Received: 4 March 2017 Revised: 8 February 2018 Accepted: 15 February 2018

DOI: 10.1002/sys.21428

SPECIAL ISSUE ARTICLE

Applying autonomy to distributed satellite systems: Trends,

challenges, and future prospects

Carles Araguz1,2 Elisenda Bou-Balust2 Eduard Alarcón2

1 Nano-Satellite and Payload Laboratory, Tech-

nical University of Catalonia – UPC Barcelona- Abstract

Tech, Barcelona, Spain While monolithic satellite missions still pose significant advantages in terms of accuracy and
2 Department Electronics Engineering, Technical
operations, novel distributed architectures are promising improved flexibility, responsiveness,
University of Catalonia and adaptability to structural and functional changes. Large satellite swarms, opportunistic satel-
Correspondence lite networks or heterogeneous constellations hybridizing small-spacecraft nodes with high-
Carles Araguz, Nano-Satellite and Pay-
load Laboratory, Technical University
performance satellites are becoming feasible and advantageous alternatives requiring the adop-
of Catalonia – UPC BarcelonaTech, tion of new operation paradigms that enhance their autonomy. While autonomy is a notion that
Barcelona, Barcelona 08034, Spain is gaining acceptance in monolithic satellite missions, it can also be deemed an integral character-
Email: [email protected]
istic in Distributed Satellite Systems (DSS). In this context, this paper focuses on the motivations
Funding information
for system-level autonomy in DSS and justifies its need as an enabler of system qualities. Auton-
H2020 LEIT Space, Grant/Award Number:
687490 omy is also presented as a necessary feature to bring new distributed Earth observation func-
tions (which require coordination and collaboration mechanisms) and to allow for novel structural
functions (e.g., opportunistic coalitions, exchange of resources, or in-orbit data services). Mission
Planning and Scheduling (MPS) frameworks are then presented as a key component to implement
autonomous operations in satellite missions. An exhaustive knowledge classification explores the
design aspects of MPS for DSS, and conceptually groups them into: components and organiza-
tional paradigms; problem modeling and representation; optimization techniques and metaheuris-
tics; execution and runtime characteristics and the notions of tasks, resources, and constraints.
This paper concludes by proposing future strands of work devoted to study the trade-offs of
autonomy in large-scale, highly dynamic and heterogeneous networks through frameworks that
consider some of the limitations of small spacecraft technologies.

KEYWORDS
autonomy, distributed satellite systems, mission planning and scheduling

1 INTRODUCTION tions where monolithic satellites have become obsolete in terms of

risk, cost, or even performance.
While transforming our planet, climate change is calling for more accu- Next-generation satellite systems are envisioned as dynamically
rate, reliable, and frequent data to study meteorological phenomena, networked formations of heterogeneous satellites designed to satisfy
monitor and predict natural disasters, or estimate fauna migratory ambitious revisit times, cover large areas with higher resolutions, or
movements. At the same time, the ever increasing human exploitation minimize data access latencies. These novel architectures leverage
of natural resources, transportation services and massive agricultural on concepts and technologies that have been widely explored for
activities are demanding remotely sensed data to monitor deforesta- on-ground applications, namely, Wireless Sensor Networks, the
tion, study sea-ice coverage or watch crops for weather damage. Even Internet of Things, multicore computing, or nowadays' pervasive
for humanitarian aid, governments and NGOs are unveiling the need to cloud services. As a result, satellite nodes are not conceived as inde-
monitor oceans and conflict zones in near-real-time. As a result, some pendent instrument-centric components, but are rather expected to
satellite-based missions are facing an architectural change of paradigm provide data services to their infrastructure, to engage in collaborative
driven by the needs of these new Earth observation, security, and endeavors or even to wirelessly exchange resources among them. Cer-
surveillance applications. Distributed Satellite Systems (DSS) encom- tainly, some of these concepts are still technologically immature and
passing several interacting spacecraft are leading the way in applica- present challenges both at the implementation, design and operation

System Engineering. 2018;1–16. wileyonlinelibrary.com/journal/sys 

c 2018 Wiley Periodicals, Inc. 1
2 ARAGUZ ET AL .

levels. This technological and architectural landscape is being comple- (Sections 4.2, 4.3, and 4.6); the optimization approach and common
mented by the growing interest in small satellite platforms and their metaheuristics (Sections 4.4 and 4.6); the runtime and execution char-
adoption in new designs. Given their cost-effectiveness in launch and acteristics (Section 4.5); the notions of a task (Section 4.7); the com-
development, small-spacecraft could be technological enablers in the monly modeled resources and constraints (Section 4.8); and, finally,
deployment of practical DSS. With the advent of highly miniaturized some remarks on network characteristics (Section 4.9). Finally, the
systems and ultra-low-power technologies, mini-, micro- and paper concludes in Section 5 by emphasizing some of the less explored
nanosatellites have become suitable platforms to implement Earth scenarios in DSS and suggesting future studies related to the applica-
observation missions with multiple observing nodes. At the same tion of autonomy in this context.
time, the use of these technologies imposes limitations on space-
craft, mostly in their communication, computation, and attitude 2 DISTRIBUTED SATELLITE SYSTEMS
control capabilities. Nevertheless, audacious ventures are already
demonstrating their suitability (e.g., Planet Labs' constellation of 120 Usually conceived as complementary space assets, DSS are defined
hyperspectral nanosatellite imagers or Sky and Space Global's commu- as mission architectures consisting of multiple space elements that
nication constellation based on 3U CubeSat with intersatellite link interact, cooperate, and communicate with each other. The term
[ISL] capabilities). comprises many architectural approaches, namely, constellations
Designing systems with multiple interacting satellites, however, (Figure 1A), satellite trains (Figure 1B), clusters or formations (Fig-
could be deemed as a nonrevolutionary approach. Notwithstanding ure 2A), satellite swarms (Figure 2B), and innovative mission concepts
that systems with multiple interacting satellites such as Globalstar or such as fractionated spacecraft1 and Federated Satellite Systems2
Iridium have already been in orbit for two decades, the mission require- (Figures 3A and 3B, respectively). Their differences and detailed tax-
ments, functions, and intersatellite interactions suggested in novel DSS onomy can be found in Ref. 3. Because DSS are not only expected
concepts pose the need for novel operation concepts capable of han- to be structurally and functionally disaggregated but are envisioned
dling the complexity, heterogeneity, and dynamism of next-generation as interconnected systems-of-systems where the nodes cooperatively
architectures. In this context, this paper explores the need to enable achieve common goals, it should not be surprising to find classification
autonomous DSS missions in order to address the latter aspects. approaches concerned about node interdependencies and the infor-
Briefly presenting their characteristics and summarizing different mation exchanged between them. In an approach to study their funda-
instances of DSS in Section 2, this paper lists some relevant features mental commonalities and elaborate a comprehensive taxonomy, Lluch
that need be designed from an operations perspective (Section 2.1). and Golkar3 have classified them in terms of: spatial distribution of
After that, Section 3 discusses the need for autonomy; first in mono- their nodes (or the so-called degree of fractionation); the dynamic nature
lithic satellite missions and then emphasizing the need for autonomy of their network and physical structure (i.e., static, dynamic, oppor-
in DSS (Section 3.1). Section 4 devotes special efforts to scrutinize the tunistic); their orbit configuration; their intersatellite communication
design and runtime characteristics of Mission Planning and Schedul- and coordination capabilities; and, finally, the type of goals they ful-
ing (MPS) frameworks by exploring previous works and classifying fill (i.e., shared with the rest, individual). In spite of the relevance of
them with regards to: the system characteristics and organizational the above-mentioned taxonomy, such classification approach can be
paradigm (Section 4.1); their problem modeling and representation considered crucial to understand DSS because it addresses the many

(A) Satellite constellaon. Each satellite is in a (B) Satellite train. Several spacecra fol-
strategic orbit (usually to guarantee global cov- low the same trajectory (i.e. they are in
erage. Communicaon between the satellites is the same orbit). The satellites include all
scarce or not performed at all (Sat-Com. are the necessary subsystems but their pay-
an excepon, such as Iridium). Each satellite loads are usually different to allow measure-
is composed of all the necessary subsystems, in- ments of mulple magnitudes with negligi-
cluding their payload. The size of the spacecra ble delays between them.
tends to be big (∼ 700–800 kg.)

FIGURE 1 Distributed satellite systems taxonomy (I): constellation and train

ARAGUZ ET AL . 3

(A) Cluster. The spacecra are maintaing (B) Swarm. A cloud of several satellites (e.g.
ght flight formaon. Therefore, they need 10, 50, 100...) Each of the satellites is in-
to communicate and exchange, at least, a- dependent: it has its own goals and control
tude control and orbital state. mechanisms. Nonetheless, they may eventu-
ally comunicate with others in order to avoid
collisions or engage in coordinated datakes.

FIGURE 2 Distributed satellite systems taxonomy (II): swarm and cluster

er
pow un
da pro
ta ces
se processed
d
data
data

(A) Fraconated spacecra.The satellite is (B) Federated Satellite System. Composed of

divided in several physically-detached mod- a network of satellites formed opportuniscally.
ules. Spacecra funconalies are dis- Satellites may engage in cooperave measure-
agregated on physically-detached modules. ments, foreign data relay and processing acv-
Each module is specialized in one or several ies or similar collaboraons. Nonetheless, each
funcons (e.g. energy, instrument, com- satellite includes all their necessary subsystems
municaons...) Data and power are shared and their own payload: when they are not part
wirelessly in close formaon. of the FSS, they have their own mission goals.

