g., uncertainties in trafﬁc lights. We

expect a much more dynamic assignment of cars to

platoons followed by continuous platoon switching.

As an overview, we give a state machine in Fig. 7 that

indicates the complexity of dynamic platooning with

respect to, e.g., join-leave-join sequences and aborted

join processes.

We tested four dynamic platoon switching strate-

gies inspired by methods to solve the multi-armed

bandit problem. The assumption is that cars need

to monitor a platoon for a short duration of time be-

fore committing to join it. Relevant features of a pla-

toon (e.g., velocity, size, distance) may change dy-

namically, especially during an initial transient phase.

We assume that querying features of a platoon can be

modelled by sampling from a ﬁxed probability den-

sity. The decision to whether to join a platoon and,

if so, which on, can be seen as a multi-armed ban-

dit problem paired with an exploitation-exploration

trade-off (when to commit). Each car stores features

of surrounding platoons in an LRU cache. The most

basic method is the ε-greedy algorithm, which selects

the best candidate with a ﬁxed probability of 1−ε. As

a second method, we test Upper conﬁdence bounds

(UCB), which uses a heuristic value to estimate the

true feature value. Third, we test Bayes UCB that

assumes Gaussian distributions for the platoon fea-

tures. Forth, we test Thompson Sampling operating

on α-β-distributions. As a baseline, we compare with

the decentralised static approach of Heinovski and

Dressler (Heinovski and Dressler, 2018), which as-

signs cars to platoons once and does not implement

platoon switching. In a conservative ‘start-to-end’

10-minute scenario we study how platoons form ini-

tially with all cars started as individual cars and how

switching platoons can be beneﬁcial as all cars have

the same (far away) destination. As simulation frame-

works we use PLEXE, VEINS, and SUMO.

The dynamic methods have an advantage at later

stages of a run, as they form bigger and thus poten-

tially more efﬁcient platoons. For example, see Fig. 8

that shows the platoon size averaged over all platoons

for a single simulation run. The give-and-take side

of dynamic platooning is shown in Fig. 9 where we

compare the beneﬁts for the switching individual car

and the compromise for the target platoon in terms of

‘happiness’ (e.g., the sum of different individual-vs.-

platoon features, such as common destination, desired

speed, platoon size). Here, we assume that platoons

cannot reject a car that wants to join. Bayes UCB (in

the tested parameterisation) turned out to be conser-

vative and rarely switched platoons. All three other

methods are more active and have a non-zero proba-

bility that even for the switching car the situation may

worsen (upper panel in Fig. 9 for < 0). However, con-

siderable improvements can be achieved, too. As seen

in the lower panel of Fig. 9, platoons almost never

improve by allowing other cars to join for the tested

scenario. Whether platoon switching is proﬁtable de-

pends considerably on the scenario and in particular

the diversity of the trafﬁc (e.g., different speeds, des-

tinations) and the density of the trafﬁc (number of

neighbouring platoons that can be joined). In sum-

mary, an intermediate result is that dynamic platoon-

ing is challenging and its beneﬁt is sensitive to many

aspects of the trafﬁc situation.

A Self-organising System Combining Self-adaptive Trafﬁc Control and Urban Platooning: A Concept for Autonomous Driving

435

Figure 6: Average travel time and number of stops for uncoordinated OTC nodes and OTC-DPSS.

SPAWNING

(START)

SINGLE CAR

PREPARE JOINING

JOINING PROCESS

PLATOON

MERGING

PREPARE LEAVING

LEAVING PROCESS

NO PLATOONING

initial phase

passed

timer

found

candidate

not feasbile

any more

candidate

in range

setup ok

better candidate

unhappy / exit close

setup ok

better candidate

unhappy or

exit close

leader found

better candidate

success or abort

END

reached

end of route

exit close

Figure 7: States in dynamic platooning with decentralised

autonomous reassignment by individual cars.

