You are on page 1of 14

Swarm Robotics:

Past, Present, and Future


By MARCO DORIGO , Fellow IEEE
Institut de Recherches Interdisciplinaires et de Développements en Intelligence Artificielle (IRIDIA),
Université Libre de Bruxelles (ULB), 1050 Brussels, Belgium
GUY THERAULAZ
Centre de Recherches sur la Cognition Animale (CRCA), Centre de Biologie Intégrative (CBI),
CNRS, Université de Toulouse-Paul Sabatier, 31062 Toulouse, France
VITO TRIANNI
Institute of Cognitive Sciences and Technologies (ISTC), National Research Council (CNR),
00185 Rome, Italy

Swarm robotics started out as


an application of swarm intelli-
gence [3], [4], that is, the com-
putational modeling of collective,
self-organizing behavior that has
resulted in several successful opti-
mization algorithms [5], [6] now
being used in fields ranging from
telecommunications [7] to simu-
lation and prediction of crowd
behavior [8]. However, it has
quickly become evident that achiev-
ing swarm behavior in robots
demands much more than simply
applying swarm intelligence algo-
rithms to existing robotic plat-
forms. In fact, it often requires
to completely rethink traditional
robotic activities, such as percep-
tion, control, localization, and the

S
very design of the robotic plat-
forms themselves. Over the last
warm robotics deals with the design, construction, and deployment
two decades, researchers working
of large groups of robots that coordinate and cooperatively solve
in swarm robotics have made sig-
a problem or perform a task. It takes inspiration from natural
nificant progress, providing proofs
self-organizing systems, such as social insects, fish schools, or bird
of concept that demonstrated the
flocks, characterized by emergent collective behavior based on simple local
potential of robot swarms, also con-
interaction rules [1], [2]. Typically, swarm robotics extracts engineering prin-
tributing to a better understanding
ciples from the study of those natural systems in order to provide multirobot
of how complex behaviors emerge
systems with comparable abilities. This way, it aims to build systems that are
in nature. Translating this research
more robust, fault-tolerant, and flexible than single robots and that can better
into practice represents a challenge
adapt their behavior to changes in the environment.
that still needs to be appropriately
tackled. As a matter of fact, as of
Digital Object Identifier 10.1109/JPROC.2021.3072740 today, only a few experiments have
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.
For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/

1152 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

day—remains the only example of


self-organized teams of robots that
cooperate to solve a complex task,
with the robots in the swarm taking
different roles over time [27]. The
Swarmanoid project (2006–2010)
extended the ideas and algorithms
developed in Swarm-bots to hetero-
Fig. 1. Citation count for the search “swarm robotics” in Google Scholar and in SCOPUS,
geneous robot swarms composed
showing an increasing trend after year 2000. of three types of robots—flying,
climbing, and ground-based—that
collaborated to carry out a search and
managed to demonstrate a large num- the year 2000 that it started to signif- retrieval task [28], [29].
ber of autonomous self-organizing icantly grow. In parallel with the successful
robots, and no real-world application Initially, the study of swarm robot- demonstration of the swarm robot-
of swarm robotics exists. More ics was aimed at testing the concept of ics paradigm, research on hardware
research is needed to establish the stigmergy (see Table 1 for the defini- miniaturization promised the deploy-
knowledge and practice required to tion of this and other concepts used ment of hundreds, possibly thousands
bring robot swarms out of the lab and in this article) as a means of indi- of cooperating robots (see Fig. 2).
into the real world. rect communication and coordination Robots became smaller and evermore
In the following, after a brief his- between robots. Following a few ini- minimalist, up to attempts of designs
tory of the field, we summarize the tial attempts [9]–[11], several stud- at the millimeter scale. Several chal-
main lessons learned during the pio- ies appeared after 2000 focused on lenges related to hardware miniatur-
neering phase of swarm robotics, tasks such as object retrieval (forag- ization and to the integration of a
we analyze the main open challenges ing [13]; stick pulling [14]), cluster- sufficient sensor suite, however, hin-
and provide examples of innovative ing [15], and sorting of objects [16]. dered progress in this direction. It was
and promising research directions. These studies started from known only a few years later that a hardware
Finally, we suggest the most likely behaviors observed in social insects concept appeared supporting exper-
fields of application for swarm robot- and deployed robot swarms demon- imentation with a thousand robots:
ics and assess its potential impact on strating similar behavior. In a few the kilobot [30]. Kilobots were con-
selected industries, by showing appli- cases, the robot swarm was exploited ceived to support the first demonstra-
cation scenarios that cannot be tack- to closely replicate the dynamics tion of a large robot swarm designed
led by a single robot, or by a few observed in biological systems (e.g., for shape formation [31] and have
robots controlled in a traditional, cen- aggregation in cockroaches [17]), been later used for several successful
tralized way. leading to the first example of a studies, allowing swarm robotics to be
mixed biological-robotic society [18]. demonstrated in physical settings with
In addition, swarms of robots have hundreds of robots [32]–[34].
I. B R I E F H I S T O R Y O F been used as a tool to address biolog- Swarm robotics is not limited to
S WA R M R O B O T I C S ical questions (e.g., what is the trail ground platforms: recent work has
In the last two decades, swarm robot- network geometry to find the shortest considered aquatic surface [38] and
ics has grown from a small domain path between a food source and a underwater robots [39], as well
initiated by a few studies with a nest [19]). as swarms of flying drones [40],
clear biological inspiration [9]–[12] One of the first international [41]. While aquatic and underwa-
to a mature research field involving projects to investigate cooperation ter technologies still need substan-
several labs and researchers world- in a swarm of robots was the tial development efforts to become
wide. A search with Google Scholar Swarm-bots project funded by the mature, drones are instead already
shows that the phrase “swarm robot- European Commission between commercialized and represent a very
ics” made its first appearance in 1991, 2001 and 2005. In this project, promising platform for remote sens-
and that its usage remained very lim- a swarm of up to 20 robots capable ing applications in different domains,
ited until 2003 when it started to of self-assembly—i.e., physically being currently hindered only by the
grow considerably. Similarly, a search connecting to each other to form a lack of a legal framework authorizing
with SCOPUS returns a comparable cooperating structure—were used to autonomous and group flight.
increasing trend (see Fig. 1). These study a number of swarm behaviors, Beyond hardware platforms,
data show that, even though the such as collective transport, area cov- controlling robot swarms has rep-
swarm robotics research field finds erage, and object search [25], [26]. resented the main focus of research.
its roots in a few seminal works The main result of the project was to An extensive report of the different
published in the 1990s, it is only with demonstrate what—at the present approaches so far available in the

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1153


Point of View

Table 1 Glossary of Key Terms Used in This Article

literature is beyond the scope of II. L E S S O N S L E A R N E D and communicating with each other,
this perspective (but see [42]–[46]). AND OPEN PROBLEMS as well as of recognizing peers and
The main directions taken so far Even though the ultimate goal of the work done by them. This entails
include the following: the devel- swarm robotics is to produce method- tailored hardware designs and specific
opment of analytical models of ologies and tools that make possi- sensing, processing, and interaction
swarm systems to guide the robotics ble the deployment of robot swarms abilities. Current limitations in
implementation [47]–[49]; the adop- for the solution of real-world prob- robot hardware and control have
tion of (evolutionary) optimization lems, currently, the focus remains on constrained the complexity of swarm
approaches where robots are guided the scientific understanding of the robotics research in two different
by minimalistic controllers (neural mechanisms that would inform such ways. On the one hand, specific
networks [50]; controllers without methodologies and tools. The first two robots have been developed to
computation [51], [52]; finite-state decades of research have taught us solve specific (toy) problems (e.g.,
machines [53]; and grammar-based a number of important lessons, also termes [56] and kilobots [30]).
controllers [54]); and the devel- raising a few open problems that need These examples have opened new
opment of design and verification to be addressed and solved. research directions but not always
methodologies [20], [55]. As will be First, we learned that the types of resulted in reusable components to
discussed in the following, the def- tasks that can currently be performed be borrowed in different contexts.
inition of a reliable and efficient by robot swarms are strongly On the other hand, generic robots
engineering methodology for robot constrained by the still limited (alice [19], [57] and e-puck [37])
swarms is still on the fringes of cur- capabilities of autonomous robots. have been used to produce proofs
rent research and will likely require To work in a swarm, the individual of concept, often addressing tasks
substantial effort in the years to come. robots must be capable of interacting that are a direct transposition in the

1154 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

for swarm robotics research. Substan-


tial effort is still needed to deploy
swarm robotics hardware that satisfies
the needs of the research community.
First, a good compromise between
cost, size, and onboard features must
be provided, ensuring a sufficiently
rich set of sensors and actuators while
keeping size limited to favor exper-
imentation with hundreds of robots
within a research lab. In this sense,
a size between that of the kilobot
and that of the e-puck (about 5-cm
diameter) could be a good compro-
mise. Modular approaches allowing to
plug-in extensions with new sensors,
actuators, or computational power
proved successful with the e-puck
but require a careful design. The
possibility to program and recharge
many robots in parallel—as done
with kilobots—strongly simplifies the
experimental activities when dealing
with large numbers, especially when
manual interventions to move robots
around are not required (e.g., wireless
recharging stations integrated into the
experimental environment, or even
electric floors for battery-less opera-
tion). A centralized system to autom-
Fig. 2. Some of the robots largely used in swarm robotics research: (a) jasmine [35]
atize experimental activities, capable
(photo: WikiMedia Commons); (b) alice [36] (photo courtesy of Simon Garnier);
(c) kilobots [30] (photo courtesy of Massimo Berruti); (d) e-pucks [37]; (e) swarm-bots [26];
of observing the robots’ state, mov-
and (f) swarmanoid [29]. ing them around, and logging data
acquired by the robots, would be of
great help to speed up research and
artificial world of analogous tasks To progress in swarm robotics would greatly benefit the community
performed by self-organized natural research, we will need to develop worldwide.
systems (e.g., foraging [13], [34]). tools that will make it easier for Simulating hardware is also a
However, when the hardware is not swarm robotics researchers to share fundamental aspect of swarm robotics
conceived for swarm robotics, daily results and replicate experiments. research but raises similar prob-
work can become very cumbersome A few general-purpose robotic plat- lems as with physical robots. Often,
due to the need to deal with dozens forms would constitute valuable tools the simulation software is developed
or possibly hundreds of robots at for the research community. The from scratch for each new robot
the same time, making mundane e-puck [37] is probably the most used swarm demonstrator. A common sim-
activities, such as recharging batteries swarm robotics platform to date, but ulation tool shared by the research
or uploading software, really tedious. research with more than 30 e-pucks community would be a significant
This has often limited the number of remains complex and costly. The kilo- step forward as it would simplify the
robots in the tested swarms, reducing bot, being conceived for swarm robot- sharing and comparison of research
the breadth and significance of the ics research, is widely used, but it results. However, to devise such a
demonstrations. Finally, it should is severely limited in its abilities, tool, we need to better understand
be added that the miniaturization so much that virtualization environ- the relation between simulation and
of hardware will be a key element ments have been proposed to increase the real world. The problem, known
for experimentation in the lab with the research possibilities [59], [60]. in robotics as the simulation–reality
large swarms and for many future Crazyflies [61] are also becoming gap [62], is that differences between
applications. Still, downscaling hard- very much used as flying platforms the models used in simulation and
ware poses extremely hard problems for swarm robotics studies [41] even their real-world counterparts cause
that, so far, have not been solved [58]. though they were not conceived a drop in performance when robot