FIGURE 3 Distributed satellite systems taxonomy (III): fractionated spacecraft and Federated Satellite System (FSS)

different characteristics and qualities of DSS architectures and pro- 2.1.1 Resilient, flexible, and highly adaptable systems
vides a broad view to their fundamental design challenges.
Owing to their distributed nature, DSS may represent an advantageous
alternative in terms of robustness and system availability. Decentral-

2.1 New mission functions and requirements izing their functionalities can eliminate single points of failure and
enable the possibility to provide acceptable levels of service upon
Some practical examples of DSS have already been demonstrated, node failures. If the necessary internal mechanisms to react to failures
namely: communications constellations (e.g., IRIDIUM, GlobalStar, are provided, the services and functions of the failing nodes could be
Orbcomm); global navigation satellite systems (e.g., GPS, Galileo, absorbed by the system and would endow architectures with enhanced
GLONASS, Beidou); trains4,5 ; and clusters.6,7 Nonetheless, practical resiliency.8 Similarly, their functional and structural segmentation also
examples of fully fledged DSS which demonstrate systemic improve- allows to replace or add new nodes, either when their on-board tech-
ments and bring new functionalities have yet not been reported. In nology becomes deprecated by new advances, when new functions
order to understand their design challenges and be able to address need to be added to the system or simply to overcome permanent
their operational requirements later in this paper, this section briefly failures. Their responsiveness and adaptability is also enabled by
discusses their expected system qualities and potential new functions. incremental developments or deployment strategies. In contrast to
4 ARAGUZ ET AL .

monolithic spacecraft, the financial outlay of distributed architectures a few. The latter are based upon Digital Elevation Modeling tech-
could also be split in several phases, making them affordable options niques, which are commonly implemented with interferometric syn-
when costs cannot be assumed all at once. By means of gradually thetic aperture radars (InSAR). By correlating images from two differ-
adding new nodes to a DSS, the requirements of its associated data ent satellite passes over the same area, Earth's topography and ground
products would be satisfied in a series of stages or with incremental surface deformations can be mapped. This type of measurement, gen-
performances. Adding new functionalities to an already deployed DSS erally performed in repeat-passes by a single satellite, could also be
could also complement their flexibility by allowing these systems to performed by several spacecraft in distributed architectures. In that
adapt to a changing context or user needs.8 Completing the set of qual- case, the time between consecutive acquisitions could be lower if the
ities shared with many other distributed systems, satellite missions delay between the two passes is shorter than that of a repeat-pass.
with several failing spacecraft would also degrade gracefully and might This would also improve the mission performance as it would mini-
still deliver data even when most of its nodes would not be functional. mize the likelihood of different atmospheric conditions (i.e., humidity),
which are known to be strong limiting factor for this technology.12

2.1.2 Resource exchange and collaboration While InSAR techniques require extremely accurate orbit determina-
tion and precise pointing, stereo photogrammetry is much more flex-
The literature describing fractionated spacecraft and Federated
ible and could also be accomplished by DSS with less accurate point-
Satellite Systems (FSS) frequently discusses the exchange of resources
ing capabilities. As a matter of fact, stereo photogrammetry could
as an essential function in these DSS approaches.2,9 In either of the
even be achieved on an opportunistic manner. Lower resolution eleva-
two propositions, satellite networks or satellite fractions are expected
tion maps can be obtained by combining optical imagery taken from
to exchange data through ISL to either coordinate among themselves
different viewing angles, as in Tack et al.13 Although photogramme-
or to provide services to such a distributed infrastructure. This fact
try techniques may require more than two images to converge, this
would also be true for satellite swarms or constellations where nodes
type of measurement could be performed by coordinating different
interact to one another. In-orbit data services could comprehend data
satellite nodes and could also improve revisit times. Finally, capturing
processing (e.g., compression, analysis, on-board data fusion) temporal
data from different viewing angles is also essential in mapping appli-
storage, transportation to other points of the network, or the data
cations that rely upon Bidirectional Reflectance Distribution Function
relay to ground. FSS concepts envision the exploitation of underuti-
(BDRF) data products. BDRF needs to be estimated in order to cap-
lized spaceborne commodities (e.g., processing power, or bandwidth)
ture the multiangular reflectance of opaque surfaces. This measure-
as a means to increase the value of already flying assets. This translates
ment allows to study land surface albedo,14 one of the most important
to the ability to collaborate with other systems by communicating with
parameters characterizing the Earth's radiative regime and its impact
them and supplying or consuming some of their on-board resources.
on biospheric and climatic processes. The design of distributed satellite
Even the bolder capability of wirelessly exchanging power is seen as a
missions that provide this functionality has been recently explored in
potential internal function of some DSS implementations.10
Refs. 15, 16 where the authors proposed to address their design with
formations of 6U-CubeSat and proved their feasibility and potential
2.1.3 Distributed Earth observation performance improvement with respect to monolithic alternatives.
Aside from the previous qualitative attributes and structural func-
tions, DSS open the door to alternative observation approaches that 2.1.4 On-board triggering of new observational requests
could improve mission performances. On the one hand, simultaneous in DSS
acquisition of data from multiple geographical locations would improve
Enabling on-board data processing may allow fast detection of areas
temporal resolutions (i.e., revisit times). In relation to this, architec-
of interest (e.g., geographical changes, natural disasters or even new
tures where nodes can download their data indirectly (e.g., through
human constructions). While this is true and has been done in mono-
ISL networks) could also reduce data-access latency and achieve
lithic missions (e.g., EO-117 and IPEX18 ) distributed systems designed
near-real-time monitoring of large, if not global, target areas. At the
to scan large target areas with wide-swath, coarse-resolution instru-
same time, increasing the number of observing satellites can also attain
ments could identify interest zones and trigger localized observational
wider coverage at high resolutions without the need of wider swaths.
requests. These secondary requests could be served by other nodes
A practical example of the latter is the large constellation of multi-
in the architecture, which could potentially embark higher resolution
spectral imagers PlanetScope, briefly introduced in Boshuizen et al.11
or specialized instruments as illustrated in Figure 4. This approach
Developed and operated by Planet, this constellation of about 120 3U
would efficiently use the resources of high-resolution spacecraft
nanosatellites will image the entire Earth every day with high spatial
and could provide high-quality data to users with shorter latencies,
resolutions once it enters normal imaging operation.
thus improving the system responsiveness in cases where having
In addition to the improvement in the temporal, spatial, and spec-
near-real-time updates is critical (e.g., disaster monitoring).
tral domain, DSS can also bring improved angular resolutions by cap-
turing data from multiple spacecraft, either simultaneously or with
shorter delays between the acquisitions. Several Earth observation 2.1.5 Based on small-spacecraft technologies
applications leverage on multiangular observations, such as hydrolog- Agencies, universities, and corporations have proved that micro- and
ical modeling, soil mapping, or the study of terrain stability, to name nanosatellites are cost-effective platforms that reduce development
ARAGUZ ET AL . 5

agile satellite However, autonomy per se is a concept positioned at the intersection

plaorm
of philosophical, scientific, and engineering interests. Autonomous sys-
tems are commonly defined as systems capable of generating their own
Change
detected laws or norms, but their modeling and design approaches are moti-
vated by different perspectives and interests, as so it is their ultimate
purpose.
Providing autonomous capabilities to monolithic spacecraft has
been a concept of growing interest for more than one decade. Known
Coarse resoluon as one of the first approaches toward autonomous operations, NASA's
measurement
Deep Space One (DS-1) included the Remote Agent eXperiment (RAX),
Interest zone a software system aimed at managing the spacecraft's resources and
activities.29 This mission triggered new developments that tackled
autonomous management at several different areas, including rover
navigation, Guidance Navigation and Control (GNC), monitoring and
Fault Detection, Isolation, and Recovery techniques (FDIR), signal
processing, and MPS. Traditionally, autonomous satellite technologies
FIGURE 4 Coordinated multipoint measurements. DSS missions
have been explored in order to address the following issues:
allow to have several coordinated spacecraft capturing data from the
same geographical regions. One of such approaches could be the one
illustrated in this figure, where one spacecraft with a wide instrument • Communication delays: Space exploration missions where the space-
swath, acquires images in low resolution and detects areas of interest craft is at huge distances from Earth (e.g., Mars exploration), present
that are later acquired by a secondary spacecraft with narrower swath
delays in the communication subsystem that precludes ground oper-
and better resolution
ators to manage the spacecraft with agility. Not being able to react to
unexpected situations may lead to catastrophic effects for the mis-
times and present moderate performances. Their technologies, capa- sion and hence has triggered the development of autonomous con-
bilities, and suitability has been explored in several relevant survey trollers that can detect and correct failures at subsystem and mis-
papers,19–22 concluding that these types of platforms are capable of sion level.30
delivering data with scientific value. Most authors have pointed out • Reduced visibility: Aligned with the previous issue, spacecraft orbiting
that despite power, size, and mass constraints, small spacecraft have in Low Earth Orbits (LEO) can only communicate with ground oper-
turned into extremely low-cost remote sensing platforms. Notwith- ators when they have visibility to ground stations. Therefore, there
standing the fact that many of the high-performance instruments (e.g., are large periods of time during which the spacecraft needs to oper-
SAR, lidar, high-resolution optical imagers, hyperspectral imagers, ate autonomously.
etc.) are generally not compatible with small spacecraft constraints,
• Improvement of science return: In addition to correcting faults,
they can address a broad variety of measurements with high societal
autonomous satellites can present improved performances provided
and scientific return.19 In that sense, architectures based upon small
that they generate their own mission plans. NASA's Earth Observ-
spacecraft could provide very high temporal resolutions at moderately
ing One (EO-1) was able to autonomously detect and respond to
high spatial resolutions.
dynamic, scientifically interesting events observed from its LEO.17
Small spacecraft are key enablers for distributed architectures
By combining on-board data processing algorithms with its Con-
where multiple spacecraft are required to attain global coverage. They
tinuous Activity Scheduling Planning Execution and Replanning
have also been deemed essential when the system needs enhanced
(CASPER), EO-1 was able to plan observation activities focusing on
spatial, spectral or temporal resolutions at a reduced costs and
autonomously detected interesting regions.
risk.19,23–26 Innovative projects such as NASA's Cyclone Global Navi-
• Mission robustness and tolerance against failures: Reacting to faults or
gation Satellite System27 are examples of how agencies are currently
ensuring the consistency of the plan computed on ground is critical
changing their paradigm and are already exploring solutions based on
and has to be done without supervision (i.e., ensuring that the values
many small satellites to address new scientific needs. Simultaneously,
predicted with models are within the expected ranges, checking that
relevant initiatives like the European QB50 project, aimed at studying
no system constraint is violated, ...). Moreover, when a given incon-
the middle and lower thermosphere,28 are also exploring the feasibility
sistency is found, the satellite needs to recalculate the plan in order
of large, heterogeneous, CubeSat-based clusters.
not to miss any observation opportunity.