6 CONCLUSION

Urban road networks are characterised by increasing

trafﬁc loads worldwide, rendering the available ca-

pacity of handling trafﬁc volumes increasingly inap-

propriate. As an alternative to extending the infras-

tructure, we presented a concept for improving the ef-

ﬁciency of utilising the existing capacity of such net-

works by introducing an integrated approach of de-

centralised dynamic platoon formation and mainte-

nance inspired by swarm robotics (Hamann, 2018), on

the one hand, as well as self-organised and platoon-

responsive coordination of trafﬁc lights to establish

‘green waves,’ on the other hand. Based on the deﬁni-

tion of the scenario and the underlying assumptions,

we described a research road map towards fully oper-

Figure 8: Dynamic decentralised platooning in an increased

uncertainty scenario comparing ﬁve algorithms (four dy-

namic platooning algorithms and one static platoon assign-

ment algorithm: ‘Heinvoski’), average platoon size over a

period of 10 minutes for different platoon switching strate-

gies based on multi-armed bandit methods.

Figure 9: Dynamic decentralised platooning in an increased

uncertainty scenario comparing four algorithms, average

happiness improvements for a switching individual car and

its target platoon (numbers in the legend give the average

improvement for each method).

able self-organised urban platooning, which respects

the autonomy of individual drivers and the overall ser-

vice goals of the infrastructure.

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

436

We presented two case studies of preliminary

work on integrated urban platooning with two dif-

ferent perspectives: On the one hand, we performed

an analysis of self-organised ‘green waves’ in sim-

ulations of real-world trafﬁc network topology with

actual trafﬁc demands. On the other hand, we pre-

sented preliminary results of comparing dynamic pla-

toon formation algorithms under increased uncer-

tainty. Our future work investigates possibilities to

close the gap between these two research directions.

REFERENCES

Aimsun SLU (2020). Aimsun Next Professional, Version 20.

Barcelona, Spain.

Bergenhem, C., Petterson, H., Coelingh, E., Englund, C.,

Shladover, S., and Tsugawa, S. (2012). Overview of

Platooning Systems. In ITSWC, pages 1393–1407.

Bhoopalam, A. K., Agatz, N., and Zuidwijk, R. (2018).

Planning of truck platoons: A literature review and di-

rections for future research. Transportation Research

Part B: Methodological, 107:212–228.

Chang, E. J. H. (1982). Echo algorithms: Depth parallel

operations on general graphs. IEEE Transactions on

Software Engineering, (4):391–401.

Eilers, S., M

artensson, J., Pettersson, H., Pillado, M.,

Gallegos, D., Tobar, M., Johansson, K. H., et al.

(2015). Companion–towards co-operative platoon

management of heavy-duty vehicles. In 18th Int. Conf.

on Int. Transp. Sys., pages 1267–1273. IEEE.

Gershenson, C. and Rosenblueth, D. A. (2012). Self-

organizing trafﬁc lights at multiple-street intersec-

tions. Complexity, 17(4):23–39.

Hall

e, S. and Chaib-draa, B. (2005). A collaborative driv-

ing system based on multiagent modelling and simu-

lations. Transp. Res. Part C, 13(4):320–345.

Hamann, H. (2018). Swarm Robotics: A Formal Approach.

Springer, Cham.

Heinovski, J. and Dressler, F. (2018). Platoon formation:

Optimized car to platoon assignment strategies and

protocols. In 2018 IEEE Vehicular Networking Con-

ference (VNC), pages 1–8.

Kalbitz, T. (2017). A comparison of approaches for pla-

tooning management. Technical report.

Kephart, J. and Chess, D. (2003). The Vision of Autonomic

Computing. IEEE Computer, 36(1):41–50.

Krupitzer, C., Lesch, V., Pfannem

uller, M., Becker, C., and

Segata, M. (2019). A Modular Simulation Frame-

work for Analyzing Platooning Coordination. In Proc.

TOP-Cars.