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1155


Point of View

controllers developed in simulation new simulation configuration should integration and refinement [55].
are used in the real world. This prob- not require expert knowledge of the In addition, current practices need
lem is particularly important in swarm inner working of the software. Finally, to scale up in the size of swarms,
robotics where it is exacerbated by the it will be important for the simula- seamlessly transitioning from small to
fact that many robots have to interact tion framework to be integrated with large groups. We need design method-
with each other [44]. The ideal robot standard robotics tools and libraries ologies that enable us to program a
swarm simulator should make sure (e.g., ROS) and to allow cross compi- robot swarm without being concerned
that such discrepancies are kept to a lation with respect to the robotic plat- with the swarm/problem size, which
minimum even though they cannot be form, which would make it possible to should instead be determined at
completely eliminated. test the code developed in simulation configuration time. Finally, providing
Among the many simulation soft- with the real robots without the need performance guarantees is very much
wares available, ARGoS [63] stands for any change or adjustment. needed, but current practices do not
out for the native support to swarm Having the right tools, the swarm address this point sufficiently, being
robotics research. ARGoS allows the robotics research community will limited to empirical assessments
real-time dynamical simulation of up need to provide solutions to the of performance statistics. We need
to 10 000 robots due to a clever design problem. Indeed, the second instead design methodologies that
modular design and the possibility lesson that we have learned is provide performance bounds to
to parallelize the simulation. More- that addressing the micro-macro meet verification and validation
over, it includes the models of problem—how to design the swarm standards and that promote the
some of the most used robots in behavior (macro-level), given we can reliability of robot swarms especially
swarm robotics (e-pucks and kilo- only directly program the individual for application domains with hard
bots). Another interesting example is robots (micro-level) that compose constraints (e.g., space applications).
Flightmare [64], a (multi-)UAV sim- the swarm—is probably the most To concretely support the research
ulator capable of photorealistic ren- difficult aspect to be considered. community, benchmarks are invalu-
dering of the environment, useful In order to address this problem, able tools to measure the progress
for studies of visual navigation and there have been several attempts to in a quantitative way and can be
remote sensing. To improve over these propose design methodologies—often used to challenge the researchers
experiences and provide a tool that guided by biological inspiration— on tasks that grow in complexity
responds to the needs of the swarm that are general-purpose and reusable (e.g., as done in RoboCup [68]). To
robotics community—also addressing in different application contexts, give an idea of the type of bench-
the simulation–reality gap—several from design patterns [20], [21] marks needed to progress in swarm
aspects need to be addressed and to automatic design methods [50], robotics research, consider a resource
improved. For example, we will [53], [65] (see glossary in Table 1). collection problem, as done in the
need to find ways to improve the However, all these approaches are for NASA Swarmathon [69] competition.
simulation of perception and of the moment not powerful enough: To move beyond current practices,
physical (robot–robot and robot– they successfully address relatively one could set up the problem so that
environment) and nonphysical (com- simple or constrained problems but its complexity can be adjusted along
munication) interactions. Simulations rapidly show their limits as the prob- several dimensions: environment size
should be deployed at different levels lem complexity increases. A complex and topology, to test the capability
of fidelity, allowing the user to choose task is made of several subtasks of the proposed solution to adapt
the balance between speed and accu- that might require cooperation and to different problem instances and
racy. In most cases, high-fidelity sim- that have mutual dependencies to scale performance with group
ulations are not mandatory, but their and time constraints [66], [67]. size; number and distribution of
availability would largely simplify the One might be tempted to apply items to be collected, to test abilities
transition from simulation to real- available approaches to each subtask, for coordinated exploration and
ity, supporting extensive tests before obtaining building blocks to compose exploitation of resources; and type
going live on real robots. It will later. However, such a divide-and- and persistence of items, to test the
also be necessary to improve the conquer approach is not sufficient ability to collaborate for recognition
usability of the simulation by increas- to deploy usable swarm robotics and retrieval, and to adapt to a
ing the simulation speed and pro- systems because this overlooks the dynamic environment. Informational
viding simpler means of handling many possible interrelations between complexity should also be varied by
and controlling the simulated robots tasks and the way in which these can allowing multiple alternative paths for
and the environment in which they be further partitioned and scheduled, task execution—this would require
are deployed. Simulations should be leading to suboptimal solutions. the swarm to gather and aggregate
highly configurable to respond to the We need design methodologies that information about the problem and its
needs of a diverse research commu- address the complex interrelations dynamics, taking collective decisions
nity. At the same time, setting up a between subtasks via continuous when needed to optimize the task

1156 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

performance. Possibly, multiple applications. Security against external It is also interesting to note that,
interrelated tasks should be identified attacks is needed to make swarms although the collaboration between
with variable constraints in their resilient to malicious users trying to biologists and roboticists has been
temporal execution (e.g., giving sneak into and seize the swarm. How very fruitful, it has often been
precedence to some item types to to command and control a swarm is unidirectional, with robotics taking
enable retrieval of other types). Clear also extremely important, in order more than what it gave back to
performance metrics must be assigned to let users interact with the robotic biology. We believe that this sit-
to track progress and support compar- system in a meaningful and effortless uation can be improved and that
ison between different approaches. way. This also requires a high level of robot swarms could truly help biol-
If a benchmark along these lines explainability, which is necessary to ogists, providing artificial, control-
were proposed and associated with foster acceptance and trust of swarms lable models to study the effects of
standard tools—both hardware and by users and laypeople. Properly embodiment, perception, action, and
simulation, as discussed above— addressing these aspects will largely the individual cognitive requirements
an open community could gather and strengthen swarm robotics and will necessary to support collective behav-
flourish, learning from best practices promote its transition from research ior [19], [73]. In addition, the pos-
and continuously improving over the to concrete applications. sibility of integrating autonomous
achieved results. The fourth lesson that we have robots into natural swarms offers
The third lesson has been to under- learned is that the “biological novel research directions that are
stand that some of the properties inspiration tool” must be used with just starting to be explored [18],
that are given for granted in a robot great care. Taking inspiration from [74]–[76].
swarm—e.g., fault tolerance and the behavior of social insects or
scalability—are not automatically group-living species has been very III. N E W D I R E C T I O N S
provided by the swarm and require valuable in many cases because AND NEW PROBLEMS
a careful design. The difficulties are these natural swarms have properties In the near future, most swarm
even larger if one wants to provide and display behaviors that are robotics research will likely be
other properties not intrinsically fundamental for any robot swarm: devoted to finding solutions to the
granted by self-organizing robot they are “living proofs” of the fact abovementioned open problems. Such
swarms, such as robustness, flexibility, that self-organization can work in research will be very important for the
or adaptivity (see the glossary general, and they provide viable furthering of the field and for pushing
in Table 1 for a definition of these solutions for specific problems, forward the state of the art. There are,
terms). Attempts to design robot such as how a robot swarm can however, some research directions
swarms featuring such properties move in a coordinated way, allocate that might allow a larger jump
have been made by means of tasks, or make collective decisions. forward as they would investigate
theoretical approaches that neglect In this respect, we foster further either completely new approaches
the embodiment of the robots and contributions from biology to provide or areas that, even though already
their specific functional aspects in novel guiding principles, as fresh identified as open problems, have
terms of sensors and actuators. The insights about the mechanisms been understudied. We first discuss
abovementioned properties have underlying swarm intelligence will how to design and control robot
been demonstrated using mathe- continue to inform swarm robotics swarms when dealing with novel
matical models, abstract particle practitioners. However, one should and challenging situations, such as
systems, or multiagent systems in the not forget that the long-term goal of extreme constraints given by small
context of swarms performing differ- swarm robotics research is to deploy sizes and large numbers of individuals
ent behaviors (e.g., aggregation [70], in the real world robot swarms that (III-A), or the opportunities given by
collective motion [71], collective perform useful tasks. Consequently, heterogeneous robot swarms, either
decision-making [20], and pattern robot swarms should be designed with in hardware/behavior (III-B) or in
formation [72]). However, translating an engineering-oriented approach their organizational structure (III-C).
theoretical findings into working if we want them to be relevant We then move to consider new
robotic systems often requires a for real-world applications. It is, directions for designing robot swarms,
complete rethink of the approaches therefore, unlikely that biological either mimicking biologically inspired
to introduce features and constraints inspiration will be able alone to examples of responsiveness and
not considered in the necessarily guide us when the behaviors required adaptability (III-D) or adopting
simplified theoretical models and to of the robot swarm become very machine learning approaches to
account for the specificity of the target application-specific. Researchers provide swarms with the ability to
application domain. In addition, there should, therefore, avoid putting too learn and improve their performance
are key aspects that did not receive much faith in the “biological inspira- (III-E). Finally, we discuss the need
sufficient attention so far but are tion tool” and be ready to devise ad for further research in robot swarm
required for deployment in real-world hoc solutions whenever necessary. security (III-F) and in human–swarm

