Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems:

Results from a Systematic Literature Review

Davide Calvaresi

1,2

, Kevin Appoggetti

, Luca Lustrissimini

, Mauro Marinoni

Paolo Sernani

, Aldo F. Dragoni

and Michael Schumacher

Scuola Superiore Sant’Anna, Pisa, Italy

University of Applied Sciences Western Switzerland, Sierre, Switzerland

Universit

a Politecnica delle Marche, Ancona, Italy

Keywords:

Software Engineering, Negotiation, Multi-Agent Systems, Cyber-Physical Systems, Real-Time Systems.

Abstract:

Cyber Physical Systems (CPS) require a multitude of components interacting among themselves and with the

users to perform automatic actions, usually under unpredictable or uncertain conditions. Multi-Agent Systems

(MAS) have emerged over the years as one of the major technological paradigms regulating interactions and

negotiations among autonomous entities running under heterogeneous conditions. As such, MAS have the

potential to support CPS in implementing a highly reconﬁgurable distributed thinking. However, some gaps

are still present between MAS’ features and the strict requirements of CPS. The most relevant is the lack of

reliability, which is mainly due to speciﬁc features characterizing negotiation protocols. This paper presents

a systematic literature review of MAS negotiation protocols aiming at providing a comprehensive overview

of their strengths and limitations, examining both the assumptions and requirements set during their develop-

ment. While this work conﬁrms the potential of MAS in regulating the interactions among CPS components,

the ﬁndings also highlight the absence of real-time compliance in current negotiation protocols. Strongly

characterizing CPS, the capability to face strict time constraints could bridge the gap between MAS and CPS.

1 INTRODUCTION

Cyber-Physical Systems (CPS) are deeply rooted

in our daily living. Interconnected electronic de-

vices of any size (from wearable to huge drivers)

compose heterogeneous systems operating in var-

ious domains (e.g., manufacturing (Hsieh, 2002),

zero-energy buildings, near-zero automotive fatali-

ties (Rajkumar et al., 2010), telerehabilitation (Cal-

varesi et al., 2017b), and e-health (Calvaresi et al.,

2014)). Scalable across time and space, with the abil-

ity to cope with a scenario’s uncertainty, privacy con-

cerns and security issues, CPS and MAS are trans-

forming the humans’ control of the physical world.

Usually, these systems employ sensors to collect data

from the real world, process them, and then provide

feedback, either to other entities, or directly affect-

ing (e.g., via actuators) the real world. Such systems

are capable and responsible for both performing hard-

coded and automatic actions and dealing with unpre-

dictable or uncertain situations requiring “intelligent”

actions. The distributed nature of such systems opens

the horizon to a multitude of possible synergies. In-

teractions among entities of same or different systems

represent a fascinating world, which has been largely

investigated by the scientiﬁc community. However,

new arising challenges have still to be faced.

On the one hand, according to Calvaresi et al (Cal-

varesi et al., 2017a), Multi-Agent Systems (MAS)

is one of the most prominent and promising “ap-

proaches” supporting Internet of Things (IoT) tech-

nologies and CPS. The adoption of a multi-agent

framework can facilitate the implementation of coop-

erative/competitive distributed thinking, robustness,

reconﬁgurability, reusability (e.g., components ca-

pabilities, functionalities, knowledge), and a par-

tial technology independence (smoother migration

among different technologies) (Bellifemine et al.,

2007; Calvaresi et al., 2016b). On the other hand,

CPS require strict dependably, stringent safety and

security policies, resources efﬁciency, and real-time

guarantees (Rajkumar et al., 2010). For example, a

safe use of personal devices (e.g., wearable blood-

sugar/pressure devices), reliable and timely informa-

tion delivery, bounded risks in receiving wrong infor-

mation (in terms of content and timing), privacy guar-

224

Calvaresi, D., Appoggetti, K., Lustrissimi, L., Marinoni, M., Sernani, P., Dragoni, A. and Schumacher, M.

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review.

DOI: 10.5220/0006594802240235

In Proceedings of the 10th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2018) - Volume 1, pages 224-235

ISBN: 978-989-758-275-2

antees and systems overall stable are features strictly

required in safety-critical CPS.

Although the advantages provided by the adoption

of MAS are remarkable, the full compliance with the

requirements of CPS is not met yet (Calvaresi et al.,

2017a). Uncertainty in the environment, security

attacks, limitations in cyber models, and errors in

physical devices make ensuring the overall system

robustness, security, and safety, a critical challenge.

The distributed decision-making process is crucial

in the above-mentioned systems, and the negotiation

process is essential for their success.

Contribution

To reach consensus or just interact, MAS need sev-

eral negotiation protocols (standard and not). To bet-

ter understand such contributions, this work performs

a Systematic Literature Review (SLR) of the most rel-

evant negotiation protocols proposed in the scientiﬁc

literature addressing the following features:

(i) assumptions have been detailed to deﬁne the char-

acteristics of environments and systems in which

the negotiation processes are operating;

(ii) requirements have been presented and related to

the assumptions to deﬁne which objectives and

constraints have been set;

(iii) Strengths, and limitations collected by the pri-

mary studies have been elaborated to highlight

achievements and still open challenges.

Elaborating and summarizing the evidence, the

criteria presented in Section 3 have been generated

and discussed. Finally, considering the reliability as

the main requirement of safety-critical CPS, the nego-

tiation’s characteristics, constraints, and bounds have

been formalized. The paper is organized as follows:

Section 2 presents the review process and data collec-

tion, Section 3 organizes and describes the obtained

results, Section 4 brieﬂy discusses the obtained results

in key CPS. Finally, Section 5 concludes the paper.

2 DATA COLLECTION AND

REVIEW PROCESS: THE

METHODOLOGY

Retrieving, selecting, and analyzing existing litera-

ture has more relevance if performed systematically.