Krupitzer, C., Segata, M., Breitbach, M., El-Tawab, S.,

Tomforde, S., and Becker, C. (2018). Towards

infrastructure-aided self-organized hybrid platooning.

In 2018 IEEE Global Conference on Internet of

Things (GCIoT), pages 1–6. IEEE.

uller-Schloer, C. and Tomforde, S. (2017). Organic Com-

puting – Technical Systems for Survival in the Real

World. Birkh

auser. ISBN: 978-3-319-68476-5.

Prothmann, H., Tomforde, S., Branke, J., H

ahner, J.,

uller-Schloer, C., and Schmeck, H. (2011). Or-

ganic Trafﬁc Control. In Organic Computing – A

Paradigm Shift for Complex Systems, pages 431 – 446.

Birkh

auser.

Rehena, Z. and Janssen, M. (2018). Towards a framework

for context-aware intelligent trafﬁc management sys-

tem in smart cities. In Companion Proceedings of the

The Web Conference 2018, pages 893–898.

Segata, M., Joerer, S., Bloessl, B., Sommer, C., Dressler, F.,

and Cigno, R. L. (2014). Plexe: A platooning exten-

sion for Veins. In IEEE Vehicular Networking Con-

ference (VNC), pages 53–60.

Shladover, S. E. (2007). Path at 20—history and major

milestones. IEEE Transactions on intelligent trans-

portation systems, 8(4):584–592.

Sommer, M., Tomforde, S., and H

ahner, J. (2015). “Learn-

ing a Dynamic Re-combination Strategy of Forecast

Techniques at Runtime. In 2015 Int. Conf. on Auto-

nomic Comp., pages 261 – 266. IEEE.

Sommer, M., Tomforde, S., and H

ahner, J. (2016a). An Or-

ganic Computing Approach to Resilient Trafﬁc Man-

agement. In Autonomic Road Transport Support Sys-

tems, pages 113 – 130. Birkh

auer.

Sommer, M., Tomforde, S., and H

ahner, J. (2016b).

Forecast-augmented Route Guidance in Urban Traf-

ﬁc Networks based on Infrastructure Observations. In

Int. Conf. on Veh. Techn. and Int. Transp. Sys., pages

177–186. INSTICC, ScitePress.

Stein, A., Tomforde, S., Rauh, D., and H

ahner, J. (2016).

Dealing with Unforeseen Situations in the Context of

Self-Adaptive Urban Trafﬁc Control: How to Bridge

the Gap. In 2016 Int. Conf. on Autonomic Comp.

The MathWorks, I. (2019). Simulink. Natick, Mas-

sachusetts, United State.

Tomforde, S., H

ahner, J., and Sick, B. (2014). Interwoven

systems. Inform. Spektrum, 37(5):483–487.

Tomforde, S., Prothmann, H., Branke, J., H

ahner, J., Mnif,

M., M

uller-Schloer, C., Richter, U., and Schmeck, H.

(2011). Observation and control of organic systems.

In Organic Computing—A Paradigm Shift for Com-

plex Systems, pages 325–338. Springer.

Tomforde, S., Prothmann, H., Branke, J., H

ahner, J.,

uller-Schloer, C., and Schmeck, H. (2010). Possibil-

ities and Limitations of Decentralised Trafﬁc Control

Systems. In WCCI’10, pages 3298–3306. IEEE.

Tomforde, S., Prothmann, H., Rochner, F., Branke, J.,

ahner, J., M

uller-Schloer, C., and Schmeck, H.

(2008). Decentralised progressive signal systems for

organic trafﬁc control. In 2008 Int. Conf. on Self-

Adaptive and Self-Organising Sys., pages 413–422.

Tsugawa, S., Kato, S., and Aoki, K. (2011). An automated

truck platoon for energy saving. In 2011 Int. Conf. on

Int. Rob. and Sys., pages 4109–4114. IEEE.

A Self-organising System Combining Self-adaptive Trafﬁc Control and Urban Platooning: A Concept for Autonomous Driving

437