3 AUTONOMY A thorough review of autonomous monolithic satellite missions

can be read in Ref. 31, where the authors explore autonomy as an on-
Systems with ever increasing degrees of autonomy are more preva- board functionality that can address both intelligent sensing, mission
lent and important today than ever before (e.g., unmanned aerial vehi- planning and execution, fault management, and distributed decision
cles, autonomous underwater vehicles, smart buildings, smart cars, ...). making. Their work suggests to take the definition of autonomy levels
6 ARAGUZ ET AL .

proposed in the European standard ECSS-E-ST-70-11C32 as a starting Therefore, autonomy is not conceived as an additional feature to
reference, and refer to the recent progress in FDIR, On-Board Control implement but rather as the solution to operate such dynamic and
Procedures (OBCP) in European missions, and autonomous spacecraft complex systems; to achieve the envisioned high-level qualities and
reconfiguration schemes based on Markov Decision Processes. to allow new functions. Coping with their operability with traditional
approaches could be considered impractical. Computing static plans
of action on-ground and then distributing time-tagged commands to
3.1 Autonomy in DSS
spacecraft may preclude the expected flexibility of DSS. Even from
Besides the benefits studied for monolithic satellite architectures, the point of view of some of their required functions, such as flight
autonomy is also considered a key enabler for distributed missions. As formation, autonomy is deemed an essential characteristic that, espe-
suggested in Refs. 31, 33 autonomy is key to operate spacecraft under cially for large-scale systems, may not be addressable with traditional
uncertainty and to cope with unexpected situations. Vamvoudakis approaches. Even responsive OBCP may be functionally too restric-
et al.34 also stated that full autonomy enables mission tailoring, con- tive to achieve control of a large-scale distributed architecture in which
trol reconfigurability to allow for safe recovery, improved responsive- nodes should interact to one another.
ness and agility, and a general adaptability to changing environmental In this scenario, autonomy in DSS is being studied under many
conditions.34 Nevertheless, autonomy in DSS goes beyond local man- perspectives. One of such perspectives understands autonomy as
agement of failures and generation of plans of action and needs be an integral characteristic of a DSS, like in the Autonomous Nano-
rather seen as an enabler for the system qualities and function dis- Technology Swarm (ANTS) project by NASA.38 With ANTS, NASA is
cussed in Section 2.1. In this sense, Iacopino and Palmer35 enumer- investigating the application of swarm intelligence and decentralized
ated up to seven motivations to autonomy that can help to visualize computing techniques to provide self-configuration, self-optimization,
the broad scope of this matter.35 While some of their motivations are self-protecting, and self-healing capabilities to their architecture of
very much related to the issues mentioned above, the rest describe pico-satellites.39 A different view is that of the achievement of
additional values and qualities that have been scarcely tackled in other autonomous functions, such as flight formation. Both for intersatel-
works, namely: decrease of operational costs; optimal management of lite communications (e.g., optical links) and remote sensing techniques
resources to reduce inactivity periods; responsiveness; flexibility and (e.g., bistatic radars), being able to maintain virtual flight structures is a
adaptability. critical enabler for DSS. Some formations have been able to automati-
This comprehensive notion of system-level autonomy is central cally maintain and correct their separation and orbits,40 while previous
to achieve the envisioned architectural functions of DSS, and needs missions like PRISMA41 or the upcoming PROBA-342 have presented
to encompass: the adaptation to context and internal changes; dis- important progresses in the area of autonomous structure build up.
tributed coordination and decentralized decision making; trading or At the same time, some authors have addressed satellite systems that
exchange of resources; and optimal management of infrastructure manage and exchange their resources, as in Ref. 43. This work delved
capacities. Moreover, it is critical that the design of autonomous dis- into the problem of resource exchange and collaboration between sys-
tributed satellite missions does not disregard two important notions of tems in FSS, an interesting topic that can be extended to multiple DSS
DSS, namely: their dynamic nature and their complexity. The dynamism instances. Such exchange of resources (e.g., power, data processing,
not only refers to internal failures, but also tries to describe funda- downlink capacity) goes beyond the technological capability and com-
mental features of these adaptive architectures, such as incremental prehends the necessary coordination mechanisms to do so. While cru-
deployments, replacement of nodes, or the creation of opportunistic cial to enable system-level autonomous operations, these mechanisms
coalitions. On the other hand, DSS can also be regarded as complex foster the agreement among spacecraft involved in resource transfers
systems-of-systems by virtue of three dominating factors: heterogene- (i.e., the service consumer and producer) and coordinate the appropri-
ity; their potentially large dimensionality; and the influence of inter- ate maneuvers to interface them. The coordination mechanisms are
mittent interactions characterized by orbital trajectories. DSS hetero- very much intertwined with the networking characteristics of DSS. The
geneity can be understood in many different ways. Some DSS studies literature in ISL and Intersatellite Networks is extensive and has been
are considering systems that hybridize traditional satellite platforms looking at several aspects, such as: the limitations of small satellite
with small spacecraft technologies.26 Others, are implicitly asserting platforms44 ; the adoption of ground Internet standards for space45 ;
the design of heterogeneous systems either by proposing opportunis- the design of new network protocols for specific DSS architectures46 ;
tic coalitions among existing systems2 or by considering segmented or the exploration of delay-tolerant network concepts in Earth-imaging
nodes in which each part implements a different functionality (i.e., frac- constellations.47
tionated spacecraft). Similarly, other approaches are also considering Many networking characteristics may influence the design of coor-
constellations with multiple specialized satellites (i.e., with different dination mechanisms in DSS, since these are the foundations of the
functions or instruments) which orbit at different planes or orbital decentralized interactions envisioned in many cases.
altitudes.36 Furthermore, interactions between two nodes in a dis- Likewise, the design of autonomous DSS, needs to cope with the
tributed satellite mission heavily depend upon their orbital parame- technological constraints in small spacecraft technologies. Among
ters. So much so that some task allocation approaches are even tak- their limitations, the impaired or extremely limited communications
ing advantage of their intermittent and predictable communication capabilities results in the most critical one for the context of this paper.
processes.37 Albeit recent studies have been able to develop optical downlinks for
ARAGUZ ET AL . 7