Hence, this paper adheres the procedure suggested

by (Kitchenham et al., 2009) and adapted by (Cal-

varesi et al., 2016a). Such a methodology is com-

posed of three stages (see Figure 1), and it is rigorous

and reproducible

Firstly, Planning the review deﬁnes steps and

constraints. Such a phase elaborates a generic

free-form question in structured research questions

(SRQs) which characterize the pillars of the whole

protocol. By doing so, the outcome will be repro-

ducible, reliable, and comparable. The second stage,

Performing the review, deals with the execution of

the planned activities: (i) papers’ collection and se-

lection, (ii) paper elaboration, and (iii) features ex-

traction. The last step, Document Review, deals with

the data analysis and reporting activities related to the

scientiﬁc dissemination.

2.1 Planning the Review

Deﬁning the review process sets the research ques-

tions and their contexts, search strategy, review

protocol, and biases and disagreement resolution.

Research Questions Deﬁnition

Investigating the scenarios presented in Section 1, the

following free-form questions arose: (i) What needs,

characterize the negotiations among agents in the sev-

eral application scenarios? (ii) Are the solutions pro-

posed by the scientiﬁc community satisfactory? (iii)

How are such solutions characterized?

The Goal-Question-Metric (GQM), proposed by

Kitchenham et al. (Kitchenham et al., 2010) and Gal-

ster et al. (Galster et al., 2014), ruled the decompo-

sition of the unstructured questions mentioned above,

into a set of three structured research questions. In

particular, the assumptions, requirements, strengths,

and limitations led the investigation and the deﬁnition

of the following questions:

SRQ1 Setting the next question we aim at understand-

ing the Step 0 of the negotiation protocol develop-

ment: What are the assumptions rooting the most

relevant approaches?

SRQ2 To identify the goals targeted by such protocol,

the following question is set: What are the re-

quirements such approaches intend to meet?

SRQ3 The adoption of a speciﬁc negotiation algorithm

would possibly bring some advantages. To name

them, the following question is set: What are the

strengths and limitations characterizing the re-

lated negotiation approaches?

Develop the Review Protocol

Once completed the deﬁnition of the structured-

research-questions, the deﬁnition of the Search Strat-

egy follows. Gray literature may introduce possible

Primary studies selected and elaborated in early 2017

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

225

Planning the Review (a)

Dissemination (c)

Deﬁne the research questions

[free form and structured question]

Develop the review protocol

[Search strategy deﬁnition]

Validate the review protocol

Data Analysis

Final report composition

Summarizing evidence

• Channel of research

• Acceptance criteria

• Set of keywords

• Inclusion criteria

• Stop collecting criteria

• Features and quality criteria

• Bias and disagreement resolution

• Expected output format

Performing Review (b)

Disagreement

resolution

Article Elaboration

[Features collection, DARE criteria]

Article Selection

[Inclusion criteria application]

Article Collection

[Systematic search execution]

Figure 1: Review Methodology Structure according to (Kitchenham et al., 2009) and (Calvaresi et al., 2016a).

biases. Thus, only peer-reviewed collectors of papers

(ieeeXplore

, Sciencedirect

, ACM Digital Library

and Citeseerx

) have been investigated.

To obtain more accurate results during the

semi-automatic research, some keywords have been

contextualized (by aggregating at least two or three

words). According to the reviewers’ rooted back-

grounds and knowledge related to the Multi-Agent

domain, the following set of keywords has been

deﬁned: multi-agent interaction protocol, multi-

agent negotiation protocol, agent-based negotiation,

multi-agent problem-solving negotiation, distributed

problem-solving negotiation, control distributed

problem-solving. For each query, the papers crawlers

produced lists of articles ordered by pertinence. The

criteria used to stop the paper collection is the same

adopted by Calvaresi et al. in (Calvaresi et al., 2016a).

Inclusion Criteria Deﬁnition

The initial research counted 200 papers. A further

coarse-grained examination reduced them to 143. The

reviewers ﬁltered them by performing a simultaneous

and autonomous check of titles and abstracts’ perti-

nence with the following inclusion criteria:

A) Context: The primary studies should deﬁne their

contributions in the context of distributed-like

systems;

B) Purpose: The purpose of primary studies should

refer to mechanisms for negotiating tasks and re-

sources or for achieving agreement or consensus

between distributed entities.

http://ieeexplore.ieee.org/Xplore/home.jsp

http://www.sciencedirect.com/

http://dl.acm.org/

http://citeseerx.ist.psu.edu/index

C) Relevance: The primary studies should provide

at least one of the following elements: [theoreti-

cal model, interaction mechanisms, practical im-

plementation, tests, critical analysis, critical eval-

uations or discussion]

In the case of a clear verdict was missing (e.g.,

R1(Yes), R2(No), R3(Maybe)) the disagreement

resolution process described below has been applied.

Features and Quality Criteria Deﬁnition

During the “Features Collection”, assessing the qual-

ity of the information provided by the primary studies

is one of the main challenges of a Systematic Litera-

ture Review (Calvaresi et al., 2016a).

Although this work deals with a well-deﬁned set

of feature, context, rationale, research justiﬁcation,

critical examination, statement of ﬁndings and

possible biases can hamper the credibility. Thus, the

retrieved features have been classiﬁed by associating

them Y - information is explicitly deﬁned / evaluated,

P - information is implicit / stated, or N - information

is not inferable (DARE critirea (Kitchenham et al.,

2009)).

Biases and Disagreement Resolution

The following expedients have been adopted to min-

imize and solve possible biases and conﬂicts. Devel-

oping the method and elaborating the articles, most

of the tasks have been cross-checked. In particular,

concerning Figure 1:

• the reviewers conducted the tasks included in 1(a)

and (b) “Planning the Review”, and “Document

Review” collaborating synchronously.

• The collected articles list has been divided into

three (number of reviewers performing the “Ar-

ticle selection”) subsets, which have been pro-

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

226

cessed (applying the inclusion criteria check) by

at least two out of three reviewers. The single re-

viewer’s choices (Yes, No, or Maybe) have been

kept hidden from each other till all of them had

completed such a task. In the case of possible un-

certainties (e.g., Yes-No, Yes-Maybe, No-Maybe)

a third reviewer has been asked an extra check to

ﬁnally decide weather include the article in the ﬁ-

nal list (to be elaborated) or not.