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1157


Point of View

interaction (III-G) that will be of [51], [52]. Solutions that design the C. Decentralization Versus
paramount importance for real-world hardware to present self-organizing Hierarchy
deployments. properties are also plausible [83], From its very beginning,
[84] although, in this case, it may swarm robotics has adopted the
A. Hardware Miniaturization be difficult to obtain flexible and self-organization paradigm, where
One of the ambitions of swarm adaptive behavior. In all these cases, the swarm control is obtained via
robotics is to design and control steering self-organization can be more simple (stochastic) rules that define
thousands of simple robots, achieving rewarding than attempting direct the way the robots interact with each
swarm-level complex behavior result- control. other and with the environment with-
ing from simple individual behaviors out exploiting any form of centralized
and numerous interactions. An aspect B. Heterogeneity control or of global knowledge. One
that can maximize the future impact The homogeneity assumption still could, however, argue that, in many
of robot swarms is the exploitation of pervades research in swarm robotics: cases, centralized or hierarchical
thousands of miniature robots, with all robots are identical and all run forms of control could make the prob-
sizes scaling down to millimeters the same control software, they are lem of designing and controlling a
or even micro- or nanometers. all replaceable, and only the indi- robot swarm easier. The introduction
Such swarms could access small vidual history of interactions with of some form of hierarchical control
confined spaces (e.g., microfluidic the (social) environment can lead might also be justified by the fact that
channels and the human body), to the expression of a somewhat hierarchies are observed in many ani-
manipulate microscopic objects specialized behavior. This assump- mal societies where they often go side
(e.g., microplastics or individual tion stems from theoretical models by side with self-organization [86],
cells), and self-organize to support of collective behavior, which often [87]. Unfortunately, these approaches
localized treatments (e.g., targeted simplify a complex phenomenon to would require the introduction of
drug delivery). To date, research gain tractability. As a matter of fact, machinery that would make the
has only scratched the surface of a self-organization in homogeneous sys- system vulnerable (single point of
domain with huge potential. However, tems has been often sufficient to failure) and difficult to scale.
downscaling the robot size brings explain experimental observations to The question of decentralization
about new challenges that need to be a great degree [1]. However, indi- versus hierarchy, or of how to inte-
addressed for swarm robotics to be viduals within natural swarms can grate these two aspects, is currently
able to offer practicable solutions. be very different from each other, understudied. A notable first step
Microrobots and nanorobots are both physically and behaviorally, with in this direction [88] proposes to
confronted with different physical individual personalities affecting the create hybrid systems where hier-
laws than at the macroscopic scale, response to environmental and social archical control structures resulting
requiring novel models of collective cues [85]. Heterogeneity is consid- from self-organizing processes can
behavior. Current approaches to ered fundamental to grant collec- appear on the fly in an ad hoc man-
microrobots and nanorobots are not tives with the flexibility of behav- ner. This would be similar to what
exploiting conventional hardware but ior, adaptivity to new conditions, occurs in some wasp colonies where
are rather made of active colloidal and resilience to external perturba- self-organizing processes lead to the
particles [77], soft-bodied (biolog- tions. All these features would ben- formation of a linear hierarchy and
ical) robots [78], bacteria-powered efit robot swarms, but heterogene- the emergence of a single reproducing
nanomachines [79], [80], and even ity is not exploited as much as it individual [87]. Mathews et al. [88]
controllable genetically engineered should. The already mentioned Swar- have created the infrastructure—
organisms [81]. Achieving and manoid project demonstrated one middleware—that allows a robot
controlling collective behavior in possible direction, by studying coor- swarm to autonomously switch from
such systems will require novel dinated collective behaviors in physi- purely self-organized control to hier-
paradigms, as the ability to precisely cally heterogeneous groups of robots archical control and back. While
governing the individual behavior will [29]. Other powerful forms of collab- experiments have demonstrated the
be forcedly limited. Also, integrating oration allow initially-homogeneous feasibility of the approach [88], [89],
conventional ways of perception and robots to learn different behaviors, much needs to be done to understand
action is extremely challenging [82], getting specialized to tasks when this how the rules that allow the creation
demanding a rethink of the strategies leads to group performance benefits of the hierarchical control structure
for designing and controlling such [54]. Taming the complexity of the should be designed as a function of
swarms. Overall, research should self-organized behavior displayed by the task that the robot swarm has to
focus on control methods that exploit heterogeneous entities is, however, perform, and how the passage from
few unreliable sensors, minimal or still very challenging but promises purely self-organized to hierarchical
completely absent computational great advances for the domain as control and back can be activated as
abilities, and unreliable actions a whole. a function of the task and of the

1158 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

environment in which the robot the availability of new deep learning F. Security
swarm is acting. techniques, could be leveraged both The use of autonomous robots out-
as a means to design the swarm side the lab will introduce security
D. Phase Transitions and behavior and to provide additional issues. Robots need to be safe while
Adaptability capabilities to individual robots to be doing their tasks [96], they should
In a real-world environment, shared within the swarm. So far, there guarantee the privacy of the data
the main challenge faced by a swarm has been little appreciation of these that they collect, and they should
of robots is to adapt to unexpected studies within the swarm robotics also be resilient to external attacks by
events, such as the presence of community. Machine learning as a malicious users trying to get control.
obstacles or changing atmospheric design methodology suffers from Such issues will be even more seri-
conditions (e.g., brightness, wind, or the problems associated with the ous in the case of robot swarms [97].
rain). All these events may prevent automatic design of robot swarms Issues such as entity authentica-
the swarm from moving forward or [44], with the additional constraints tion, data confidentiality, and data
accomplishing some tasks. In these given by online learning of behaviors integrity are amplified by the mere
conditions, the swarm must collec- by trial and error [93], with episodic presence of potentially hundreds of
tively adapt its behavior and auto- rewards and coordination problems. robots interacting with each other.
matically change its strategy. Such Model-free approaches (see Table 1) In addition, disruption in the work-
collective capabilities are observed in may be very demanding in terms ing of the swarm might be caused
some species of group-living animals of computational requirements by just a few malicious robots sneak-
(swarms of midges, schools of fish, although they can be very powerful in ing into the group [97]. Research in
and herds of sheep). In these species, handling the complex, unpredictable robot swarm security is still in its
the interactions between individuals contingencies that characterize swarm infancy. Initial work is investigating
give rise to group properties similar behavior. Model-based approaches how traditional (e.g., cryptographic
to those of a physical system close could be valuable, as learning a Merkle trees [98]) and less traditional
to a phase transition between two model of the (current) collective (blockchain [99]) approaches to secu-
macroscopic states (see the glossary behavior could lead to an efficient rity can be exploited either to add
in Table 1), resulting in an extreme design of the individual policies. The security layers or to be fully inte-
sensitivity to changes in the behavior combination of the two is currently grated in the control architecture of
of a small number of individuals [90], sought for in several domains and robots swarms. These initial works
[91]. In such conditions, the reaction could be relevant also for swarm allow to address issues such as how
of a few individuals that have detected robotics research. Besides design- to keep information in a swarm pri-
a change in the environment can ing the swarm behavior, machine vate [98], [100], how to avoid disrup-
spread to all the other group mem- learning and, especially, deep learning tion due to the presence of malicious
bers, allowing them to react efficiently approaches could find space in swarm robots [101], and how to counter
to such disturbances as a predator robotics research to provide advanced Sybil attacks [99]. Much research will
attack. Such collective adaptive capabilities to individual robots that be needed to extend these simple,
capabilities do not only result from sustain the individual and collective proof-of-concept solutions so that they
the particular form of interactions behavior. In this respect, it would can be ported to large swarms of
between individuals but also from a be important to identify methods robots acting in the real world.
modulation of the relative intensities that can leverage the information
of these interactions [92]. The available to the collective to support G. Human–Swarm Interaction
transposition of this type of properties a more efficient interpretation of the While the interaction with a single
in swarms of robots could significantly world. For instance, deep networks machine/robot is a very well-studied
increase their level of autonomy and represent the state of the art for image problem [102], the interaction with
would be a promising line of research. classification, a feature that is needed a robot swarm opens completely new
in many applications brought forth avenues. The main difficulty is given
E. Machine Learning for Robot by robot swarms. By leveraging by the fact that, with the swarm being
Swarms the presence of multiple robots self-organized, there is no clear entity
As of today, the only prominent observing the same scene, possibly with which a human could estab-
use of machine learning in swarm from different perspectives and at lish communication. Human–swarm
robotics has been the exploitation of different times, more accurate and interaction (HSI) will be necessary to
evolutionary algorithms (see the glos- computationally efficient solutions provide the swarm with information
sary in Table 1) for the development could be provided [94], [95]. Much about goals to be achieved or tasks to
of simple neural controllers driving work is needed to define the network be performed [103], [104]. A swarm
the behavior of individual robots in architectures and learning paradigms could be controlled indirectly by
the swarm. However, recent advances to support swarm-level operations of means of a few user-driven robots
in machine learning, in particular this kind. embedded within the swarm. Recent