small satellites,48 the ISL capabilities of these platforms still require been applied to satellite missions and what are the necessary ingredi-
many advancements to allow the deployment of large and intercon- ents to ultimately achieve fully autonomous DSS.
nected swarms or constellations. In contrast, the use of off-the-shelf
components (COTS) in their development may confer them with com- 3.3 From autonomous systems to MPS
putational capabilities slightly above their traditional counterparts.
Either autonomous or orchestrated by human operators, decision
However, it is also true that their reduced power generation could pre-
making in satellite missions has been implemented and performed in
clude them from processing large volumes of data or executing of other
Mission Planning and Scheduling systems. Satellite MPS software
algorithms that require high computing power.
transform high-level requests and system constraints into a set of com-
In summary, the need for autonomy in DSS is driven by the fac-
mands that satisfy the former without violating the latter. The litera-
tors presented herein. Autonomy is indeed essential to cope with prob-
ture is abundant with examples of MPS that present different design
lems that are present in monolithic satellites; to improve performance
and execution characteristics. Still, all of them serve the same purpose:
and tolerance to failures. Nonetheless, DSS are dynamic architectures
allocate tasks and control the actions of the spacecraft. In line with pre-
where components with different capabilities and capacities interact
vious works,31 the authors of this paper propose that autonomous rea-
to achieve global goals. Their operation must be reliable and their
soning, node coordination, and planning of activities has to be designed
nodes need to optimally manage shared resources. Autonomy is hence
as part of the mission planning framework.
seen as the cornerstone of their operability and an enabler for many of
Despite the availability of commercial MPS solutions dedicated to
their qualities and proposed functions.
Earth observation constellations, swarms, or clusters, these products
are not convenient for fully distributed satellite missions.63 Current
MPS designs provide automated scheduling and plan execution mech-
3.2 Designing autonomous systems anisms that satisfy the requirements of traditional missions (i.e., allo-
cate tasks and control system constraints). However, only a few of
One common notion in the design of autonomous system is to under-
them has been designed to exploit some of the qualities of a DSS (e.g.,
stand them as systems capable of reasoning, deciding and executing
Refs. 63–68).
their activities without human intervention, in pursuit of a given set
of goals set by operators. In many cases, this very concept has led
to describing the system in terms of a range of behaviors associated
4 MISSION PLANNING SYSTEMS
with its agents: their ability to sense the world, maintain their state
and make decisions about their actions.33 Modeling autonomous sys-
4.1 System, architectural, and design characteristics
tems as Multiagent Systems (MAS) is a common trend in many indus-
trial and academic works (including aerospace systems and satellite The noticeable growth in the number of Earth Observation missions
missions) and is applied when problems are either too large or when has lead to the development of many Mission Planning Systems.
the information sources are spatially distributed. MAS frameworks can In the literature, it is easy to find both commercial products (e.g.,
provide performance improvements to the system and allow complex GMV's flexplan,69 DEIMOS's plan4EO70 ), and a myriad of academic
functions.49,50 approaches. While the former tend to include sophisticated models,
In line with this, works in artificial intelligence have studied and detailed mission-specific constraints, security policies, and multiuser
implemented bioinspired MAS in which the entities present complex access methods, the latter tend to focus on the design aspects, the algo-
collective behaviours (e.g., self-organization) but where none of the rithms employed, and their performance. Because the literature is so
entities was explicitly programmed or designed for such purpose.51 abundant but at the same time so diverse, this section will start by iden-
Certainly, among all species of the animal kingdom, the behaviors tifying and categorizing their similarities from a system/mission per-
of some insect families have fruitfully given birth to many works in spective and from a software architecture design point of view.
the areas of swarm intelligence and collective organization. Collec-
tive behaviors are seen to emerge from direct or indirect interac- 4.1.1 Components and architecture
tions among entities and usually involve no sophisticated cognitive Whether executed on-board or on-ground, among the components
processes.52,53 Moreover, self-organized control inspired by biological of an MPS, it is very common to find a clear separation between the
systems has been receiving considerable attentions as a promising con- planning algorithm, the models that capture the dynamics of the sys-
cept for realizing robustness, scalability, and adaptability.54 Although tem and the constraint checkers that ensure that there is no violation
entities' behaviors can be simple, it is through a series of nonlinear of safety conditions or resources (e.g., Refs. 17, 64, 71, 72). Surrey
interactions that their behavior outcomes are accumulated and scaled- Space Centre's NEAT algorithm, for instance, divides the planning
up so that some collective behaviors or patterns emerge.∗ As a result, software between an heuristic component that generates schedules
approaches that mimic some of their characteristics and mechanisms with a decoupled resource allocator that validates the correctness of
(e.g., stigmergy) have been exhaustively applied in robotics (e.g., Refs. the generated plan of action and feeds the heuristic back with a fitness
55–60), task allocation61 or optimization62 and can also be adopted value.72 In some cases, such as the scheduling software for EO-1,17
to implement autonomous systems. The following sections will present the MPS also encompasses on-board data processing algorithms that
how some of these concepts and autonomous system paradigms have analyze scientific data and provide inputs to the scheduler.
8 ARAGUZ ET AL .

4.1.2 MPS for distributed missions Tripp and Palmer68 presented a mission planning scheme grounded

Since the appearance of the first Earth-observing constellations and on stigmergy.68 In this simulation-based approach, the satellites were

distributed missions, researchers have focused on specialized MPS precluded of any direct communication capability (i.e., ISL) and were

targeted for DSS.36,65,66,68,69,71,73–75 Two remarkable examples of forced to self-organize through indirect interactions. This lack of

those are the MPS for the Sentinel-S1 constellation69 and the one capability was owed to and justified by the inherent technological

for TerraSAR-X/TanDEM-X66 . Although these satellite missions only limitations of nanosatellites. A similar proposal was published later

encompass two spacecraft, the level of complexity in their models and by Iacopino et al.71 In their work, the authors presented a mission

constraints make them a practical illustration of automated distributed planning system based on the Ant Colony Optimization metaheuristic,

satellite operations. In addition, the fact that these two missions have inspired by pheromone trails left by ants and termites and extensively

been launched in the past and are operating in nominal conditions also employed in other optimization contexts. The proposal is instantiated

demonstrates the feasibility of autonomous missions at some degree. in real scenarios that take the Disaster Monitoring Constellation as

One notable feature of the TerraSAR-X/TanDEM-X planning system is case scenario. They represent the planning problem as a binary chain

the fact that it was designed to schedule thousands of observational where each task corresponds to the node of a graph (i.e., the colony)

requests per day, without human supervision. and satellites can choose whether to schedule the (potential) obser-

On the other hand, the planners presented in Refs. 36, 65, 68, 71, vation requests or not. Tasks are generated with the predicted ground

74, 75 have been designed for complex satellite systems encompass- tracks; each satellite has its own orbit and can propagate its position

ing, in most of them, tens or hundreds of heterogeneous satellites and determine which observation requests correspond to their future

in multiple configurations (i.e., GEO, LEO, several orbit planes, etc.). geographical location. This binary chain encoding not only allowed

Because of that, these works tend to simplify the modeling of subsys- them to solve the problem as an Ant Colony Optimization (ACO) but

tems and instruments. Instead of considering the modes of operation also provided organic coordination mechanisms: tasks that present

that would be needed to operate very-high-resolution imagers, or conflicts (e.g., because can be scheduled by two satellites) are modeled

complex attitude dynamics, these approaches focus on the complexity as shared nodes between two colonies. Furthermore, their encoding

of the architecture and define simplified arbitrary components. also allowed to define dynamic scenarios in which new observations
are inserted in the binary chain at any given moment, in order to
represent disaster events (e.g., fires, volcanic eruptions) or weather
4.1.3 Multiagent-based MPS
updates. While it is unclear whether their current approach could be
A recent study for the European Space Agency50 has explored and val-
executed on-board, the main drawback is that the binary encoding of
idated the benefits of MAS technologies in the context of Earth Obser-
observation requests and ACO algorithm does not account for other
vation constellations. The Distributed Agents for Autonomy study
structural functions in DSS (e.g., exchange of resources, multiangular
(DAFA) demonstrated how MAS-oriented frameworks can improve
observations).
several system qualities and presented negotiation-based self-
organizing mechanisms as means to improve the autonomous capabili-
ties of a system. Aligned with this vision, some works have addressed 4.2 Problem division
MPS for DSS that could be regarded as MAS.36,65,73–75 Whereas
Similarly to the architectural separation of components mentioned
traditional heuristic approaches are fine-tuned and optimized
above, some works try to reduce the computational complexity of the
for a known number of tasks and satellites, multiagent approaches
problem by dividing it into several subproblems or levels of definition.
allow for an adaptability and scalability that is not comparable.65 The
On the one hand, the approach presented in Ref. 76 has two sepa-
system proposed in Ref. 74 defines nine types of agents that allow to
rated modules with two different goals. The weekly module, executed
model as many elements of the architecture as the mission requires.
on ground, takes advantage of the computational resources of large
The authors have designed three cooperative agents that interact
servers and generates weekly schedules 3–5 days in advance. These
to find a solution to the scheduling problem: the satellite, request,
schedules are refined several times until they become available to the
and mesh agents. In addition, three active agents and three passive
mission control center. A second part of the problem is solved on-
agents allow them to model observational constraints (cloud coverage,
board, where the spacecraft executes a replacement scheduling algo-
solar ephemeris, ground stations) and the system resources (battery,
rithm. Much more constrained than the weekly phase, the second stage
attitude, memory), respectively. This configuration allowed to achieve
takes the weekly schedule as an input and replaces it with a more accu-
optimized results and to adapt to environmental circumstances and
rate solution with a time horizon of only 8 hours. This system, imple-
structural characteristics. Noteworthy, Ref. 73 presented an MPS
mented by researchers at NASA JPL, allowed to port their MPS to the
that was tested against scalability, showing reasonable results for a
IPEX CubeSat mission.77 On the other hand, Refs. 75, 78 also present
constellation of 300 CubeSat-like satellites.
algorithms that split the problem into many subproblems that are then
recombined. This very idea is also implemented in Refs. 36, 79 where
4.1.4 Bioinspired, self-organizing MPS the problem is divided into a global part and a local part. Global prob-
Aside from heuristic and MAS-based designs, a few MPS have been lems take into account the whole constellation and all the observation
influenced by swarm intelligence and collective organization schemes requests, while local problems are those limited to one spacecraft. In
and have been based upon bioinspired algorithms. On the one hand, both cases, at the time of solving the problem the global part of the
ARAGUZ ET AL . 9

algorithm is unaware of the actual state and resource capacities of the merit that feeds the sequencer in order to fully suit the mission needs.
spacecraft. The difference between these two approaches lays in the Similarly, the MPS for the Sentinel-S1 constellation has been designed
algorithm used in the global part. The global algorithm in Ref. 36 is exe- to optimize both the usage of the recording device and the ground sta-
cuted on ground with global, real-time information from all the obser- tion network usage69 .
vation requests and the environment. Once the observation requests Although there are very few planning systems not optimizing their
are distributed among the satellites in the constellation, each satellite outcomes, some MPS simply perform task distribution and try to opti-
uses this plan as a recommendation and generates its own local sched- mize a given figure of merit with abstract constraints (e.g., minimize
ule taking into account its resources. On the other hand, the approach repeated tasks). This kind of optimization is the one observed in Refs.
presented in Ref. 79 is always executed in the space segment. A coordi- 68, 75 and in the ground-segment process of Damiani et al.36 Similarly,
nating satellite receives the set of observation requests and distributes negotiation-based algorithms in multiagent approaches, present opti-
all of them to the rest. Then, local schedulers provide n sorted subsolu- mization algorithms that are computed in a collective and decentral-
tions that are merged by the coordinating satellite with a combinatorial ized manner (e.g., Refs. 36, 65, 74).
optimization algorithm.80