• During the “Article Elaboration”, in the case rel-

evant doubts arose, periodical collaborative dis-

agreement resolution meetings have been orga-

nized.

3 RESULTS PRESENTATION

This section discusses the outcomes obtained by per-

forming the methodology presented in Section 2. The

main investigated issues are the assumptions on which

the studied protocols rely on, the subsequent require-

ments set by the authors of the primary studies to

identify and proﬁle the proposed algorithms, and ﬁ-

nally, the elaborated strengths and limitations, to sum-

mon the state of the art and identify future challenges.

3.1 Assumptions

The assumptions have been clustered to elicit abstract

categories thus facilitating presentation and under-

standing (see Table 1). Most of the systems com-

posed by distributed entities are based on the inter-

actions among the available components. In MASs,

such interactions have always been assumed asyn-

chronous (Smith, 1980; Smith and Davis, 1981)

strengthening the autonomy of single agents (e.g.,

their ability to execute without a direct human inter-

vention and with full control over their own thread).

Despite the communication-delay can be a crucial

component, some studies neglect it, referring to the

hypothesis of instantaneous message delivery (Ak-

nine, 1998). In most cases, the authors refer to

a general multi-agent architecture, even if few of

the analyzed papers base their agents on the BDI

paradigm (Atkinson et al., 2005). The design of a

negotiation protocol mainly relies on the capability of

taking autonomous decisions to pursue beliefs or di-

rectly self-interested or common goals. Indeed, the

rationality (e.g., the ability of agents to always exe-

cute to achieve their goals, and never to prevent them

from being achieved) and autonomy of agents are the

most common assumptions in the analyzed studies.

For example, in a group choice design support system

(GCDSS), the agents negotiate on behalf of their user

trying to persuade other agents according to their im-

posed or independently developed knowledge (Rus-

sell et al., 1995; Ito and Shintani, 1997).

Often, such autonomy has to face the impossi-

bility of having agents ready with complete knowl-

edge. Although dealing with partial knowledge might

lead to possible deception, it is the most studied sce-

nario in both cooperative and competitive MAS (Ak-

nine et al., 2004; Zlotkin and Rosenschein, 1991;

Smith and Davis, 1981). Having a competitive rather

than cooperative agents’ community, frames com-

pletely different scenarios and conditions which are

even more complex in the case they are both cooper-

ative and competitive at the same time. Some prac-

tical examples of negotiating limited knowledge in

cooperative scenarios are the control of UAVs’ task

scheduling (Budaev et al., 2016), monitoring elec-

tricity transformation networks, and scheduling meet-

ings (Kraus, 1997). Agents can collaborate by fol-

lowing self-organizing policies or relying on an or-

chestrator/coordinator (Wang et al., 2014) (the spec-

ular role in competitive scenarios is named “moder-

ator” (Hanachi and Sibertin-Blanc, 2004)). Agents

have to be “certiﬁed” or “trusted” (Alberti et al.,

2004). Thus, the collaboration is more secure and can

be applied in crucial activities such as decision mak-

ing, coordination, and control processes. The bid-

based negotiation approach is the most diffused, de-

spite the involvement of simple or complex tasks (Ak-

nine et al., 2004). In this approach, each agent

can play two main roles: (i) the initiator (who calls

for bids) and (ii) the contractor (who bids) in 1-to-

1, 1-to-many scenarios, or auction based many-to-

many (Wang et al., 2014). It can be predicted to

last for short (Faratin et al., 1998) or long (Collins

and Wolfgang Ketter, 2002) periods of time. In the

scenario where the negotiation is still not converg-

ing, it might be considered as failed (Aknine et al.,

2004). During a single instance of the bid-based

protocols, an agent can play one of the two roles.

Nevertheless, during the system execution, several

negotiations of several tasks or resources can hap-

pen, and then, agents can play both (i) and (ii) (as-

suming a community of agents playing exclusively

either (i) or (ii) is a rare scenario). In collabora-

tive scenarios, due to their inner mechanisms, par-

ticular negotiation protocols need to prevent agents

from over-bidding (e.g., very high rates in the Pre-

Bidding phase). The solutions have been “bound-

ing” the cooperation with the introduction of self-

interested agents (Aknine et al., 2004), imposing “se-

quentiality” (Hanachi and Sibertin-Blanc, 2004), or

limiting the number of issues to be possibly negoti-

ated (Faratin et al., 1998).

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

227

Table 1: Assumptions overview.

Assumption Class Assumption Class Assumption Class

No-commitment AU Agents multi-role AR Stationary/mobile agents AR

Customizable neg. FL Customizable interaction prot. FL No comm-delay AR

Mobile agents AR Autonomous agents AU agent as service provider AR

Cooperative agents CP Neighborhood limited comm. IN Instantaneous messaging IN

Partial knowledge RA Cooperative decision making CP Multi-crit. decision making RA

Cooperative control CP BDI agents AR No bids on conﬂicting plans RB

Cooperative computation CP Certiﬁed agents RL One bid per agent per time FL

Multilateral comm. IN Limited resources RA Low-level comm. protocol IN

Shared resources AR Delegation FL No explicit utility transfer RA

Concurrent agents CM Castable (constraints/agents) FL Guaranteed resource alloc. RL

Sequential neg. RB Feedback mechanism RL Information Completeness AR

Roles re-deﬁnable

FL Bounded behaviors RA Competitive agents CM

Coordinated agents AR Stationary agents AR Self-interested agents CM

1-to1 negotiation IN Static environment AR Indivisible resources AR

1-to-n negotiation IN Tasks fully preemptable RB Loosely-coupled agents AR

Time efﬁciency PR Fault-tolerance RB Asynchronous agents AR

Failing negotiations RB simple tasks AR Sub-optimality PR

Neg. topic related AR Ontology IN Timed negotiation N

Long-time negotiation FL Agents’ speciﬁc role AR Limited services/issues FL

Indirect interactions IN Short-time negotiation FL Multi-negotiation FL

Allowed counter-offers FL No preemption AR Independent tasks AR

Penalized de-committing RB Extendable agents FL Rational Agents AR

Uncertainty RB

Legend

AR Architectural FL Flexibility RA Rationality CM Competition RB Robustness

AU Autonomy CP Cooperation IN Interaction PR Performance RL Reliability

The “pool” of agents able to take part in a nego-

tiation might be subject to some constraints. For ex-

ample, it can be restricted by the concept of neighbor-

hood (Olfati-Saber et al., 2007; Budaev et al., 2016)

which can have completely different outcomes if con-

sidering stationary agents (e.g., agents which execute

always in the same node of a network), mobile agents

(e.g., agents able to migrate to different nodes at run-

time), or hybrid scenarios (Ferber and Gutknecht,

1998; Wang et al., 2014). In (Aknine et al., 2004),

the agent selection for a task execution is based on

several factors such as the position of the agent in its

environment and its capacity to process information.