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1159


Point of View

research in several disciplines [92], swarm robotics research will be key to choice of swarm robotics solutions
[105]–[107] has shown that a minor- address complex coordination prob- when dealing with a concrete applica-
ity of committed agents can determine lems of future robotics applications, tion problem.
the overall behavior of a group. Sim- taking into account both cooperative The first very general guideline
ilar mechanisms represent interesting scenarios (i.e., robots coordinating for is that the use of a multirobot
means for the control of robot swarms accomplishing a common task) and system—and by extension, of a
although they may introduce security semicooperative scenarios (i.e., robots robot swarm—should be envisioned
challenges that must be dealt with that are self-interested but benefit only if the problem considered
to avoid a few malicious robots from a globally efficient organization cannot be (efficiently) solved via a
taking control of the entire swarm. of activities, such as autonomous single-robot solution because it is
Alternatively, robot swarms could be vehicles). Hence, we firmly believe either too complex or too demanding
controlled or steered directly by the that pushing forward swarm robotics considering the available technology
user, and different ways have been research will benefit not only the and the application constraints.
proposed, such as through gestures field in itself but also the domains For instance, the surveillance of
[108], [109] or EEG signals [110]. of robotics, cyber–physical systems, a large area with a single robot
Direct control of a swarm by a user and sociotechnical systems at large. might not be feasible, and the only
is complicated by the fact that under- In this section, we first discuss option might be to use many robots
standing what the swarm is doing what are the general criteria that at the same time [119]. Another
might be very challenging due to would justify the use of robot swarms example is the exploration of a
the multitude of interactions happen- to solve problems or perform tasks large collapsed building by UAVs in
ing within the swarm, which might in real-world applications, and then, a search-and-rescue scenario: even
be hard to “read” for a human we give an overview of what we though, in this case, a single UAV
observer. Therefore, explainability is believe to be the main potential appli- might perform the task, this might
crucial. Possible solutions might be cation domains for swarm robotics. not be efficient enough due to the
built into the self-organizing mecha- This overview is of a speculative limited flight time and the need
nisms of the swarm, so as to make the nature since—as we said—real-world to fly back for recharging. In such
current state and goal of the swarm applications are still to come. We try, conditions, a multirobot solution
visible to users. Interfaces to swarm however, to motivate our choices by could be more efficient due to parallel
behaviors, possibly enabled by aug- critically evaluating the benefits that operation [41].
mented reality, may collect and visu- a swarm robotics approach could con- Once the suitability of a multirobot
alize information from the swarm, cretely bring into play in the different system is established, one should con-
while models of the collective behav- application domains considered. sider what type of control approach
ior could be integrated in order to would be the most appropriate for
provide predictions that could support A. General Criteria for a Robot the considered problem. For exam-
the user to take action (e.g., by issuing Swarm Solution ple, when it is not possible or advis-
new commands to the swarm). The In principle, the first question to able to coordinate the robots in a
design of any HSI solution will also ask when considering the application centralized way [120], the use of
require an understanding of the psy- of a robot swarm to the solution of a robot swarm might be the right
chological effects induced on humans a real-world problem is whether a way to go. In some cases, centralized
who interact with a robot swarm in robot swarm is indeed the best way replanning could be feasible to deal
order to favor interaction modalities to go. However, this is a very diffi- with task uncertainty and environ-
that reduce stress [111], [112] and cult question, especially so given that mental unpredictability [121]. How-
improve usability and trust [113]. swarm robotics is a young discipline ever, a strong demand for online
and that, as discussed above, there are recognition of features and adap-
IV. H O W F U T U R E still many open research questions. As tation to experienced contingencies
A P P L I C AT I O N S W I L L a consequence, the current practice might be better approached through
GUIDE RESEARCH consists of evaluating the suitability decentralized, self-organized meth-
The great interest in swarm robot- of swarm robotics solutions on the ods. Even in this case, however, one
ics research recorded to date basis of the expected advantages with should consider if other approaches,
[114]–[116] is due to the expected respect to other solutions, factoring such as distributed model predictive
forthcoming ubiquity of autonomous in the constraints imposed by avail- control [122], [123], could be used,
robots in real-world applications able technologies.1 Given the lack of 1 A notable exception is the work of
and to the challenge of letting them working methodologies to move from Kazadi [117], [118], who explicitly addresses
cooperate with each other and with problem specification to robot swarm the question of whether a robot swarm is an
their human users avoiding the implementation and deployment, in appropriate technology for a given problem;
however, his methodology is still at the stage
pitfalls of centralized control. The the following we discuss a few gen- of proposal and has not been applied on any
knowledge and practice produced by eral guidelines that should direct the real robot swarm implementation.

1160 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

which might not be the case when it applications can guide future swarm extent, disposable. Here, however,
is impossible or too difficult to cre- robotics research. Conversely, applica- the human component remains
ate simple enough models of both tions such as precision agriculture inevitably central. Hence, defense
the problem to be solved and the or infrastructure inspection and applications need to consider the
environment in which the robots are maintenance require dealing with human in the loop, and advanced HSI
going to operate. an unstructured, unpredictable strategies will be crucial for effective
Another aspect to consider is environment—often covering exten- deployment [113]. Also, safety and
whether the given problem is sive areas—and can benefit from security aspects need to be at the
decomposable in a fixed number parallelization and collaboration highest level to guarantee that robot
of well-defined tasks that can be between robots in a swarm. For swarms do not get out of control
addressed by a team of robots, each instance, early identification of the or maliciously seized [96]. Similar
with a specific role, as is the case outbreak of diseases within a crop aspects are fundamental in other
for instance in an assembly line or in field requires information sharing application areas, such as civil protec-
robotic soccer [68]. Once again, if this between robots to make global tion, where the need to face natural or
is not the case, then a swarm robotics patterns emerge from coupled local anthropogenic disasters requires agile
approach might be conceivable. views, supporting suitable responses robots capable of dealing with emer-
In other words, even if a problem and better strategic planning [95], gency situations, with no external
is better solved with a multirobot [125]. Similarly, reliable identification infrastructure or reliable maps. Such
system, this does not necessarily imply of defects in a large infrastructure applications set the bar very high,
the need for a robot swarm. The latter requires efficient search abilities as robot swarms should be capable
is better justified by tasks that have that could be best implemented of guaranteeing the highest possible
no predefined partitioning in subtasks by means of swarms [126]. Both performance and reliability, because
or that allow diverse allocations of precision agriculture and infrastruc- no victim should be left behind.
roles to the available robots [27], ture inspection happen in somewhat Space missions introduce other
[29]. Finally, a swarm robotics static environments (the crop field constraints on robotics applications
approach might be the right choice or the infrastructure to inspect). that might be successfully addressed
if a beneficial collaboration between Despite this, decentralization and by swarm robotics. In space, the com-
the robots is expected. In fact, due self-organization can be useful in putational power of computers
to collaboration, a swarm robotics improving efficiency—thanks to has to remain limited because of
system can aim for a superlinear parallel and coordinated operations— cosmic radiation burning modern
increase in performance that would and accuracy—thanks to adaptive CPUs [128]. A swarm of robots
justify the overhead necessary to set strategies for collective perception of limited computational power
up the collaboration [124]. that allow to respond to the sensed might, therefore, be a better design
contingencies and determine the choice than a single more powerful
optimal mission plan so as to robot [69], [129], [130]. Robots that
B. Applications, Needs, and maximize the likelihood that all are sent out in space cannot be easily
Future Research relevant features are observed with repaired or substituted, which is well
With these considerations in mind, care. In this respect, future research addressed by the swarm robotics
potential application domains for should focus on strategies to make focus on redundant systems where
swarm robotics should be critically sense of complex features by means the failure of one of the robots does
evaluated for the benefits that a of information fusion among multiple, cause only a graceful degradation
swarm robotics approach can con- possibly heterogeneous, robots. of the swarm performance. Finally,
cretely bring into play. For instance, In addition, tailored intervention in space, it might be extremely
although service robots are usually and manipulation abilities need costly or even impossible to build an
not organized as a swarm, the coor- to be devised (e.g., for harvesting external infrastructure to support the
dination of activities and allocation fruits or maintenance), opening to coordination of robots, again a typical
of tasks performed by each robot can new opportunities for decentralized, situation that robot swarm can effec-
be decentralized and self-organized collaborative activities. tively deal with. Accordingly, space
to some extent. Still, the specific task The application of robot swarms agencies, such as NASA and ESA,
itself may not require coordination is sought by defense agencies have started to show interest in swarm
or collaboration between robots. worldwide, that find extremely technologies, for example, with activ-
Similarly, logistics (e.g., in large appealing a system that cannot be ities such as the already mentioned
warehouses), autonomous cars, and easily shut down [127]. A system Swarmathon competition [69] and
smart mobility can surely benefit that is fault-tolerant to external with research directed at the control
from the decentralized coordination attacks can support operations in of swarms of picosatellites [130]. The
strategies studied in swarm robotics. adversarial settings, especially when great challenge brought forward by
It is, however, unlikely that these robots are replaceable and, to some space applications is the necessary