4.3 Timeline representation 4.5 Execution and runtime characteristics

The modeling of a scheduling problem can be done in a vari- 4.5.1 On-board/on-ground

ety of ways. One of such ways is using timeline representations. One of the above-mentioned open questions when designing auton-
This approach is used in most mission planning systems for mono- omy systems for satellites is whether their intelligence has to be placed
lithic spacecraft64,76,78,81 as well as in some satellite constellation on-board or on-ground. Traditionally, deep space missions have been
missions66,71 .† Timeline-based modeling is grounded on the definition commanded with OBCP: automated programs or complex routines
of each system variable in the time domain (i.e., resources, instrument that are decomposed into time-tagged commands on-board, and which
modes, constraints, and activities). Thus, the plan generated by the implement specific control actions. These actions need be executed
scheduling algorithm is composed of several timelines the values of without human supervision and generally encompass some kind of
which are generated during the planning process through system mod- fault management to prevent loss of contact with the spacecraft.84
els that reflect their time dynamics as illustrated in Figure 5. This com- OBCP are the solution to critically long propagation delays in which
monality was identified by Chien et al. in a survey paper published human operators cannot operate their spacecraft reliably. For similar
in 2012 that was devoted to automated mission operations.18 There, reasons, one could argue that missions with long propagation delays
they explored the state-of-the-art in MPS and highlighted the com- would also be inclined to issue autonomous decisions on-board. While
mon types of constraints, resources, tasks, and dependencies. Inter- this could be true for these cases, the solution can be debatable for
estingly enough, task decomposition is commented as a common fea- Earth observation satellites that are frequently in contact with ground
ture in most MPS. However, the decomposition of high-level activities operators without delays.
into lower-level ones is referred to post- and preactivities that need to Recent developments30,68,77 leverage on the benefits exploited in
be performed when a given instrument or subsystem has to be oper- previous on-board systems36,75,76 and are designed to run in the space
ated (e.g., thermal stabilization of an instrument). Hierarchical Task segment. These systems take advantage of having real updated data
Networks (HTN) are a known and suitable method to represent and with which they can predict their schedule more accurately. On-board
model such task decompositions. Noteworthy, the very same concept MPS for DSS can utilize intersatellite communication capabilities (ISL)
could be applied for distributed applications where complex and syn- to perform truly decentralized and distributed mission planning, even
chronized activities describe collaborative tasks.‡ Furthermore, Chien without direct visibility with ground control centers. On the other
et al.18 also note that dependencies usually exist between actions and hand, they are limited by the inevitable computational and network
state variables (e.g., an SAR instrument should not be enabled until the constraints. These limitations need also be considered along their
satellite is in a safe attitude mode). While some of the matters dis- lack of real-time operational goals. The system has its own variables
cussed in their paper will be used as a reference in the following sec- updated in real-time and can react to internal and environmental
tions, the reader is directed to the survey for detailed analysis on the changes with agility. However if a change in mission goals occurs, an
functionalities and characteristics presented by timeline-based MPS. on-board system needs this information to be propagated to all the
nodes.
On the contrary, ground-based MPS66,69,71,73,74,81–83 can be exe-
4.4 Optimization
cuted in powerful computers but present a reduced responsiveness
Another aspect of MPS is whether they perform some kind of optimiza- for similar reasons. One could say that both on-board and on-ground
tion and which are the optimized objectives. The algorithms used in alternatives can be used to provide autonomy to the system, since its
Refs. 36, 69, 71, 72, 74, 81–83 all solve optimization problems, usually definition does not force the algorithm to be executed in any particu-
with multiple objectives. The Capacity Analysis and Mission Planning lar physical location. However, one could also argue that the adaptabil-
Tool developed by DEIMOS81 calculates the most optimum sequence ity claimed by MAS-based schemes is worsened as long as the system
of observations and allows the end-user to decide the type of figure of is executed in partially updated environments. The truth is that some
10 ARAGUZ ET AL .

Agent i Agent j Agent k

t t t
i j
TL 1 TL 1
t t
S y s t em s t at e

i j k
TL 2 TL 2 TL 2
t t t
i k
TL 3 TL 3
t t
τ τ τ

F I G U R E 5 Timeline representations model system states and variables and visualize them together with the timeline of the scheduled tasks (top
timelines). While some variables can be represented by integer values as in TL2 , others (e.g., instrument states) can only be represented with a
finite set of values (TL3 ). Timelines can also represent constraints, usually as maximum or minimum values or delays. Abstract constraints, such as
mutual exclusion, can also be modeled as a resource timeline the values of which are binary (e.g., TL1 ). MPS that control several spacecraft, would
necessitate to represent a set of timelines for each satellite (in the figure represented as Agents i, j, and k)

developments tend to divide the system in two parts: one executed in which the system can operate solely with a reactive algorithm that
on-ground and the other on-board.36,76 In our point of view, there is produces instantaneous solutions based on system rules. Given the
a clear trade-off between the advantages and drawbacks presented by runtime required by a deliberative algorithm, the system is constantly
both on-board, on-ground, and hybrid approaches. controlled by the reactive module and receives deliberations in an
asynchronous manner. Therefore, the deliberative layer only enhances
the solutions and acts as an advisor.
4.5.2 Reactive control, deliberative reasoning, and hybrid The mission planners presented in Refs. 66, 78 could also be
approaches considered hybrid in these terms. In these works, the deliberative
On-line mission planners are those which are constantly producing planner is executed periodically, allowing the algorithm to replan part
an output (i.e., the actions to be performed). Conversely, off-line of the spacecraft activities with updated system values. Such modes of
schedulers are executed during a certain amount of time and gen- operation normally present interleaved scheduling windows as shown
erate a plan of actions to be performed until a given future time in Figure 7.
horizon (Figure 6). A similar distinction can also be made between
schedulers that perform reactive control or deliberative reasoning about 4.6 Problem modeling, types of algorithms, and
the task to perform. While the first ones react to inputs from the metaheuristics
environment to perform the scheduling function (i.e., on-line), delib-
erative planners predict future states and plan the tasks accordingly In general, mission planning problems can be formulated as multidi-

(i.e., off-line). Traditionally, Mission Planning Systems have been mensional Knapsack problems like Equation 1a where X is a vector of

deliberative.66,78,81–83 The accuracy and feasibility of their solutions decision variables xi ∈ {0, 1}, i = 1, … , n that indicate if request i has

relies on the fidelity of their system models, which is proportional to been scheduled (xi = 1) or not (xi = 0). Equation 1b expresses arbitrary

their complexity. Because of that, some off-line MPS have reported resource constraints. Most schedulers have to deal with m resources,

to take up to 2 days of runtime to output a schedule.82 On-line yielding at least m constraints. In addition, scheduling constraints that

approaches,68,72 present shorter execution times to produce a valid express relative timing or ordering of activities can be added.18 Allen's

output but are unable to predict future states. This condition may interval algebra defines up to seven different time relations that could

prevent them from fully optimizing the operations, specially in a DSS be applied in different cases (Table 1). Many MPS explicitly formulate

context. the problems as a Knapsack36,71,83,85 which are, in fact, a subset of the

For this characteristic, hybrid approaches exist as well, yielding NP-Hard Traveling Salesman Problem (TSP).

better results and balancing the benefits of both approaches.17,36,71,78 ∑

n

Upon the detection of an inconsistency, a system failure or the arrival max. f(X) = wi xi (1a)
i=1
of new science data (i.e., reactive behavior), the planners described in
∑
n
Refs. 17, 36, 64, 71 activate a deliberative algorithm which recomputes subject to: ri xi ≤ c. (1b)
or modifies an existing schedule. The fact that the system is designed i=1

to modify an existing solution could be advantageous in terms of com-

putational costs, given that the algorithm does not start with the same Other characteristics in the way these problems are tackled are
initial conditions. An interesting approach is that of Beaumet et al.,64 the algorithms used to solve them. Dynamic Programming has been
ARAGUZ ET AL . 11

T w : scheduling window (only for off-line)

MPS
Off-line scheduling
execuon task j
(deliberave)
me task k

τf : soluon τs : start of the τh : scheduling me

found execuon horizon

On-line scheduling
task j
(reacve)
task k

MPS execuon me

me

τnow : current
mission me

F I G U R E 6 Whereas off-line (i.e., deliberative) systems allocate all tasks within a time window that is in the future, reactive systems constantly
determine which tasks have to be enabled. Combinations of both approaches can exist, in which the on-line execution simply corrects or refines a
previously generated schedule. These usually reduce the computational complexity of the reactive part, while improving the system reactiveness
and schedule uncertainly

τh : scheduling
horizon

T w : scheduling window n Schedule n

Schedule n + 1

Refined/updated T w : scheduling window n + 2 Schedule n + 2

schedule.

me

F I G U R E 7 There are cases of off-line MPS in which their scheduling windows are interleaved, yielding to valid parts of a previously generated
schedule to be recalculated. This approach tries to solve one of the main drawbacks of purely off-line/deliberative systems: because models are not
perfect and faults can occur, the later in time a task has been scheduled the more uncertainty is accumulated from all the previous actions. Thus,
when the scheduling window is fixed (e.g., a given number of satellite orbits), interleaved schedule windows allows to reduce the uncertainty levels
for late tasks

used in Earth observation missions36 as well as negotiation-based 4.7 Tasks

approaches.74,75 In Ref. 65, the authors implement a reinforcement
Tasks (or activities) are the basic inputs for Mission Planning Systems.
learning strategy to solve this problem, implementing it with neu-
In EO applications, tasks are usually mapped to observational requests.
ral networks encoded in chromosome. However, the most common
They are represented as objects which, upon their execution, consume
approach to solving these problems is to implement some kind of meta-
a given resource. As mentioned above, datatake activities can actually
heuristic or combination of metaheuristics. While Iacopino et al.71
be decomposed into several tasks or activities18 (linked with Allen's
use Ant Colony Optimization,86 combines ACO with a mechanism of
interval algebra relations, Table 1). Depending on the framework and
Local Search. Classical metaheuristics like Local Search17 or Greedy
mission, the tasks and their level of abstraction vary. Nonetheless, it is
algorithms64,66 are also common when producing suboptimal solutions
common to find tasks related to data acquisition, attitude control, or
is not critical. Tabu Search87 and Simulated Annealing88 have also been
data download (e.g., Refs. 17, 36, 72, 75, 81, 83). Some works actually
used for EO missions, although, above all methods, the most common
map each system state to a task, as in the DEIMOS-2 MPS,81 where the
ones are Genetic Algorithms.68,72,81–83,89
five high-level system modes of operation are mapped to tasks that can
12 ARAGUZ ET AL .