Reza et al. (Olfati-Saber et al., 2007) give cru-

cial importance to the agents’ autonomy, especially

in the presence of possible link/node failures unex-

pected time-delay and possible changes in the net-

work topology. The assumption of having a system

capable of operating as expected even in the case one

or more failures happen is quite strong. However, sev-

eral studies such as (Aknine et al., 2004) adopted it,

facing scenarios where faults are most likely to hap-

pen. Several studies made assumption enforcing the

ﬂexibility, but hampering (in some cases impeding)

the reliability. For example, the possibility of break-

ing a commitment (the promise made for a task exe-

cution in the bidding phase), with (Wu, 2008; Zhou

et al., 2004) or without penalty, is not remotely al-

lowed (Odell et al., 2001; Odell et al., 2000). Assum-

ing the possibility of delegating tasks to other agents,

it would boost ﬂexibility and efﬁciency but limit re-

liability and rationality. The possibility of preempt-

ing tasks/behaviors is reasonable. However, assum-

ing complete preemptability coupled with the absence

of explicit deadlines, and allowing the possibility of

failing negotiations, identical outcomes might be gen-

erated: multiple deadlines missing or direct starva-

tion (Krothapalli and Deshmukh, 1999; Aknine et al.,

2004). Sharing resources is a common practice to en-

hance system ﬂexibility, bounded by their availabil-

ity (Wellman and Wurman, 1998). Several protocols

consider the customization of the negotiation interac-

tions (Mazouzi et al., 2002) possible by also provid-

ing a pre-set personalization mechanism (Demazeau,

1995; Purvis et al., 2003). The agents’ roles might be

assumed static or dynamic (Wang et al., 2014; Faratin

et al., 1998).

3.2 Requirements

Once the most common and relevant assumptions

have been framed, the next step is to investigate the

prevailing requirements set for negotiation protocols

in MAS (see Table 2). The agents’ interaction leading

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

228

to the achievement of consensus and self/community

goals captured the most concerns. Many contribu-

tions provide only negotiation-baselines, and thus re-

quire the implementation of generic/ad-hoc heuris-

tics (Wanyama and Far, 2007). According to Mazouzi

et al. (Mazouzi et al., 2002), being able to identify

how and when to validate protocols, evaluate their

success, and explain the relationships between agents,

are outstanding requirements that must be considered.

Nonetheless, deciding whom to interact with (e.g.,

agents with a higher reputation should have better

bearing than others) and when initiating the interac-

tion in certain scenarios is also crucial (Ramchurn

et al., 2004).

On one hand, having an organized structure (Fer-

ber and Gutknecht, 1998) and a ﬂexible and auto-

mated agent community (Kraus, 1997) capable of

achieving desired goals without affecting somebody

else autonomy (Marzougui and Barkaoui, 2013) are

the most common elements characterizing the envi-

ronments in which the negotiation protocols have to

operate in. On the other hand, having feasible, bal-

anced, converging and preserved individual rational-

ity and privacy are the most common elements that

the protocols should present (Wellman and Wurman,

1998). For example, feasibility (basic assumption or

requirement associating all the approaches) involves

the need for setting functionalities such as check-and-

validation of task assignment (Hsieh, 2002). Some

approaches resulted in being extremely tailored on

certain use-cases. Thus, they set very precise re-

quirements to address a relatively broad multitude of

goals. For example, the impossibility for the con-

tractor to quit a task after having started it (Aknine

et al., 2004), the non-retractability of bids, and the

non-returnability of products (Guttman and Maes,

1998) are requirements set to foster reliability, espe-

cially in time-dependent solutions (Collins and Wolf-

gang Ketter, 2002). Moreover, although insufﬁcient

to fully provide real-time guarantees, some solutions

seek for the respect of deadlines and schedulability

guarantees (Shen and Norrie, 1998).

To enhance stability, some authors set the

compliance with precedence and temporal con-

straints (Wanyama and Far, 2007). The time depen-

dency has also been interpreted as the agents’ capa-

bility of conceding more rapidly if the deadline ap-

proaches (Faratin et al., 1998). Regarding resources,

they are assumed limited. Thus, setting a require-

ment regulating resources access and consumption

regarding the agent community and their environ-

ment is mandatory. In trusted and collaborative en-

vironments, setting some policies is required to pro-

tect agents from exploiting each other (Faratin et al.,

1998) and to discourage counter-speculations (Collins

et al., 1998b). Other approaches to avoid security

issues propose the requirement to speciﬁcally de-

ﬁne payment and permission mechanisms (Collins

et al., 1998b), transactions and market architec-

tures (Collins et al., 1998a), mandatory penalty poli-

cies (e.g., non-penalization for new entrance and

changing agents’ identity (Ramchurn et al., 2004)),

agent reputation update rate, and formal speciﬁcation

for processes validation (Mazouzi et al., 2002).