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1161


Point of View

autonomy of the swarm system, which Overall, the relationship between technologies and to foster trust in
cannot rely on reliable and constant the requirements from potential them, hence creating the conditions
intervention from human operators. application domains and future for massive real-world deployment.
The entertainment sector has also research challenges for swarm Even though many of these issues
good potential for the employment robotics is indisputable. We, there- are being addressed more generally
of robot swarms. There are already fore, envisage a close collaboration within the artificial intelligence field,
several examples of light shows between researchers and the rele- their complexity might be increased
performed with drone swarms both vant stakeholders from the various by the high number of autonomous
outdoors and indoors [131], which, application domains, who can provide entities and by their numerous
however, are generally based on cen- concrete examples to push novel interactions with each other that are
tralized preorchestrated trajectories developments and contribute to set typical of swarm robotics systems.
for the drones. In a similar way, the agenda of swarm robotics research If these challenges are overcome,
other attempts to exploit multiro- in the years to come. we expect swarm robotics to success-
bot systems for entertainment have fully transition from laboratories to
relied on some centralized solution V. C O N C L U S I O N real-world applications within the cur-
to finely control the system [132], The design and implementation of rent decade, as suggested above. Such
[133]. New avenues are possible if a effective robot swarms is among the a transition will not be immediate
decentralized approach is considered, greatest challenges that lie ahead but will progressively involve more
especially if users can actively par- for robotics, as well as one of the and more application domains in the
ticipate in the entertainment activity most promising research avenues, identification of new challenges and
by engaging with the robot swarm, as acknowledged in [116]. In this the creation of a demand for new
changing its dynamics according to article, we have briefly summarized technological solutions that will drive
positions, movements, and even emo- the state of the art and identified what research and innovation in the years
tions [134]. In this context, research we believe to be the most promising to come.
can experiment with radically new research directions and main open
modalities for HSI that can be after- problems. It should, however, be con- Acknowledgment
ward borrowed by other applica- sidered that significant advances in The authors acknowledge support
tion domains. For instance, all sorts swarm robotics are bound to progress from the Office of Naval Research
of HSI interfaces can be imagined, made outside the field. For example, Global (ONRG) for funding the work-
from wearable devices [135], aug- new materials, biohybrid solutions, shop “Swarm Robotics: Pushing the
mented and virtual reality [136], and and new ways of storing and trans- State of the Art” held in Rome, Italy,
brain–computer interaction modali- mitting energy would help address on October 25 and 26, 2018 (Award
ties [110]. some of the current issues related N62909-18-1-2156). The discussions
Finally, swarms of nanobots might, to the hardware of robot swarms. and interactions spurred from the
in the future, become a new and The development of AI techniques, workshop have been very valuable
powerful tool in precision medicine, in particular of distributed learning to inspire this article. This work was
making possible targeted interven- algorithms that require limited com- also supported by the Office of Naval
tions within the human body, such as putation and can work with the CPUs Research Global (Award N62909-18-
minimally invasive surgery or poly- of small inexpensive robots, will allow 1-2093 to Vito Trianni and Award
therapy delivery directly to cancerous robot swarms to gradually increase N62909-19-1-2024 to Marco Dorigo)
cells [137], [138]. However, the coor- their autonomy. Swarms will have to and by an ARC Project funded by the
dination of huge numbers of robots ensure explainability, now a major French Community of Belgium. Marco
with extremely limited computational issue for the whole field of robotics Dorigo acknowledges support by the
and communication capabilities will and artificial intelligence. In other Belgian Fonds de la Recherche Scien-
stretch the swarm robotics approach words, the user will need to be able tifique (F.R.S.-FNRS) of which he is
to its limits and will require the to understand the decision-making a Research Director. Guy Theraulaz
development of new conceptual tools, processes without detailed knowledge gratefully acknowledges the Indian
in addition to the development of of the underlying mechanisms—a Institute of Science to serve as Infosys
microscopic hardware or biorobotics paramount requirement to ensure Visiting Professor at the Centre for
devices [58]. the acceptability of new intelligent Ecological Sciences, Bengaluru, India.
REFERENCES
[1] S. Camazine, J.-L. Deneubourg, N. R. Franks, New York, NY, USA: Oxford Univ. Press, 1999. [7] G. Di Caro and M. Dorigo, “AntNet: Distributed
J. Sneyd, G. Theraulaz, and E. Bonabeau, [4] M. Dorigo and M. Birattari, “Swarm intelligence,” stigmergetic control for communications
Self-Organization in Biological Systems. Princeton, Scholarpedia, vol. 2, no. 9, p. 1462, 2007. networks,” J. Artif. Intell. Res., vol. 9, pp. 317–365,
NJ, USA: Princeton Univ. Press, 2001. [5] M. Dorigo, “Ant colony optimization,” Dec. 1998.
[2] D. J. T. Sumpter, Collective Animal Behavior. Scholarpedia, vol. 2, no. 3, p. 1461, 2007. [8] M. Moussaïd, D. Helbing, and G. Theraulaz, “How
Princeton, NJ, USA: Princeton Univ. Press, 2010. [6] M. Dorigo, M. A. M. de Oca, and A. P. Engelbrecht, simple rules determine pedestrian behavior and
[3] E. Bonabeau, M. Dorigo, and G. Theraulaz, Swarm “Particle swarm optimization,” Scholarpedia, crowd disasters,” Proc. Nat. Acad. Sci. USA,
Intelligence: From Natural to Artificial Systems. vol. 3, no. 11, p. 1486, 2008. vol. 108, no. 17, pp. 6884–6888, 2011. [Online].

1162 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

Available: https://www.pnas.org/content/ [28] M. Dorigo, M. Birattari, and R. O’Grady, perspectives,” Frontiers Robot. AI, vol. 4,
108/17/6884 “Swarmanoid, the movie,” in Proc. AAAI-11 Video, Mar. 2017, doi: 10.3389/frobt.2017.00009.
[9] R. Beckers, O. E. Holland, and J. L. Deneubourg, 2011. [Online]. Available: http://www. [47] M. Massink, M. Brambilla, D. Latella, M. Dorigo,
“From local actions to global tasks: Stigmergy and youtube.com/watch?v=M2nn1X9Xlps and M. Birattari, “On the use of bio-PEPA for
collective robotics,” in Proc. 4th Int. Workshop [29] M. Dorigo et al., “Swarmanoid: A novel concept modelling and analysing collective behaviours in
Synthesis Simulation Living Syst. Artif. Life IV, R. A. for the study of heterogeneous robotic swarms,” swarm robotics,” Swarm Intell., vol. 7, nos. 2–3,
Brooks and P. Maes, Eds. Cambridge, MA, USA: IEEE Robot. Autom. Mag., vol. 20, no. 4, pp. 201–228, Sep. 2013.
MIT Press, 1994, pp. 181–189. pp. 60–71, Dec. 2013. [48] A. Prorok, N. Correll, and A. Martinoli,
[10] O. Holland and C. Melhuish, “Stigmergy, [30] M. Rubenstein, C. Ahler, N. Hoff, A. Cabrera, and “Multi-level spatial modeling for stochastic
self-organization, and sorting in collective R. Nagpal, “Kilobot: A low cost robot with scalable distributed robotic systems,” Int. J. Robot. Res.,
robotics,” Artif. Life, vol. 5, no. 2, pp. 173–202, operations designed for collective behaviors,” vol. 30, no. 5, pp. 574–589, Apr. 2011.
Apr. 1999. Robot. Auton. Syst., vol. 62, no. 7, pp. 966–975, [49] K. Elamvazhuthi and S. Berman, “Mean-field
[11] C. R. Kube and H. Zhang, “Collective robotics: 2014, reconfigurable Modular Robotics. [Online]. models in swarm robotics: A survey,”
From social insects to robots,” Adapt. Behav., Available: http://www.sciencedirect.com/science/ Bioinspiration Biomimetics, vol. 15, no. 1, 2019,
vol. 2, no. 2, pp. 189–218, Sep. 1993. article/pii/S0921889013001474 Art. no. 015001, doi:
[12] C. Kube and E. Bonabeau, “Cooperative transport [31] M. Rubenstein, A. Cornejo, and R. Nagpal, 10.1088/1748-3190/ab49a4.
by ants and robots,” Robot. Auton. Syst., vol. 30, “Programmable self-assembly in a thousand-robot [50] V. Trianni, “Evolutionary swarm robotics,” in
no. 1, pp. 85–101, 2000. [Online]. Available: swarm,” Science, vol. 345, no. 6198, pp. 795–799, Evolving Self-Organising Behaviours in Groups of
http://www.sciencedirect.com/science/article/ 2014. [Online]. Available: https://science. Autonomous Robots (Studies in Computational
pii/S0921889099000664 sciencemag.org/content/345/6198/795 Intelligence), vol. 108. Berlin, Germany:
[13] M. J. B. Krieger, J.-B. Billeter, and L. Keller, [32] G. Valentini, E. Ferrante, H. Hamann, and Springer-Verlag, 2008.
“Ant-like task allocation and recruitment in M. Dorigo, “Collective decision with 100 kilobots: [51] M. Gauci, J. Chen, W. Li, T. J. Dodd, and R. Groß,
cooperative robots,” Nature, vol. 406, no. 6799, Speed versus accuracy in binary discrimination “Self-organized aggregation without
pp. 992–995, Aug. 2000. problems,” Auton. Agents Multi-Agent Syst., computation,” Int. J. Robot. Res., vol. 33, no. 8,
[14] A. J. Ijspeert, A. Martinoli, A. Billard, and vol. 30, no. 3, pp. 553–580, May 2016. pp. 1145–1161, 2014.
L. M. Gambardella, “Collaboration through the [33] I. Slavkov et al., “Morphogenesis in robot [52] A. Ozdemir, M. Gauci, S. Bonnet, and R. Groß,
exploitation of local interactions in autonomous swarms,” Sci. Robot., vol. 3, no. 25, 2018, “Finding consensus without computation,” IEEE
collective robotics: The stick pulling experiment,” Art. no. eaau9178. [Online]. Available: https:// Robot. Autom. Lett., vol. 3, no. 3, pp. 1346–1353,
Autonom. Robots, vol. 11, no. 2, pp. 149–171, robotics.sciencemag.org/content/3/25/eaau9178 Jul. 2018.
2001. [34] M. S. Talamali, T. Bose, M. Haire, X. Xu, [53] G. Francesca et al., “AutoMoDe-chocolate:
[15] W. Agassounon, A. Martinoli, and K. Easton, J. A. R. Marshall, and A. Reina, “Sophisticated Automatic design of control software for robot
“Macroscopic modeling of aggregation collective foraging with minimalist agents: swarms,” Swarm Intell., vol. 9, nos. 2–3,
experiments using embodied agents in teams of A swarm robotics test,” Swarm Intell., vol. 14, pp. 125–152, Sep. 2015.
constant and time-varying sizes,” Auton. Robots, no. 1, pp. 25–56, Mar. 2020. [54] E. Ferrante, A. E. Turgut, E. Duéñez-Guzmán,
vol. 17, nos. 2–3, pp. 163–192, Sep. 2004. [35] S. Kornienko, O. Kornienko, and P. Levi, M. Dorigo, and T. Wenseleers, “Evolution of
[16] M. Wilson, C. Melhuish, A. B. Sendova-Franks, “Minimalistic approach towards communication self-organized task specialization in robot
and S. Scholes, “Algorithms for building annular and perception in microrobotic swarms,” in Proc. swarms,” PLOS Comput. Biol., vol. 11, no. 8,
structures with minimalist robots inspired by IEEE/RSJ Int. Conf. Intell. Robots Syst., Aug. 2005, Aug. 2015, Art. no. e1004273.
brood sorting in ant colonies,” Auton. Robots, pp. 2228–2234. [55] M. Brambilla, A. Brutschy, M. Dorigo, and
vol. 17, nos. 2–3, pp. 115–136, Sep. 2004. [36] G. Caprari and R. Siegwart, “Mobile micro-robots M. Birattari, “Property-driven design for robot
[17] S. Garnier et al., “The embodiment of cockroach ready to use: Alice,” in Proc. IEEE/RSJ Int. Conf. swarms: A design method based on prescriptive
aggregation behavior in a group of micro-robots,” Intell. Robots Syst., Aug.2005, pp. 3295–3300. modeling and model checking,” ACM Trans. Auton.
Artif. Life, vol. 14, no. 4, pp. 387–408, Oct. 2008. [37] F. Mondada et al., “The e-puck, a robot designed Adapt. Syst., vol. 9, no. 4, pp. 17:1–17:28, 2015.
[18] J. Halloy et al., “Social integration of robots into for education in engineering,” in Proc. 9th Conf. [56] J. Werfel, K. Petersen, and R. Nagpa, “Designing
groups of cockroaches to control self-organized Auton. Robot Syst. Competitions, 2009, pp. 59–65. collective behavior in a termite-inspired robot
choices,” Science, vol. 318, no. 5853, [38] M. Duarte et al., “Evolution of collective behaviors construction team,” Science, vol. 343, no. 6172,
pp. 1155–1158, 2007. [Online]. Available: for a real swarm of aquatic surface robots,” PLoS pp. 754–758, 2014. [Online]. Available: https://
https://science.sciencemag.org/content/318/5853/ ONE, vol. 11, no. 3, Mar. 2016, Art. no. e0151834. science.sciencemag.org/content/343/6172/754
1155 [39] P. Zahadat and T. Schmickl, “Division of labor in a [57] S. Garnier, J. Gautrais, M. Asadpour, C. Jost, and
[19] S. Garnier, M. Combe, C. Jost, and G. Theraulaz, swarm of autonomous underwater robots by G. Theraulaz, “Self-organized aggregation triggers
“Do ants need to estimate the geometrical improved partitioning social inhibition,” Adapt. collective decision making in a group of
properties of trail bifurcations to find an efficient Behav., vol. 24, no. 2, pp. 87–101, Apr. 2016. cockroach-like robots,” Adapt. Behav., vol. 17,
route? A swarm robotics test bed,” PLoS Comput. [40] G. Vásárhelyi, C. Virágh, G. Somorjai, T. Nepusz, no. 2, pp. 109–133, Apr. 2009.
Biol., vol. 9, no. 3, Mar. 2013, Art. no. e1002903. A. E. Eiben, and T. Vicsek, “Optimized flocking of [58] M. Sitti, Mobile Microrobotics. Cambridge, MA,
[20] A. Reina, G. Valentini, C. Fernández-Oto, autonomous drones in confined environments,” USA: MIT Press, 2017.
M. Dorigo, and V. Trianni, “A design pattern for Sci. Robot., vol. 3, no. 20, 2018, Art. no. eaat3536. [59] A. Reina, A. J. Cope, E. Nikolaidis,
decentralised decision making,” PLoS ONE, [Online]. Available: https://robotics.sciencemag. J. A. R. Marshall, and C. Sabo, “ARK: Augmented
vol. 10, no. 10, Oct. 2015, Art. no. e0140950. org/content/3/20/eaat3536 reality for kilobots,” IEEE Robot. Autom. Lett.,
[21] G. Valentini, Achieving Consensus in Robot [41] K. N. McGuire, C. De Wagter, K. Tuyls, vol. 2, no. 3, pp. 1755–1761, Jul. 2017.
Swarms: Design and Analysis of Strategies for the H. J. Kappen, and G. C. H. E. de Croon, “Minimal [60] G. Valentini et al., “Kilogrid: A novel experimental
Best-of-n Problem (Studies in Computational navigation solution for a swarm of tiny flying environment for the Kilobot robot,” Swarm Intell.,
Intelligence). Cham, Switzerland: Springer, 2017. robots to explore an unknown environment,” Sci. vol. 12, no. 3, pp. 245–266, Sep. 2018.
[22] P.-P. Grassé, “La reconstruction du nid et les Robot., vol. 4, no. 35, 2019, Art. no. eaaw9710. [61] W. Giernacki, M. Skwierczyński, W. Witwicki,
coordinations interindividuelles chez [Online]. Available: https://robotics. P. Wroński, and P. Kozierski, “Crazyflie 2.0
Bellicositermes natalensis et Cubitermes sp. La sciencemag.org/content/4/35/eaaw9710 quadrotor as a platform for research and
théorie de la stigmergie: Essai d’interprétation du [42] M. Brambilla, E. Ferrante, M. Birattari, and education in robotics and control engineering,” in
comportement des termites constructeurs,” M. Dorigo, “Swarm robotics: A review from the Proc. 22nd Int. Conf. Methods Models Autom.
Insectes Sociaux, vol. 6, no. 1, pp. 41–80, swarm engineering perspective,” Swarm Intell., Robot. (MMAR), Aug. 2017, pp. 37–42.
Mar. 1959. vol. 7, no. 1, pp. 1–41, Mar. 2013. [62] N. Jacobi, P. Husbands, and I. Harvey, “Noise and
[23] G. Theraulaz and E. Bonabeau, “A brief history of [43] V. Trianni and A. Campo, “Fundamental collective the reality gap: The use of simulation in
stigmergy,” Artif. Life, vol. 5, no. 2, pp. 97–116, behaviors in swarm robotics,” in Springer evolutionary robotics,” in Proc. 3rd Eur. Conf. Adv.
1999, doi: 10.1162/106454699568700. Handbook of Computational Intelligence, Artif. Life. Berlin, Germany: Springer-Verlag, 1995,
[24] S. Garnier, J. Gautrais, and G. Theraulaz, J. Kacprzyk and W. Pedrycz, Eds. Berlin, Germany: pp. 704–720.
“The biological principles of swarm intelligence,” Springer, 2015, pp. 1377–1394, doi: [63] C. Pinciroli et al., “ARGoS: A modular, parallel,
Swarm Intell., vol. 1, no. 1, pp. 3–31, Oct. 2007. 10.1007/978-3-662-43505-2_71. multi-engine simulator for multi-robot systems,”
[25] M. Dorigo et al., “Evolving self-organizing [44] G. Francesca and M. Birattari, “Automatic design Swarm Intell., vol. 6, no. 4, pp. 271–295,
behaviors for a swarm-bot,” Auton. Robots, of robot swarms: Achievements and challenges,” Dec. 2012.
vol. 17, nos. 2–3, pp. 223–245, Sep. 2004. Frontiers Robot. AI, vol. 3, p. 29, Mar. 2016. [64] Y. Song, S. Naji, E. Kaufmann, A. Loquercio, and
[26] F. Mondada, L. M. Gambardella, D. Floreano, [Online]. Available: https://www.frontiersin. D. Scaramuzza, “Flightmare: A flexible quadrotor
S. Nolfi, J. Deneubourg, and M. Dorigo, org/article/10.3389/frobt.2016.00029 simulator,” 2020, arXiv:2009.00563. [Online].
“The cooperation of swarm-bots–physical [45] L. Garattoni and M. Birattari, “Swarm robotics,” in Available: https://arxiv.org/abs/2009.00563
interactions in collective robotics,” IEEE Robot. Wiley Encyclopedia of Electrical and Electronics [65] M. Birattari et al., “Automatic off-line design of
Autom. Mag., vol. 12, no. 2, pp. 21–28, Jun. 2005. Engineering, J. G. Webster, Ed. 2016, robot swarms: A manifesto,” Frontiers Robot. AI,
[27] S. Nouyan, R. Groß, M. Bonani, F. Mondada, and 10.1002/047134608X.W8312. vol. 6, p. 59, Jul. 2019.
M. Dorigo, “Teamwork in self-organized robot [46] G. Valentini, E. Ferrante, and M. Dorigo, [66] B. P. Gerkey and M. J. Matarić, “A formal analysis
colonies,” IEEE Trans. Evol. Comput., vol. 13, “The best-of-n problem in robot swarms: and taxonomy of task allocation in multi-robot
no. 4, pp. 695–711, Aug. 2009. Formalization, state of the art, and novel systems,” Int. J. Robot. Res., vol. 23, no. 9,