TA B L E 1 Allen's interval algebra relations

t2
Relaon Representaon t1
X before Y X X
Y
X
X equal Y Y
t1
X meets Y X Y

X
X overlaps Y Y

X during Y X
Y
X
X starts Y Y
X
X ends Y Y

t2
be repeated, namely: Sun-pointing for housekeeping, nadir-pointing,
image acquisition, orbit maintenance and control, and download.
F I G U R E 8 Networks implemented through Intersatellite Links
present a dynamic structure mostly subjugated to the orbits of the
4.8 Resources and constraints satellites

The diversity in the modeling of resources is almost as large as

the diversity of EO applications. Resources are abstract objects that
encapsulate a system constraint, usually represented as a capacity.
be found in the literature, visibility with the ground station is one
Researchers at NASA JPL analyzed them thoroughly18 and described
of the most common one. In addition, safety constraints are also
the following characteristics, which have been used in other works as
defined to prevent damage to the spacecraft subsystems (e.g., direct
well:72
Sun exposure,17 sensor mutual exposure90 ) to impose technologi-
cal restrictions (e.g., multiple simultaneous downloads to the same
1. Atomic resources: Tasks or activities that cannot overlap, will require
ground station90 ) or to reflect environmental conditions (e.g., cloud
resources with binary capacity. Thus if a task is scheduled for a
coverage,83 sunlight64 ).
given period of time, the atomic resource to which it is linked will
have capacity 0 and will prevent other tasks from being scheduled
at the same time. 4.9 Network considerations
2. Depletable resources: On the contrary, resources that can take an Very few Mission Planning Systems oriented to DSS consider network
integer capacity, will be those which can be consumed by more than topologies, the dynamism influenced by orbit trajectories or their lossy
one activity at a time (if and only if the sum of task consumptions nature (especially for missions that rely on small spacecraft technolo-
is lower or equal to the resource capacity). This resources are also gies). Figure 8 tries to emphasize the fact that node interactions (i.e.,
called “consumable” in Ref. 72 where their capacity Ci is formulated transfer of data, coordination messages, etc.) happen in an intermittent
as a normalized real number (0 < Ci ≤ 1). manner. In addition to that, different nodes in the constellation may
3. Intrinsic maximum/minimum values: Given that resources are, in be characterized by different ISL bandwidths, ranges, or even tech-
some cases, representing physical variables, some of them will nologies (i.e., optical, RF bands). Nevertheless, many of the consulted
present intrinsic restrictions that have to be modeled accordingly. references propose MPS that do not rely or consider these network
When modeling the resource energy, the designer has to take into features: they are executed on-ground and then send the solution to
account that the devices that store energy (e.g., batteries) can store the network of satellites. A notable exception is the one published in
up to a maximum amount of charge; when the state-of-charge of a Refs. 37, 75 These researchers from the French CNES and ONERA
battery reaches 100%, energy in excess will be lost. propose a cooperative framework based on a negotiation approach
that actually takes advantage of the fact that orbits can be considered
Typically, satellite MPS encompass resource models to represent quasi-periodic. With this primitive assumption, they designed a decen-
power, energy and memory (data storage) (e.g., Refs. 17, 36, 74, 75, 81, tralized, on-board planning scheme where satellites predict future
83, 90). Some missions have been seen to include complex models to interactions based on orbit period information. At every satellite
represent thermal90 or attitude74,81 as a resource. encounter, spacecraft exchange information in order to allow the nego-
Finally, most missions include temporal and spatial constraints tiation scheme to converge to a feasible planning solution. They imple-
to reduce the solution space of the problem and represent regions mented an epidemic intersatellite communication protocol (known to
of unavailability. Complex operations constraints can include: instru- propagate a given information unit in O(log n) communication rounds)
ment and subsystem timing and synchronization; thermal; power; data to allow for indirect communication between the satellites. Their
volume; antenna visibility (i.e., line-of-sight and/or distance); space- simulations show how, on average, a satellite constellation with three
craft pointing; and priority. Among the common constraints that can satellites is able to absorb more than 70% of the requests in less than
ARAGUZ ET AL . 13

48 hours (with the constellation initialized with 150 requests and two to be one of the commonalities of negotiation-based and bioinspired
blocks of 200 requests triggered during the simulation). Nonetheless, solutions. Similarly, this classification has also attempted to distinguish
their work does neither analyze the network overhead nor show the segmentation between on-board and on-ground mission planning
the actual convergence time for this algorithm, making it difficult designs. Moreover, the paper has also explored many of the meta-
to evaluate the suitability for demanding contexts requiring short heuristic families used in the solving of DSS scheduling problems when
revisit times or deadlines. Moreover, despite centralized, the algorithm these are formulated as a multidimensional Knapsack problem.
they presented assumes a fixed and known constellation, precluding Albeit some works have addressed autonomous operations in
dynamic or opportunistic concepts such as FSS. regards to flight-formation, failure detection and mitigation, or the
Novel approaches such as the one in Ref. 68 propose hierarchical generation of observational requests (through processing of on-board
networks of indirectly communicated satellites in order to provide a data), achieving autonomous operations at all levels is still an open
scale-free solution to the planning problem in huge constellations or field of research. Several fundamental questions remain open and
satellite swarms. This idea, however, was not explored in their pub- should be addressed in future works. Undoubtedly thought-provoking,
lished work and does not seem to be developed in further publications one of such questions continues to revolve around what is the most
of the authors. appropriate organizational paradigm to implement autonomous DSS
Finally, it is interesting to mention that the MPS presented in Ref. (i.e., multiagent-based, bioinspired, based on heuristics and optimiza-
36 was validated for a complex network consisting of several ground tion techniques, based on negotiation algorithms, hybrid constitution).
stations, three GEO satellites to provide data rely capabilities and 12 Related to that, future studies may also inquire further into the actual
LEO satellites performing Earth Observation datatakes. Interestingly, trade-offs of decentralized on-board decision making in contrast to
despite dating from 2005, this work seems to capture the network centralized on-ground alternatives. Furthermore, attaining the system
complexity of future DSS, at least in a very primitive way (i.e., small con- qualities of a DSS through autonomous operations has still not been
stellation, unconstrained ISL, etc.). demonstrated. Future research is expected to tackle these aspects,
which may also comprehend the ability to provide in-orbit data ser-
vices or, more generally put, the exchange of on-board resources.
5 CONCLUSIONS Assessing the impact of small spacecraft technologies and their lim-
itations in regards to autonomous operations is equally pivotal. In
This paper has tried to lay out the potential benefits of autonomy in particular, new satellite coordination mechanisms need to circumvent
DSS by revisiting and structuring its traditional motivations and the the constraints enforced by the low data rate ISL inherent to small
current prospects. Autonomy is presented as an integral characteris- satellite designs. Ultimately, it is important to highlight the need to
tic of DSS that could enable coordination of spacecraft in distributed put all these questions in the context of disruptive DSS designs: large-
Earth observation endeavors. Particularly, the need for autonomous scale, dynamic networks in which heterogeneous satellites coopera-
DSS has been discussed in the context of innovative space systems; tively perform distributed Earth observation.
those that foresee reusable, increasingly complex and large-scale sys-
tems, composed of interconnected spacecraft with heterogeneous
capabilities. As such, autonomy is not only targeted at solving task ACKNOWLEDGMENTS

scheduling problems but rather at becoming an all-pervasive opera- The work presented in this paper has been carried out in the frame of
tional scheme that can contribute to the achievement of many sys- the Operational Network of Individual Observation Nodes (ONION)
tem qualities. In line with this, the paper highlighted that operating project, and has received funding from the European Union's Hori-
DSS autonomously could improve responsiveness of the system, and zon 2020 research and innovation program under grant agreement
even bring about other qualities like resiliency, adaptability, and in- 687490, coordinated by Thales Alenia Space France.
orbit exchange of resources.
Autonomous operations can be designed as part of next-generation
MPS frameworks. Accordingly, the paper devoted several sections ENDNOTES
∗ The reader is directed to Ref. 51 for a comprehensive classification of col-
to present an exhaustive knowledge classification that has aimed
lective behaviors as well as detailed study on the characteristics of self-
at identifying the architectural, modeling, and execution character-
organization in biological systems.
istics of current MPS solutions. Among the explored characteristics
† Their task chains are described by the authors as a discrete timeline repre-
of MPS, several coordination and organization approaches toward
sentation.
increasing autonomy have been presented, namely: bioinspired and ‡
It should be possible to include such type of interactions in the binary
self-organizing systems; MAS-based designs; and MPS that generate chains used in Iacopino et al.71
mission plans in batch. Problem division, the type and modeling of
resources, and the timeline representation have been presented as
some of the design traits of multiple MPS. In addition, their runtime
characteristics have been classified into deliberative, reactive, and ORCID

hybrid approaches. With the former being common for MPS that gen- Carles Araguz http://orcid.org/0000-0001-6305-811X
erate mission plans in batch, reactive decision making has been found Elisenda Bou-Balust http://orcid.org/0000-0002-5853-5900
14 ARAGUZ ET AL .