Regarding robustness, systems are required to ei-

ther avoid failures or to keep working if they do oc-

cur below a certain threshold. One solution pro-

posed in the primary studies is to supply informa-

tion about the contractor during task execution (Ouel-

hadj et al., 2005). In particular, Collins et al. (Collins

et al., 1998b) and Hsieh et al. (Hsieh, 2002) propose

the requirement of a robust exception handler and a

method to solve resource conﬂicts. Architectural re-

quirements have been another important and recurrent

element in the primary studies. For example, to over-

come orchestration and autonomy limitations, a mod-

erator could be compulsory (supporting community’s

fairness) (Hanachi and Sibertin-Blanc, 2004). Finally,

to enhance or attain a certain performance, scenario-

driven converging time and maximum execution time

per task set are required (Vulkan and Jennings, 2000).

Despite the lack of critical analysis found in

many scientiﬁc contributions (Calvaresi et al., 2016a),

the analyzed papers have often proposed interesting

clues. The more practical the proposed solutions are,

the more detailed is the analysis of strengths and lim-

itations. The mainly theoretical contributions pre-

sented a broad range of claims from the more ex-

plicit and easily understandable to the more ambi-

tious and ambiguous. By looking at the big pic-

ture, common traits also associate entirely different

approaches. Moreover, clustering strength allowed to

deﬁne a sort of hierarchical relevance of the arisen

categories. Due to space restrictions, the above-

mentioned process will not be addressed in this paper.

Nevertheless, such categories can be easily under-

stood, since they reﬂect the structures of Section 3.3

and Section 3.4

3.3 Strengths

Table 3 collects all the features identiﬁed as

“strengths” by the primary studies. Although feasi-

bility is at the base of every process/protocol, it is not

always guaranteed, and thus many studies consider it

a “strength”. Hence, it is not trivial having a con-

verging negotiation protocol (Hanachi and Sibertin-

Blanc, 2004; Matt et al., 2006) and guaranteeing that

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

229

Table 2: Requirements Overview.

Requirement Class Requirement Class Requirement Class

Speciﬁcs formalization VL Protocols validation VL Planned maintenance AR

Protocol evaluation VL Relating agents RA Tradeoff community/autonomy AU

Agents goal isolation RL Entity-change discouraged RL (who/when)-heuristics for neg IN

Promotion community join FL Agents’ reputation balancing FL Interactions reputation-based RL

Organized structures AR Automated agent AU Fake transactions penalization RL

Privacy preservation RL Individual rationality RA Increasing GDSSs intelligence PR

Efﬁciency PR Policies feasibility RL Services/gods non-returnability AR

Increase of compatibility FL Convergence & equilibrium PR Reasonable Converging Time PR

Ontology-based neg. IN Presence of moderators AR Manager operating in parallel IN

Context-based interactions IN time-limited neg. IN Heterogeneous transactions IN

Online tasks introduction FL Enanching inter-connections AR Complete agents’ knowledge RA

Interconnected managers

AR Limited managers visibility RA Reasonable Execution Time AR

Auction strictly-ruled RL Bids non-retractability AR Unbreakable commitment RL

global goals AR Precedence constraints AR Time-dependent neg. IN

Fault-tolerant neg. RL Resource-dependent neg. IN No-unbalanced exploitation RL

Complex neg. contracts AR Anti-frauds control RL Energy-balancing heuristics PR

Global social goals IN NO counter-speculations RL Secure resource supply RL

Optimal neg. PR Payment mechanisms IN Enable rich-semantic language IN

Enabled alliances IN Robust exception handling RL Multiple/Parallel neg. FL

Scalability AR retro-compatibility IN Common time reference AR

Competitiveness CM Shared knowledge AR Estimating due dates FL

Shared policies AR Costs estimable FL Multiple providers per service AR

Competitive negotiation CM Free community In/Out FL Heuristic-based bids FL

Cooperative framework CP Optimized coordination PR High-level comm. lagns IN

Norms taxonomy VL trust mechanism RL Mass customization PR

Holonic dynamics AR Conﬂict resolution proc. RL Breakable contracts RL

Deadlines respect RL Overview methods RL Comm-trafﬁc reduction PR

Legend

AR Architectural FL Flexibility RA Rationality CM Competition VL Validation

CP Cooperation IN Interaction PR Performance RL Reliability AU Autonomy

a deal can always be achieved (Faratin et al., 1998).

Vice-versa, in the case of failures, detection and ex-

planation of success/failure are possible (El Fallah-

Seghrouchni et al., 1999). A possible way to avoid

failures due to computational intractability is to nego-

tiate throughout a centralized scheduling unit (Kan-

chanasevee et al., 1999). Seeking for effectiveness

and efﬁciency, many analyzed solutions are extremely

specialized and employable only in speciﬁc situa-

tions (Sun and Wu, 2009; Wu, 2008). Nevertheless, it

is possible to mention cases that allow language inde-

pendence (El Fallah-Seghrouchni et al., 1999), con-

text independence (Cardoso and Bordini, 2016) and

protocol re-utilization (Mazouzi et al., 2002), even

in diametrically opposed scenarios (e.g., cooperative

and competitive) (Sandholm, 1993). Some protocols

can deal with uncertain environments, avoiding un-

expected behaviors (Ito et al., 2008) and providing a

high level of formalization (Kraus, 1997) (relatively

ﬂexible (Alberti et al., 2004)).

Moreover, having a controllable protocol size and

a tractable complexity (Mazouzi et al., 2002) helps

to enhance the system’s stability (Olfati-Saber et al.,

2007). Supporting agent autonomy (Hanachi and

Sibertin-Blanc, 2004), one has to cope with a broad

set of constraints. For example, they are radically

different if the scenarios considered are ﬁrmly struc-

tured and automated (Wang et al., 2014) (hierarchical

MAS (Wellman and Wurman, 1998)) or less struc-

tured, but considerably dynamic (e.g., the system just

requires to observe juridical, common-sense, and be-

havioral laws (Wu, 2008), or admits rule re-deﬁnition

on the ﬂy (Purvis et al., 2003)). Finding an opti-

mal trade-off between completeness (the capability of

ﬁnding the optimal solution) (Ito et al., 2008) and the

computational cost is always needed.