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1163


Point of View

pp. 939–954, Sep. 2004. perspectives for active matter,” Annu. Rev. scenario,” in Proc. 17th Int. Conf. Auton. Agents
[67] E. Nunes, M. Manner, H. Mitiche, and M. Gini, Condens. Matter Phys., vol. 11, no. 1, Multiagent Syst. (AAMAS), M. Dastani,
“A taxonomy for task allocation problems with pp. 441–466, Mar. 2020. G. Sukthankar, E. André, and S. Koenig, Eds.
temporal and ordering constraints,” Robot. Auton. [85] R. Jeanson and A. Weidenmüller, “Interindividual Richland, SC, USA: International Foundation for
Syst., vol. 90, pp. 55–70, Apr. 2017. [Online]. variability in social insects–proximate causes and Autonomous Agents and Multiagent Systems,
Available: http://www.sciencedirect. ultimate consequences,” Biol. Rev., vol. 89, no. 3, 2018, pp. 541–549.
com/science/article/pii/S0921889016306157 pp. 671–687, 2014. [Online]. Available: https:// [102] C. Bartneck, T. Belpaeme, F. Eyssel, T. Kanda,
[68] S. R. Hedberg, “Robots playing soccer? RoboCup onlinelibrary.wiley.com/doi/abs/10.1111/brv.12074 M. Keijsers, and S. Šabanović, Human-Robot
poses a new set of challenges in intelligent [86] I. D. Chase, “Social process and hierarchy Interaction: An Introduction. Cambridge, U.K.:
distributed computing,” IEEE Concurrency, vol. 5, formation in small groups: A comparative Cambridge Univ. Press, 2020.
no. 4, pp. 13–17, Oct. 1997. perspective,” Amer. Sociol. Rev., vol. 45, no. 6, [103] D. S. Brown, M. A. Goodrich, S.-Y. Jung, and
[69] L. A. Nguyen, T. L. Harman, and C. Fairchild, pp. 905–924, 1980. [Online]. Available: S. C. Kerman, “Two invariants of human swarm
“Swarmathon: A swarm robotics experiment for http://www.jstor.org/stable/2094909 interaction,” J. Human-Robot Interact., vol. 5,
future space exploration,” in Proc. IEEE Int. Symp. [87] G. Theraulaz, E. Bonabeau, and J.-L. Deneubourg, no. 1, pp. 1–31, Aug. 2015.
Meas. Control Robot. (ISMCR), Sep. 2019, “Self-organization of hierarchies in animal [104] A. Kolling, P. Walker, N. Chakraborty, K. Sycara,
pp. B1-3-1–B1-3-4. societies: The case of the primitively eusocial and M. Lewis, “Human interaction with robot
[70] H. Hamann and H. Wörn, “A framework of wasp Polistes dominulus Christ,” J. Theor. Biol., swarms: A survey,” IEEE Trans. Human-Mach.
space–time continuous models for algorithm vol. 174, no. 3, pp. 313–323, 1995. [Online]. Syst., vol. 46, no. 1, pp. 9–26, Feb. 2016.
design in swarm robotics,” Swarm Intell., vol. 2, Available: http://www.sciencedirect.com/ [105] J. Gautrais, C. Jost, R. Jeanson, and G. Theraulaz,
nos. 2–4, pp. 209–239, Dec. 2008. science/article/pii/S0022519385701014 “How individual interactions control aggregation
[71] T. Vicsek and A. Zafeiris, “Collective motion,” Phys. [88] N. Mathews, A. L. Christensen, R. O’Grady, patterns in gregarious arthropods,” Interact. Stud.,
Rep., vol. 517, nos. 3–4, pp. 71–140, Aug. 2012. F. Mondada, and M. Dorigo, “Mergeable nervous vol. 5, no. 2, pp. 245–269, 2004. [Online].
[72] M. Coppola, J. Guo, E. Gill, and systems for robots,” Nature Commun., vol. 8, Available: https://www.jbe-platform.com/
G. C. H. E. de Croon, “Provable self-organizing no. 1, p. 439, Dec. 2017. content/journals/10.1075/is.5.2.05gau
pattern formation by a swarm of robots with [89] W. Zhu, M. Allwright, M. K. Heinrich, S. Oğuz, [106] I. D. Couzin, J. Krause, N. R. Franks, and
limited knowledge,” Swarm Intell., vol. 13, no. 1, A. L. Christensen, and M. Dorigo, “Formation S. A. Levin, “Effective leadership and
pp. 59–94, Mar. 2019. control of UAVs and mobile robots using decision-making in animal groups on the move,”
[73] L. Lei, R. Escobedo, C. Sire, and G. Theraulaz, self-organized communication topologies,” in Nature, vol. 433, no. 7025, pp. 513–516,
“Computational and robotic modeling reveal Proc. 12th Int. Conf. Swarm Intell. (ANTS) (Lecture Feb. 2005.
parsimonious combinations of interactions Notes in Computer Science), vol. 12421. Berlin, [107] A. Baronchelli, “The emergence of consensus:
between individuals in schooling fish,” PLOS Germany: Springer, 2020, pp. 306–314. A primer,” Roy. Soc. Open Sci., vol. 5, no. 2,
Comput. Biol., vol. 16, no. 3, Mar. 2020, [90] A. Attanasi et al., “Finite-size scaling as a way to Feb. 2018, Art. no. 172189.
Art. no. e1007194. probe near-criticality in natural swarms,” Phys. [108] G. Podevijn, R. O’Grady, Y. S. G. Nashed, and
[74] J. Krause, A. F. Winfield, and J.-L. Deneubourg, Rev. Lett., vol. 113, Dec. 2014, Art. no. 238102, M. Dorigo, “Gesturing at subswarms: Towards
“Interactive robots in experimental biology,” doi: 10.1103/PhysRevLett.113.238102. direct human control of robot swarms,” in Proc.
Trends Ecol. Evol., vol. 26, no. 7, pp. 369–375, [91] M. A. Muñoz, “Colloquium: Criticality and 14th Annu. Conf. Towards Auton. Robotic Syst.
2011. [Online]. Available: http://www. dynamical scaling in living systems,” Rev. Mod. (TAROS) (Lecture Notes in Computer Science),
sciencedirect.com/science/article/pii/S016953 Phys., vol. 90, Jul. 2018, Art. no. 031001, doi: vol. 8069, A. Natraj, S. Cameron, C. Melhuish,
4711000851 10.1103/RevModPhys.90.031001. and M. Witkowski, Eds., 2013, pp. 390–403.
[75] S. Mitri, S. Wischmann, D. Floreano, and L. Keller, [92] D. S. Calovi, U. Lopez, P. Schuhmacher, H. Chaté, [109] J. Nagi, A. Giusti, L. M. Gambardella, and
“Using robots to understand social behaviour,” C. Sire, and G. Theraulaz, “Collective response to G. A. Di Caro, “Human-swarm interaction using
Biol. Rev., vol. 88, no. 1, pp. 31–39, 2013, doi: perturbations in a data-driven fish school model,” spatial gestures,” in Proc. IEEE/RSJ Int. Conf.
10.1111/j.1469-185X.2012.00236.x. J. Roy. Soc. Interface, vol. 12, no. 104, 2015, Intell. Robots Syst., Sep. 2014, pp. 3834–3841.
[76] F. Bonnet et al., “Robots mediating interactions Art. no. 20141362. [Online]. Available: [110] L. Mondada, M. E. Karim, and F. Mondada,
between animals for interspecies collective https://royalsocietypublishing.org/doi/abs/ “Electroencephalography as implicit
behaviors,” Sci. Robot., vol. 4, no. 28, 2019, 10.1098/rsif.2014.1362 communication channel for proximal interaction
Art. no. eaau7897. [Online]. Available: https:// [93] L. Busoniu, R. Babuska, and B. De Schutter, between humans and robot swarms,” Swarm
robotics.sciencemag.org/content/4/28/eaau7897 “A comprehensive survey of multiagent Intell., vol. 10, no. 4, pp. 247–265, Dec. 2016.
[77] H. Xie et al., “Reconfigurable magnetic microrobot reinforcement learning,” IEEE Trans. Syst., Man, [111] G. Podevijn, R. O’Grady, N. Mathews, A. Gilles,
swarm: Multimode transformation, locomotion, Cybern. C, Appl. Rev., vol. 38, no. 2, pp. 156–172, C. Fantini-Hauwel, and M. Dorigo, “Investigating
and manipulation,” Sci. Robot., vol. 4, no. 28, Mar. 2008. the effect of increasing robot group sizes on the
2019, Art. no. eaav8006. [Online]. Available: [94] E. Price et al., “Deep neural network-based human psychophysiological state in the context of
https://robotics.sciencemag.org/content/4/28/ cooperative visual tracking through multiple human–swarm interaction,” Swarm Intell., vol. 10,
eaav8006 micro aerial vehicles,” IEEE Robot. Autom. Lett., no. 3, pp. 193–210, Sep. 2016.
[78] S. Kriegman, D. Blackiston, M. Levin, and vol. 3, no. 4, pp. 3193–3200, Oct. 2018. [112] G. Podevijn, R. O’Grady, C. Fantini-Hauwel, and
J. Bongard, “A scalable pipeline for designing [95] F. Magistri, D. Nardi, and V. Trianni, “Using prior M. Dorigo, “Human responses to stimuli produced
reconfigurable organisms,” Proc. Nat. Acad. Sci. information to improve crop/weed classification by robot swarms—The effect of the reality-gap on
USA, vol. 117, no. 4, pp. 1853–1859, 2020. by MAV swarms,” in Proc. 11th Int. Micro Air psychological state,” in Proc. 13th Int. Symp.
[Online]. Available: https://www.pnas. Vehicle Competition Conf., Madrid, Spain, 2019, Distrib. Auto. Robotic Syst. (Springer Proceedings
org/content/117/4/1853 pp. 67–75. in Advanced Robotics), vol. 6, R. Groß, A. Kolling,
[79] R. D. Leonardo et al., “Bacterial ratchet motors,” [96] E. R. Hunt and S. Hauert, “A checklist for safe S. Berman, E. Frazzoli, A. Martinoli, F. Matsuno,
Proc. Nat. Acad. Sci. USA, vol. 107, no. 21, robot swarms,” Nature Mach. Intell., vol. 2, no. 8, and M. Gauci, Eds. Cham, Switzerland: Springer,
pp. 9541–9545, May 2010. [Online]. Available: pp. 420–422, Aug. 2020. 2018, pp. 531–543.
https://www.pnas.org/content/107/21/9541 [97] F. Higgins, A. Tomlinson, and K. M. Martin, [113] C. Nam, P. Walker, H. Li, M. Lewis, and K. Sycara,
[80] K. E. Peyer, L. Zhang, and B. J. Nelson, “Survey on security challenges for swarm “Models of trust in human control of swarms with
“Bio-inspired magnetic swimming microrobots for robotics,” in Proc. 5th Int. Conf. Auton. Auton. varied levels of autonomy,” IEEE Trans. Human-
biomedical applications,” Nanoscale, vol. 5, no. 4, Syst., 2009, pp. 307–312. Mach. Syst., vol. 50, no. 3, pp. 194–204, Jun. 2020.
pp. 1259–1272, 2012. [Online]. Available: [98] E. C. Ferrer, T. Hardjono, M. Dorigo, and [114] M. Dorigo, G. Theraulaz, and V. Trianni,
http://dx.doi.org/10.1039/C2NR32554C A. S. Pentland, “Secure and secret cooperation in “Reflections on the future of swarm robotics,” Sci.
[81] I. Rabinowitch, M. Chatzigeorgiou, B. Zhao, robotic swarms,” 2019, arXiv:1904.09266. Robot., vol. 5, no. 49, Dec. 2020,
M. Treinin, and W. R. Schafer, “Rewiring neural [Online]. Available: https://arxiv.org/abs/ Art. no. eabe4385.
circuits by the insertion of ectopic electrical 1904.09266 [115] M. Dorigo and E. Sahin, Eds., “Special issue on
synapses in transgenic C. elegans,” Nature [99] V. Strobel, E. C. Ferrer, and M. Dorigo, “Blockchain swarm robotics,” Auton. Robots, vol. 17, nos. 2–3,
Commun., vol. 5, no. 1, 2014, Art. no. 4442. technology secures robot swarms: A comparison pp. 111–246, Sep. 2004.
[Online]. Available: https://app.dimensions.ai/ of consensus protocols and their resilience to [116] G.-Z. Yang et al., “The grand challenges of science
details/publication/pub.1050134945 and Byzantine robots,” Frontiers Robot. AI, vol. 7, robotics,” Sci. Robot., vol. 3, no. 14, 2018,
https://www.ncbi.nlm.nih.gov/pmc/articles/ p. 54, May 2020. Art. no. eaar7650. [Online]. Available: https://
PMC4109004 [100] A. Prorok and V. Kumar, “Towards differentially robotics.sciencemag.org/content/3/14/eaar7650
[82] M. Z. Miskin et al., “Electronically integrated, private aggregation of heterogeneous robots,” in [117] S. Kazadi, “Swarm engineering,”
mass-manufactured, microscopic robots,” Nature, Proc. 13th Int. Symp. Distrib. Auton. Robotic Syst., Ph.D. dissertation, California Inst. Technol.,
vol. 584, no. 7822, pp. 557–561, Aug. 2020. R. Groß, A. Kolling, S. Berman, E. Frazzoli, Pasadena, CA, USA, 2000.
[83] N. Kumar, H. Soni, S. Ramaswamy, and A. Martinoli, F. Matsuno, and M. Gauci, Eds. [118] M. Li et al., “Generating swarm solution classes
A. K. Sood, “Flocking at a distance in active Cham, Switzerland: Springer, 2018, pp. 587–601, using the Hamiltonian method of swarm design,”
granular matter,” Nature Commun., vol. 5, no. 1, doi: 10.1007/978-3-319-73008-0_41 in Proc. 9th Int. Conf. Agents Artif. Intell., vol. 2.
p. 4688, Dec. 2014. [101] V. Strobel, E. C. Ferrer, and M. Dorigo, “Managing Setúbal, Portugal: SciTePress, 2017, pp. 145–152.
[84] M. Bär, R. Großmann, S. Heidenreich, and Byzantine robots via blockchain technology in a [119] A. Renzaglia, L. Doitsidis, A. Martinelli, and
F. Peruani, “Self-propelled rods: Insights and swarm robotics collective decision making E. B. Kosmatopoulos, “Multi-robot