REFERENCES 21. Bouwmeester J, Guo J. Survey of worldwide pico-and nanosatel-

1. Brown O, Eremenko P. Application of value-centric design to space lite missions, distributions and subsystem technology. Acta Astronaut.
architectures: the case of fractionated spacecraft. AIAA SPACE Confer- 2010;67:854–862.
ences and Exposition; 2008. 22. Poghosyan A, Golkar A. CubeSat evolution: analyzing CubeSat capa-
2. Golkar A, Lluch I. The Federated Satellite Systems paradigm: concept bilities for conducting science missions. Prog Aerosp Sci. 2017;88:59–
and business case evaluation. Acta Astronaut. 2015;111:230–248. 83.

3. Lluch I, Golkar A. The value proposition of novel distributed space sys- 23. Nag S, LeMoigne J, de Weck O. Cost and risk analysis of small satellite
tems: a literature survey on Ilities. Proceedings of 3rd Federated Satellite constellations for earth observation. 2014 IEEE Aerospace Conference;
Systems Workshop, 2015. 2014:1–16.
4. Graziano MD. Overview of distributed missions. In: D'Errico M, ed. Dis- 24. Sandau R. Status and trends of small satellite missions for Earth obser-
tributed Space Missions for Earth System Monitoring, vol. 31. New vation. Acta Astronaut. 2010;66:1–12.
York, NY: Springer Science & Business Media; 2012:375–386.
25. Barnhart DJ, Vladimirova T, Sweeting MN. Very-small-satellite design
5. Schoeberl MR. The afternoon constellation: a formation of Earth for distributed space missions. J Spacecr Rockets 2007;44:1294–
observing systems for the atmosphere and hydrosphere. Proceedings of 1306.
the IEEE International Geoscience and Remote Sensing Symposium, vol. 1;
2002:354–356. 26. Araguz C, Llaveria D, Lancheros E, et al. Optimized model-based
design space exploration of distributed multi-orbit multi-platform
6. Zink M, Krieger G, Fiedler H, Hajnsek I, Moreira A. The TanDEM-X mis-
Earth observation spacecraft architectures. 2018 IEEE Aerospace Con-
sion concept. Proceedings of the 7th European Conference on Synthetic
ference; 2018.
Aperture Radar; 2008:1–4.
27. Rose R, Ruf C, Rose D, Brummitt M, Ridley A. The CYGNSS flight
7. Muñoz S, Hornbuckle RW, Lightsey EG. FASTRAC early flight results. J
segment; A major NASA science mission enabled by micro-satellite
Small Satellites 2012;1:49–61.
technology. Proceedings of the 2013 IEEE Aerospace Conference; 2013:
8. Hastings D, La Tour P, Putbrese BL. Value-driven analysis of new 1–13.
paradigms in space architectures: an Ilities-based approach. AIAA
28. Gill E, Sundaramoorthy P, Bouwmeester J, Zandbergen B, Reinhard R.
SPACE Conferences and Exposition; 2014.
Formation flying within a constellation of nano-satellites: the QB50
9. Brown O, Eremenko P. The value proposition for fractionated space mission. Acta Astronaut. 2013;82:110–117.
architectures. SPACE Conferences and Exposition; 2006:1–22.
29. Gamble EB Jr, Simmons R. The impact of autonomy technology
10. Porter AK, Alinger DJ, Sedwick RJ, et al. Demonstration of electromag- on spacecraft software architecture: A case study. IEEE Intell Syst.
netic formation flight and wireless power transfer. J Spacecr Rockets. 1998;5:69–75.
2014;51:1914–1923.
30. Wojtkowiak H, Balagurin O, Fellinger G, Kayal H. ASAP: autonomy
11. Boshuizen C, Mason J, Klupar P, Spanhake S. Results from the planet through on-board planning. Proceedings of the 2013 6th Interna-
labs flock constellation. 28th Annual AIAA/USU Conference on Small tional Conference on Recent Advances in Space Technologies. IEEE;
Satellite; 2014. 2013:377–381.
12. Ding Xl, Li ZW, Zhu JJ, Feng GC, Long JP. Atmospheric effects 31. Tipaldi M, Glielmo L. A survey on model-based mission planning
on InSAR measurements and their mitigation. Sensors 2008;8:5426– and execution for autonomous spacecraft. IEEE Syst J. 2017;99:1–13.
5448. https://doi.org/10.1109/JSYST.2017.2720682
13. Tack F, Goossens R, Buyuksalih G. Assessment of a photogrammet- 32. Space segment operability. European Cooperation for Space Standard-
ric approach for urban DSM extraction from tri-stereoscopic satellite ization; 2008.
imagery. Photogramm Rec. 2012;27:293–310. 33. Jónsson A, Morris RA, Pedersen L. Autonomy in space. Current capa-
14. Lucht W, Schaaf CB, Strahler AH. An algorithm for the retrieval of bilities and future challenges. AI Mag. 2007;28:27–42.
albedo from space using semiempirical BRDF models. IEEE Trans Geosci 34. Vamvoudakis KG, Antsaklis PJ, Dixon WE, et al. Autonomy and
Remote Sens. 2000;38:977–998. machine intelligence in complex systems: a tutorial. Proceedings of the
15. Nag S, Gatebe C, Hilker T. Simulation of multiangular remote sensing IEEE 2015 American Control Conference; 2015:5062–5079.
products using small satellite formations. IEEE J Sel Top Appl Earth Obs 35. Iacopino C, Palmer P. Autonomy. In: D'Errico M, ed. Distributed
Remote Sens. 2017;10:638–653. Space Missions for Earth System Monitoring, vol. 31. Springer,
New York, NY: Springer Science & Business Media; 2012:309–329.
16. Nag S, Gatebe CK, Miller DW, de Weck O. Effect of satellite forma-
https://doi.org/10.1007/978-1-4614-4541-8
tions and imaging modes on global albedo estimation. Acta Astronaut.
2016;126:77–97. 36. Damiani S, Verfaillie G, Charmeau MC. An earth watching satellite con-
stellation: how to manage a team of watching agents with limited com-
17. Chien S, Sherwood R, Tran D, et al. Using autonomy flight software to
munications. Proceedings of the 4th International Joint Conference
improve science return on earth observing one. J Aerosp Comput Inform
on Autonomous Agents and Multiagent Systems, ACM; 2005:455–
Commun. 2005;2:196–216.
462.
18. Chien S, Johnston M, Frank J, et al. A generalized timeline representa- 37. Bonnet G, Tessier C. Collaboration among a satellite swarm. Proceed-
tion, services, and interface for automating space mission operations. ings of the 6th International Joint Conference on Autonomous Agents
Space Operations Symposium, Stockholm, Sweden; 2012. and Multiagent Systems, ACM; 2007:54.
19. Selva D, Krejci D. A survey and assessment of the capabilities of Cube- 38. Vassev E, Sterritt R, Rouff C, Hinchey M. Swarm technology at NASA:
sats for Earth observation. Acta Astronaut. 2012;74:50–68. building resilient systems. IT Prof Mag. 2012;14:36–42.
20. Woellert K, Ehrenfreund P, Ricco AJ, Hertzfeld H. Cubesats: cost- 39. Sterritt R, Rouff CA, Hinchey MG, Rash JL, Truszkowski W. Next gener-
effective science and technology platforms for emerging and develop- ation system and software architectures: challenges from future NASA
ing nations. Adv Space Res. 2011;47:663–684. exploration missions. Sci Comput Program. 2006;61:48–57.
ARAGUZ ET AL . 15

40. Montenbruck O, Kahle R, D'Amico S, Ardaens JS. Navigation and con- 62. Crina G, Ajith A. Stigmergic optimization: inspiration, technologies and
trol of the TanDEM-X formation. J Astronaut Sci. 2008;56:341–357. perspectives. Stigmergic Optimization. Berlin, Heidelberg: Springer
41. Gill E, Montenbruck O, D'Amico S. Autonomous formation flying for Berlin Heidelberg; 2006:1–24.
the PRISMA mission. J Spacecr Rockets 2007;44:671–681. 63. Iacopino C, Harrison S, Brewer A. Mission planning systems for com-
42. Llorente JS, Agenjo A, Carrascosa C, et al. PROBA-3: precise formation mercial small-sat Earth observation constellations. Proceedings of the
flying demonstration mission. Acta Astronaut. 2013;82:38–46. 2015 International Workshop on Planning and Scheduling for Space,
Buenos Aires, Argentina; 2015.
43. Grogan PT, Golkar A, Shirasaka S, de Weck O. Multi-stakeholder inter-
active simulation for federated satellite systems. 2014 IEEE Aerospace 64. Beaumet G, Verfaillie G, Charmeau MC. Feasibility of autonomous
Conference; 2014. decision making on board an agile Earth-observing satellite. Comput
Intell. 2011;27:123–139.
44. Muri P, McNair J. A survey of communication sub-systems for inter-
satellite linked systems and cubesat missions. J Commun. 2012;7:290– 65. Wang C, Li J, Jing N, Wang J, Chen H. A distributed cooperative
308. dynamic task planning algorithm for multiple satellites based on multi-
agent hybrid learning. Chin J Aeronaut. 2011;24:493–505.
45. Wood L, Ivancic W, Stewart D, Northam J, Jackson C, da Silva Curiel
A. IPv6 and IPsec on a satellite in space. 58th International Astronautical 66. Lenzen C, Wörle MT, Mrowka F, Geyer MP, Klaehn R. Automated
Congress; 2007. scheduling for TerraSAR-X/TanDEM-X. Proceedings of the 2011 Interna-
46. Lluch I, Grogan PT, Pica U, Golkar A. Simulating a proactive ad-hoc net- tional Workshop on Planning and Scheduling for Space; 2011.
work protocol for Federated Satellite Systems. 2015 IEEE Aerospace 67. Miller E, Paulsen P, Pasciuto M. Autonomous satellite operations via
Conference, IEEE; 2015:1–16. secure virtual mission operations center. Proceedings of IEEE Inter-
47. Ivancic W, Eddy WM, Stewart D, Wood L, Northam J, Jackson C. Expe- national Geoscience and Remote Sensing Symposium (IGARSS), IEEE;
rience with delay-tolerant networking from orbit. Int J Satellite Com- 2010:2607–2610.
mun Netw. 2010;28:335–351. 68. Tripp H, Palmer P. Stigmergy based behavioural coordination for satel-
48. Rose TS, Janson S, LaLumondiere W, et al. LEO to ground optical com- lite clusters. Acta Astronaut. 2010;66:1052–1071.
munications from a small satellite platform. Proceedings of Free-Space 69. Tejo JA, Garrigues Rua A, Arregui JP. Planning the operations for
Laser Communication and Atmospheric Propagation XXVII, vol. 9354; Sentinel-1 satellite: how to fit a complex puzzle. SpaceOps 2014 Con-
2015:1–9. ference; 2014:1727.
49. Sycara KP. Multiagent systems. AI Mag. 1998;19:79. 70. González JA, Gonzalez O, Monge Á, Cillero B, Pirondini F, Ortiz A.
50. Ocón J, Rivero E, Sanchez Montero A, Cesta A, Rasconi R. Multi-agent DEIMOS-2 advanced mission planning capabilities. SpaceOps 2014
frameworks for space applications. Proceedings of the 11th International Conference. American Institute of Aeronautics and Astronautics;
Conference on Space Operations, Huntsville, AL; 2010. 2014.