MAS are considered distributed by nature, thus

guaranteeing low computational costs (Olfati-Saber

et al., 2007; Collins and Wolfgang Ketter, 2002;

Hong-tao and Kang, 2016; Golfarelli et al., 1997)

is broadly recognized as a major strength. Concern-

ing agent interactions, the overall performance of the

community can be enhanced by shortening global

negotiation processes (Aknine et al., 2004), avoid-

ing inﬁnite plan expansion for recursive plans (Car-

doso and Bordini, 2016), generally reducing traf-

ﬁc (Smith, 1980), avoiding the broadcast of request

messages to all the agents (Shen and Norrie, 1998),

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

230

Table 3: Strengths overview.

Strength Class Strength Class

Deal always possible RL Convergence of conversation RL

Success/failure detection VL NO computational intractability RL

Improved efﬁciency/effectiveness PR Communication lang/tech-independent AR

Allocations context-independent FL Protocol reuse FL

Cooperative/competitive compliant AR Dealing with uncertain environments FL

High-level of formalization AR Flexible speciﬁcation FL

Controllable protocol size FL Tractable complexity PR

Stability

RL Ensure autonomy AU

Allows automated negotiation AR Hierarchical agents AR

Juridical/common-sense compliance RL Rules changing on the ﬂy FL

Low computational costs PR Success/failure explanation RA

Shorter global negotiation processes PR No diverging/recursive plans RL

Net trafﬁc reduced PR Avoidance of broadcasting requests PR

Reduced negotiation rounds PR Dynamic task allocation AR

Fast reaction to unpredictability PR Contract compliance veriﬁable RL

Preventable neg. with blocked agents RB Tasks-sets atomically negotiable AR

Better resource utilization PR Services description not required RA

Multiple heuristics employable AU Possibile parallel negotiations PR

De-commitment reduction PR Complex interactions observable RA

Qualitative/quantitative analysis VL Conﬂict Resolution in Natural Language IN

Trusted neg. sessions RL Increased task execution probability PR

Legend

AR Architectural FL Flexibility RA Rationality CM Competition VL Validation

RB Robustness IN Interaction PR Performance RL Reliability AU Autonomy

and reducing rounds (Wanyama and Far, 2007) and

messages-per-negotiation (Garcia et al., 2017). En-

abling dynamic task allocation (Ouelhadj et al., 2005)

is crucial. Thus, increasing the probability of task

execution (Budaev et al., 2016) is highly appreci-

ated. In terms of performance, the capacity of check-

ing contract compliance (Vok

ınek et al., 2007), and

preventing negotiations with blocked agents (Aknine

et al., 2004), can limit unpredictability (further re-

duced in (Budaev et al., 2016) by decreasing the re-

action time to unpredictable events). Moreover, other

relevant studies mentioned the capability of: negoti-

ating sets of tasks considering them as atomic bar-

gaining items (Sandholm, 1993), improving the re-

source utilization (Xueguang and Haigang, 2004), re-

laxing some constraints in “trusted” negotiation ses-

sions (e.g., no need for services description) (Collins

et al., 1998b), implementing different heuristics (Car-

doso and Bordini, 2016), reducing the decommitment

ratio, and paralleling the negotiation processes (Ak-

nine et al., 2004).

Finally, some approaches permit to be evaluated

by executing formal studies (El Fallah-Seghrouchni

et al., 1999) such as qualitative and quantitative anal-

ysis (Mazouzi et al., 2002), and conﬂict resolution in

natural Language (Demazeau, 1995).

3.4 Limitations

Gathering and analyzing the limitations have been the

most challenging step of the whole review process.

They emerge in three main ways: related to the pro-

posed solution (often implicit and hidden between the

lines), to other approaches presented in the state of

the art, or to speciﬁc solutions used as comparison

terms.

The data elaboration, performed to avoid dupli-

cated elements and to simplify their understanding,

added a considerable overhead in the elaboration pro-

cess. Although several primary studies share the same

limitations, more than a hundred different instances

can be enumerated. The output of such aggregation is

summarized in Table 4.

Sorted by relevance, only the most relevant per

class are presented. The main limitation that affects

some elaborated protocol is the possibility of ending

up in a deadlock (Mazouzi et al., 2002; Aknine et al.,

2004; Golfarelli et al., 1997) which can entail catas-

trophic consequences. In the case of short bidding

windows, both initiators and contractors may lose op-

portunities. In the opposite scenario, with long bid-

ding windows, the whole system might be congested,

thus collecting a cascade of failures. Particularly for

those protocols only suitable for single issue nego-

tiation (Chang and Woo, 1994) or unable to handle

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

231

Table 4: Limitations Overview.

Limitation Class Limitation Class

Risk deadlock RL Limited to single-issue neg. PR

Limited to sequential neg. PR Risk of not reaching stability RL

Single Point Of Failure RL Limited Knowledge access IN

Impossibility of any-time tactics PR Statistic constraints and system’s features PR

High net-trafﬁc PR Not scalable FL

Additional Overhead neglected PR High computational cost PR

Strictly domain-dependent FL Competitive scenarios neglected AR

Semantic neglected IN Protocol limiting interactions IN

Low efﬁciency PF Optimal distribution unreachable PR

Conﬂicting sub-optimal allocations RL No dynamic rescheduling PR

Bounded applicability (issues/agents/interactions) PR Dynamics Non-analyzable RL

Feasibility non-observable RL Execution’s correctness non-observable RL

Risk of injection RL Risk of collusion RL

Legend

AR Architectural FL Flexibility PR Performance IN Interaction RL Reliability

parallel negotiations (Sandholm, 1993). This insta-

bility (Ito et al., 2008; Golfarelli et al., 1997) does

not come alone. Hence, some approaches introduce

single points of failure (Krothapalli and Deshmukh,

1999) such as the coordinator or moderator which can

also be affected by a limited knowledge (Hanachi and

Sibertin-Blanc, 2004; Vulkan and Jennings, 2000).