1164 P ROCEEDINGS OF THE IEEE | Vol. 109, No. 7, July 2021


Point of View

three-dimensional coverage of unknown areas,” of industrial machinery,” IEEE Robot. Autom. Mag., Int. J. Robot. Res., vol. 31, no. 6, pp. 753–773,
Int. J. Robot. Res., vol. 31, no. 6, pp. 738–752, vol. 16, no. 1, pp. 103–112, Mar. 2009. May 2012.
May 2012. [127] J. J. Dawkins, F. L. Crabbe, and D. Evangelista, [133] M. Le Goc, L. H. Kim, A. Parsaei, J.-D. Fekete,
[120] W. Hönig, J. A. Preiss, T. K. S. Kumar, “Deployment and flight operations of a large scale P. Dragicevic, and S. Follmer, “Zooids: Building
G. S. Sukhatme, and N. Ayanian, “Trajectory UAS combat swarm: Results from DARPA service blocks for swarm user interfaces,” in Proc. 29th
planning for quadrotor swarms,” IEEE Trans. academies swarm challenge,” in Proc. Int. Conf. Annu. Symp. User Interface Softw. Technol.
Robot., vol. 34, no. 4, pp. 856–869, Aug. 2018. Unmanned Aircr. Syst. (ICUAS), Jun. 2018, New York, NY, USA: Association for Computing
[121] W. Hönig, S. Kiesel, A. Tinka, J. W. Durham, and pp. 1271–1278. Machinery, Oct. 2016, p. 97.
N. Ayanian, “Persistent and robust execution of [128] G. C. Messenger, “Radiation hardening,” in [134] M. Alhafnawi, S. Hauert, and P. O’Dowd,
MAPF schedules in warehouses,” IEEE Robot. AccessScience. McGraw-Hill, 2020. [Online]. “Robotic canvas: Interactive painting onto
Autom. Lett., vol. 4, no. 2, pp. 1125–1131, Available: https://www.accessscience.com/ robot swarms,” in Proc. Conf. Artif. Life, 2020,
Apr. 2019. content/566850 pp. 163–170.
[122] R. R. Negenborn and J. M. Maestre, “Distributed [129] T. M. Roehr, F. Cordes, and F. Kirchner, [135] E. Tsykunov, R. Agishev, R. Ibrahimov,
model predictive control: An overview and “Reconfigurable integrated multirobot exploration L. Labazanova, A. Tleugazy, and D. Tsetserukou,
roadmap of future research opportunities,” IEEE system (RIMRES): Heterogeneous modular “SwarmTouch: Guiding a swarm of
Control Syst., vol. 34, no. 4, pp. 87–97, Aug. 2014. reconfigurable robots for space exploration,” micro-quadrotors with impedance control using a
[123] C. E. Luis, M. Vukosavljev, and A. P. Schoellig, J. Field Robot., vol. 31, no. 1, pp. 3–34, 2014. wearable tactile interface,” IEEE Trans. Haptics,
“Online trajectory generation with distributed [Online]. Available: https://onlinelibrary.wiley. vol. 12, no. 3, pp. 363–374, Jul. 2019.
model predictive control for multi-robot motion com/doi/abs/10.1002/rob.21477 [136] W. Hönig, C. Milanes, L. Scaria, T. Phan, M. Bolas,
planning,” IEEE Robot. Autom. Lett., vol. 5, no. 2, [130] C. Pinciroli et al., “Self-organizing and scalable and N. Ayanian, “Mixed reality for robotics,” in
pp. 604–611, Apr. 2020. shape formation for a swarm of pico satellites,” in Proc. IEEE/RSJ Int. Conf. Intell. Robots Syst.
[124] H. Hamann and A. Reina, “Scalability in Proc. NASA/ESA Conf. Adapt. Hardw. Syst., (IROS), Sep. 2015, pp. 5382–5387.
computing and robotics,” 2020, Los Alamitos, CA, USA: IEEE Computer Society [137] S. Jeon et al., “Magnetically actuated microrobots
arXiv:2006.04969. [Online]. Available: Press, Jun. 2008, pp. 57–61. as a platform for stem cell transplantation,” Sci.
http://arxiv.org/abs/2006.04969 [131] M. Waibel, B. Keays, and F. Augugliaro, “Drones Robot., vol. 4, no. 30, May 2019,
[125] D. Albani, T. Manoni, D. Nardi, and V. Trianni, shows: Creative potential and best practices,” Art. no. eaav4317.
“Dynamic UAV swarm deployment for Verity Studios, Zurich, Switzerland, White Paper, [138] X. Dong and M. Sitti, “Controlling
non-uniform coverage,” in Proc. 17th Int. Conf. Jan. 2017. two-dimensional collective formation and
Auton. Agents MultiAgent Syst. (AAMAS), 2018, [132] J. Alonso-Mora, A. Breitenmoser, M. Rufli, cooperative behavior of magnetic microrobot
pp. 523–531. R. Siegwart, and P. Beardsley, “Image and swarms,” Int. J. Robot. Res., vol. 39, no. 5, 2020,
[126] N. Correll and A. Martinoli, “Multirobot inspection animation display with multiple mobile robots,” Art. no. 027836492090310.

ABOUT THE AUTHORS


Marco Dorigo (Fellow, IEEE) received the in the study of collective animal behavior. He is also a pioneering
Laurea and M.Tech. degrees in industrial researcher in the field of swarm intelligence. He has uncovered
technologies engineering and the Ph.D. many biological principles that govern swarm intelligence in social
degree in electronic engineering from the insects and fish schools, and used them as a source of inspira-
Politecnico di Milano, Milan, Italy, in 1986 tion to design distributed algorithms in artificial systems. He has
and 1992, respectively. He received the coauthored five books, among which Swarm intelligence: From
title of Agrégé de l’Enseignement Supérieur Natural to Artificial Systems (Oxford University Press, 1999) and
from the Université Libre de Bruxelles (ULB), Self-Organization in Biological Systems (Princeton University Press,
Brussels, Belgium, in 1995. 2001) are considered as reference textbooks. His research focuses
From 1992 to 1993, he was a Research Fellow with the Interna- on the understanding of a broad spectrum of collective behaviors
tional Computer Science Institute, Berkeley, CA, USA. Since 1996, in animal societies and human groups by quantifying and then
he has been a tenured Researcher with the FNRS, the Belgian modeling the individual-level behaviors and interactions, thereby
National Funds for Scientific Research, Brussels, and a Co-Director elucidating the mechanisms generating the emergent, group-level
of IRIDIA, the Artificial Intelligence Laboratory of ULB. He is also the properties.
inventor of the ant colony optimization metaheuristic. His current Dr. Theraulaz was awarded the Bronze Medal of the CNRS
research interests include swarm intelligence, swarm robotics, and in 1996 and the Scientific Excellence Award of the CNRS
metaheuristics for discrete optimization. in 2010 and 2017.
Dr. Dorigo was a NATO-CNR Fellow in 1993. From 1994 to 1996,
he was a Marie Curie Fellow. He is also a Fellow of the Association
for the Advancement of Artificial Intelligence (AAAI) and European
Association for Artificial Intelligence (EurAI). He was awarded the
Italian Prize for Artificial Intelligence in 1996, the Marie Curie
Vito Trianni received the M.Sc. degree
Excellence Award in 2003, the Dr. A. De Leeuw-Damry-Bourlart
in computer science engineering from the
Award in Applied Sciences in 2005, the Cajastur International Prize
Politecnico di Milano, Milan, Italy, in 2000,
for Soft Computing in 2007, the ERC Advanced Grant in 2010,
the master’s degree in information and com-
the IEEE Frank Rosenblatt Award in 2015, and the IEEE Evolutionary
munication technology from CEFRIEL, Milan,
Computation Pioneer Award in 2016. He is the Editor-in-Chief of
in 2001, and the Ph.D. degree in applied sci-
Swarm Intelligence and an associate editor or a member of the
ences from the Université Libre de Bruxelles,
editorial board of many journals on computational intelligence and
Brussels, Belgium, in 2006.
adaptive systems.
He is currently a Senior Researcher with
the Institute of Cognitive Sciences and Technologies of the Italian
Guy Theraulaz received the Ph.D. degree National Research Council (ISTC-CNR), Rome, Italy. He pioneered
in neurosciences and ethology from the domain of evolutionary swarm robotics, that is, the synthesis of
Aix-Marseille University, Marseille, France, collective behaviors for robotic swarms through the exploitation of
in 1991, and the Habilitation degree in evolutionary robotics techniques. More recently, he is also conduct-
quantitative behavioral biology from Paul ing research on the analysis and design of large-scale decentralized
Sabatier University, Toulouse, France, systems, not limited to swarm robotics systems (e.g., cognitive
in 2002. radio networks and cyber–physical and sociotechnical systems). His
He is currently a Senior Research Fellow research mainly involves swarm intelligence and swarm robotics,
with the National Center for Scientific with a particular emphasis on the design and analysis of complex
Research (CNRS), Toulouse, France, where he is also an expert self-organizing systems and distributed cognitive processes.

Vol. 109, No. 7, July 2021 | P ROCEEDINGS OF THE IEEE 1165

You might also like