51. Garnier S, Gautrais J, Theraulaz G. The biological principles of swarm 71. Iacopino C, Palmer P, Policella N, Donati A, Brewer A. How ants can
intelligence. Swarm Intell. 2007;1:3–31. manage your satellites. Acta Futura. 2014;9:59–72.

52. Tummolini L, Mirolli M, Castelfranchi C. Stigmergic cues and their uses 72. Carrel A, Palmer P. An evolutionary algorithm for near-optimal
in coordination: an evolutionary approach. In: Uhrmacher A, Weyns D, autonomous resource management. 8th International Symposium on
eds. Agents, Simulation and Applications. Taylor & Francis, Boca Raton, Artificial Intelligence, Robotics and Automation in Space; 2005.
FL: CRC Press; 2009:243–265. 73. HolmesParker C, Agigino A, Tumer K. Evolving distributed resource
53. Monekosso ND, Remagnino P. A collective can do better. Design of sharing for cubesat constellations. Proceedings of the Genetic and Evo-
Intelligent Multi-Agent Systems. New York, NY: Springer; 2005:215– lutionary Computation Conference; 2012.
238. 74. Bonnet J, Gleizes MP, Kaddoum E, Rainjonneau S, Flandin G. Multi-
54. Ding Y, Jin Y, Ren L, Hao K. An intelligent self-organization scheme for satellite mission planning using a self-adaptive multi-agent system.
the internet of things. IEEE Comput Intell Mag. 2013;8:41–53. Proceedings of the 2015 IEEE 9th International Conference on Self-
Adaptive and Self-Organizing Systems (SASO); 2015:11–20.
55. Ducatelle F, Di Caro GA, Pinciroli C, Gambardella LM. Self-organized
cooperation between robotic swarms. Swarm Intell. 2011;5:73–96. 75. Bonnet G, Tessier C. Coordination despite constrained communica-
tions: a satellite constellation case. 3rd National Conference on Control
56. Parker LE. Distributed intelligence: overview of the field and its appli-
Architectures of Robots; 2008:89–100.
cation in multi-robot systems. J Phys Agents 2008;2:5–14.
76. Chien S, Tran D, Rabideau G, Schaffer S, Mandl D, Frye S. Improving the
57. Johansson R, Saffiotti A. Navigating by stigmergy: a realization on an
operations of the Earth Observing One mission via automated mission
RFID floor for minimalistic robots. Proceedings of the 2009 IEEE Interna-
planning. Proceedings of the 11th International Conference on Space Oper-
tional Conference on Robotics and Automation; 2009:245–252.
ations; 2010.
58. Werfel J, Nagpal R. Extended stigmergy in collective construction. IEEE
77. Chien S, Doubleday J, Ortega K, et al. Onboard autonomy and ground
Intell Syst. 2006;21:20–28.
operations automation for the Intelligent Payload Experiment (IPEX)
59. Werfel J, Petersen K, Nagpal R. Designing collective behavior in a CubeSat mission. Proceedings of the International Symposium on Artificial
termite-inspired robot construction team. Science 2014;343:754–758. Intelligence, Robotics, and Automation for Space; 2012.
60. Tsankova D, Georgieva V, Zezulka F, Bradac Z. Immune network con- 78. Ceballos A, Bensalem S, Cesta A, et al. A goal-oriented autonomous
trol for stigmergy based foraging behaviour of autonomous mobile controller for space exploration. Proceedings of 11th Symposium on
robots. Int J Adapt Control Signal Process. 2007;21:265–286. Advanced Space Technologies in Robotics and Automation, vol. 11;
61. Cornejo A, Dornhaus A, Lynch N, Nagpal R. Task allocation in 2011.
ant colonies. Distributed Computing. Springer, Berlin, Heidel- 79. Araguz C, Alvaro A, Del Portillo I, Root K, Alarcón E, Bou-Balust E. On
berg: Springer; 2014:46–60. https://doi.org/10.1007/978-3-662- autonomous software architectures for distributed spacecraft: a local-
45174-8_4 global policy. 2015 IEEE Aerospace Conference, IEEE; 2015:1–9.
16 ARAGUZ ET AL .

80. Rodrigo Muñoz S. A scalable distributed autonomy system for frac- AUTHOR'S BIOGRAPHIES
tionated satellite missions. Technical University of Catalonia – UPC
BarcelonaTech. Barcelona; 2016.
Carles Araguz received his BSc. and MSc. degrees in electronics engi-
81. Tonetti S, Cornara S, Pirondini F. Fully automated mission planning and
neering in 2014 from Technical University of Catalonia, where he
capacity analysis tool for the DEIMOS-2 Agile satellite. Proceedings
is currently pursuing his PhD. In 2012 he joined the Nano-Satellite
of the 13th International Conference on Space Operations. American
Institute of Aeronautics and Astronautics; 2014. and Payload Laboratory as a graduate researcher. His research inter-
82. Englander JA, Conway BA, Williams T. Automated mission planning via ests include autonomous distributed satellite systems, mission plan-
evolutionary algorithms. J Guid Control Dyn. 2012;35:1878–1887. ning and scheduling algorithms and the development of model-based
83. Baek SW, Han SM, Cho KR, et al. Development of a scheduling algo- simulators.
rithm and GUI for autonomous satellite missions. Acta Astronaut.
2011;68:1396–1402.
Elisenda Bou-Balust M.Sc. Telecom Engineering UPC'2011, M.Sc. Elec-
84. Steiger C, Furnell R, Morales J. On-board control procedures for ESA's
tronic Engineering ULPGC'2011, PhD UPC'2017. Adjunct professor at
Deep space missions Rosetta and Venus Express. Proceedings of the
Data Systems in Aerospace Conference, vol. 60; 2005. UPC and CTO at Vilynx. Since 2007 she has specialized in large-scale
AI systems, coauthoring more than 40 articles and 5 patents and par-
85. Grasset-Bourdel R, Verfaillie G, Flipo A. Planning and replanning for
a constellation of agile Earth observation satellites. Proceedings of the ticipated in national and international R&D projects.
ICAPS-11 Workshop on Scheduling and Planning Applications; 2011.
86. Wu G, Liu J, Ma M, Qiu D. A two-phase scheduling method with the Eduard Alarcón is a faculty member of the EE School at UPC
consideration of task clustering for earth observing satellites. Comput
BarcelonaTech, where he endeavors as educator and mentor of
Oper Res. 2013;40:1884–1894.
researchers. He is currently Vice President of Technical Activities
87. Bianchessi N, Cordeau JF, Desrosiers J, Laporte G, Raymond V. A
for the IEEE Circuits and Systems Society and Editor-in-Chief of the
heuristic for the multi-satellite, multi-orbit and multi-user manage-
ment of Earth observation satellites. Eur J Oper Res. 2007;177:750– IEEE Journal of Emergent Topics in Circuits and Systems. His research
762. interests include the areas of small satellites, nanotechnology-
88. Wu G, Pedrycz W, Li H, Ma M, Liu J. Coordinated planning of hetero- enabled wireless communications and onchip energy management and
geneous Earth observation resources. IEEE Trans Syst Man Cybern Syst. harvesting.
2016;46:109–125.
89. Mansour MA, Dessouky MM. A genetic algorithm approach for solving
the daily photograph selection problem of the SPOT5 satellite. Comput
Indust Eng. 2010;58:509–520. How to cite this article: Araguz C, Bou-Balust E, Alar-
cón, E. Applying autonomy to distributed satellite systems:
90. Geyer MP, Mrowka F, Lenzen C. TerraSAR-X/TanDEM-X mission
planning–handling satellites in close formation. Proceedings of the AIAA Trends, challenges and future prospects. Syst Eng. 2018;1–16.
International Conference on Space Operations; 2010. https://doi.org/10.1002/sys.21428