In the “Open-For-All environment” (Vulkan and Jen-

nings, 2000), there is a more pronounced incapability

to apply tactics at any instant (Faratin et al., 1998),

difﬁculties in deﬁning/updating constraints and sys-

tem features (Hanachi and Sibertin-Blanc, 2004; Jen-

nings et al., 2001), an uncontrolled network traf-

ﬁc growth (Jennings et al., 2001; Faratin et al.,

1998), expansion issues (Krothapalli and Deshmukh,

1999), and neglected additional overheads (Singh

et al., 2010) (e.g., due to increasing computational

costs (Ito et al., 2008; Wan et al., 2007)) hamper

dramatically the systems’ scalability. In terms of

reusability, certain approaches present limited appli-

cation domain (Krothapalli and Deshmukh, 1999;

Aknine, 1998) (e.g., not considering competitive

agents (Sandholm, 1993)). Low level and techno-

logically committed approaches do not consider the

semantic (Smith, 1980), thus concurring to gener-

ate interaction issues (Mazouzi et al., 2002; Jian,

2008). In term of performance, several studies re-

fer to a general “low performance” (Krothapalli and

Deshmukh, 1999; Ito et al., 2007), inefﬁciency (Ito

and Shintani, 1997), and “non-optimality” (Vulkan

and Jennings, 2000; Zhou et al., 2004). In particu-

lar, some approaches do not offer automatic mecha-

nisms (Shen and Norrie, 1998) for task/resource run-

time rescheduling. In same cases, scaling issues and

agents (Wan et al., 2007) may arise problems as well

(e.g., in (Ito et al., 2008), no more than two agents

and seven issues can be properly handled). For exam-

ple, in (Wellman and Wurman, 1998) there is a lack of

in-depth analysis mechanisms, and in (Hsieh, 2002)

checking the feasibility can be difﬁcult or impossi-

ble(referred to cooperative communities). Finally, in

terms of security, checking or enforcing the course

of conversation is not always possible (Hanachi and

Sibertin-Blanc, 2004). Some protocols leave the door

open to possible injections, allowing “strategic lying”

(tricking agents into believing the liars are trustwor-

thy. Thus, they can exploit the unaware agents) (Ram-

churn et al., 2004). Agents collusion is also a lim-

itation and hence, a limited amount of mechanisms

deal with “agent reputation” preventing such unde-

sired circumstances (Ramchurn et al., 2004).

4 DISCUSSION

Exploiting the MAS’ capability of negotiating in CPS

represents a great potential, and it will be one of the

main challenges for MAS in the upcoming years. Ac-

cording to Calvaresi et al. (Calvaresi et al., 2017a),

MAS are still not ready to face strict timing con-

straints which strongly characterize the CPS. Never-

theless, many characteristics of the investigated nego-

tiation protocols conﬁrm such a potential. The agents

in MAS can be seen as distributed nodes in CPS.

Hence, they are assumed as autonomous, concurrent,

coordinated, rational, multi-role, self-interested and

loosely coupled. Computational and functional ca-

pabilities, communication (asynchronous), resources,

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

232

and knowledge are considered limited. Resources can

be shared, tasks in the system can be independent, ar-

chitectures can be heterogeneous, and a mechanism

for fault-tolerance has to be feasible. Sub-optimal re-

source allocations have to be reached in polynomial

time. Unfortunately, some assumptions profoundly

characterizing many negotiation protocols make them

unable to cope with the requirements of CPS. In par-

ticular, in the presence of safety-critical CPS, assump-

tions such as “no-commitment is required, the possi-

bility of delegations, and only a vaguely deﬁned time

efﬁciency” hamper the system reliability. In terms of

requirements, the impossibility to quit a running task,

the non retractability of bidding, the possibility of us-

ing different agent heuristics, the desired guarantee

of respecting deadlines (for manufactured goods), and

the presence of precedence constraints, go in the same

direction of many CPS requirements.

Nevertheless, requirements such as the introduc-

tion of a mediator mechanism to “simplify” the sys-

tem dynamics, the possibility for the agents of chang-

ing their nature/identity, and unconstrained permis-

sion of agents to participate in multiple bids and

tasks, cannot be accepted. Strength is strongly sub-

jected to the combination of requirements and as-

sumptions. Thus, given such biases, anything inferred

may result in inconsistent hypothesis. Instead, in the

same situation, analyzing the limitations gives already

important clues. The algorithms can be deﬁned as in-

adequate to be employed in safety-critical CPS due to

the lack of commitment constraints, the difﬁculties in

checking the feasibility, breaking contracts allowed

by simply “paying” penalties, admission of a single

point of failure, and impossibility of being scalable.

5 CONCLUSIONS

This paper proposed an SLR applied to 143 primary

studies to explore the assumptions standing behind

the negotiation protocol in MAS and the requirements

the different approaches set. Finally, strengths and

limitations have been investigated to understand what

has been done and what is still missing from the

safety-critical CPS perspective.

The negotiation process in such systems involves

smart nodes in distributed networks. The conven-

tional decision-making processes performed in CPS

are subject to more stringent constraints with respect

to the ones characterizing traditional agent-based ap-

plications. The limitations presented in 3.4 and dis-

cussed in Section 4 depict a scenario in which the

most relevant missing feature is the reliability.

Under the same assumption, bridging the gap be-

tween MAS and CPS (e.g., enabling the respect of

strict timing constraints) can unveil new application

scenarios in domestic, manufacturing, and healthcare

domains. Finally, the analyzed techniques assume to

operate in trusted environments. So far, if such a hy-

pothesis is missing, the risk of injections and collu-

sions is quite high. Hence, security challenges ap-

peared to be still open, requiring to secure the systems

at several levels.

Further work shall include the identiﬁcation of the

reliability of the primary objective, and the sets of

assumptions and requirements that have to be rede-

ﬁned accordingly. Consequently, MAS would have

to be purged from the inadequate components, which

consist of several interventions in terms of theoret-

ical contributions and practical development of new

mechanisms. The proposed enhancements regard the

agent local scheduler, and the communication middle-

ware properly coupled with a new negotiation proto-

col based on concepts such as utilization factor and

resource reservation (Calvaresi et al., 2017a).

REFERENCES

Aknine, S. (1998). Issues in cooperative systems: Ex-

tending the contract net protocol. In Intelligent Con-

trol (ISIC), 1998. Held jointly with IEEE Interna-

tional Symposium on Computational Intelligence in

Robotics and Automation (CIRA), Intelligent Systems

and Semiotics (ISAS), Proceedings, pages 582–587.

Aknine, S., Pinson, S., and Shakun, M. F. (2004). An ex-

tended multi-agent negotiation protocol. Autonomous

Agents and Multi-Agent Systems.

Alberti, M., Daolio, D., Torroni, P., Gavanelli, M., Lamma,

E., and Mello, P. (2004). Speciﬁcation and veriﬁcation

of agent interaction protocols in a logic-based system.

In Proceedings of the 2004 ACM symposium on Ap-

plied computing, pages 72–78. ACM.

Atkinson, K., Bench-Capon, T., and Mcburney, P. (2005). A

dialogue game protocol for multi-agent argument over

proposals for action. Autonomous Agents and Multi-

Agent Systems, 11(2):153–171.

Bellifemine, F. L., Caire, G., and Greenwood, D. (2007).

Developing multi-agent systems with JADE, volume 7.

John Wiley & Sons.

Budaev, D., Amelin, K., Voschuk, G., Skobelev, P., and

Amelina, N. (2016). Real-time task scheduling for

multi-agent control system of uav’s group based on

network-centric technology. In Control, Decision

and Information Technologies (CoDIT), 2016 Interna-

tional Conference on, pages 378–381. IEEE.

Calvaresi, D., Cesarini, D., Sernani, P., Marinoni, M., Drag-

oni, A., and Sturm, A. (2016a). Exploring the ambient

assisted living domain: a systematic review. Journal

of Ambient Intelligence and Humanized Computing,

pages 1–19.

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

233

Calvaresi, D., Claudi, A., Dragoni, A., Yu, E., Accattoli, D.,

and Sernani, P. (2014). A goal-oriented requirements

engineering approach for the ambient assisted living

domain. In Proceedings of the 7th International Con-

ference on PErvasive Technologies Related to Assis-

tive Environments, PETRA ’14, pages 20:1–20:4.

Calvaresi, D., Marinoni, M., Sturm, A., Schumacher, M.,

and Buttazzo, G. (2017a). The challenge of real-

time multi-agent systems for enabling iot and cps. in

proceedings of IEEE/WIC/ACM International Confer-

ence on Web Intelligence (WI’17).

Calvaresi, D., Schumacher, M., Marinoni, M., Hilﬁker, R.,

Dragoni, A., and Buttazzo, G. (2017b). Agent-based

systems for telerehabilitation: strengths, limitations

and future challenges. In proceedings of X Workshop

on Agents Applied in Health Care.

Calvaresi, D., Sernani, P., Marinoni, M., Claudi, A., Balsini,

A., Dragoni, A. F., and Buttazzo, G. (2016b). A

framework based on real-time os and multi-agents for

intelligent autonomous robot competitions. In 2016

11th IEEE Symposium on Industrial Embedded Sys-

tems (SIES), pages 1–10.

Cardoso, R. C. and Bordini, R. H. (2016). Allocating social

goals using the contract net protocol in online multi-

agent planning. In Intelligent Systems (BRACIS), 2016

5th Brazilian Conference on, pages 199–204. IEEE.

Chang, M. K. and Woo, C. (1994). A speech-act-based ne-

gotiation protocol: design, implementation, and test

use. ACM Transactions on Information Systems.

Collins, J., Tsvetovat, M., Mobasher, B., and Gini, M.

(1998a). Magnet: A multi-agent contracting system

for plan execution. In Proc. of SIGMAN.

Collins, J. and Wolfgang Ketter, M. G. (2002). A multi-

agent negotiation testbed for contracting tasks with

temporal and precedence constraints. International

Journal of Electronic Commerce, 7(1):35–57.

Collins, J., Youngdahl, B., Jamison, S., Mobasher, B., and

Gini, M. (1998b). A market architecture for multi-

agent contracting. In Proceedings of the second in-

ternational conference on Autonomous agents, pages

285–292. ACM.

Demazeau, Y. (1995). From interactions to collective be-

haviour in agent-based systems. In Proceedings of the

1st. European Conference on Cognitive Science.

El Fallah-Seghrouchni, A., Haddad, S., and Mazouzi, H.

(1999). Protocol engineering for multi-agent inter-

action. In European Workshop on Modelling Au-

tonomous Agents in a Multi-Agent World, pages 89–

101. Springer.

Faratin, P., Sierra, C., and Jennings, N. R. (1998). Ne-

gotiation decision functions for autonomous agents.

Robotics and Autonomous Systems, 24:159–182.

Ferber, J. and Gutknecht, O. (1998). A meta-model for

the analysis and design of organizations in multi-agent

systems. In Multi Agent Systems, 1998. Proceedings.

International Conference on, pages 128–135. IEEE.

Galster, M., Weyns, D., Tofan, D., Michalik, B., and Avge-

riou, P. (2014). Variability in software systems-a sys-

tematic literature review. IEEE Transactions on Soft-

ware Engineering, 40(3):282–306.

Garcia, E., Cao, Y., and Casbeer, D. W. (2017). Periodic

event-triggered synchronization of linear multi-agent

systems with communication delays. IEEE Transac-

tions on Automatic Control, 62(1):366–371.

Golfarelli, M., Maio, D., and Rizzi, S. (1997). A task-

swap negotiation protocol based on the contract net

paradigm. DEIS, Universit

a di Bologna.

Guttman, R. H. and Maes, P. (1998). Cooperative vs. com-

petitive multi-agent negotiations in retail electronic

commerce. In International Workshop on Coopera-

tive Information Agents, pages 135–147. Springer.

Hanachi, C. and Sibertin-Blanc, C. (2004). Protocol mod-

erators as active middle-agents in multi-agent sys-

tems. Autonomous Agents and Multi-Agent Systems,

8(2):131–164.

Hong-tao, L. and Kang, F.-j. (2016). Distributed task allo-

cation modeling based on agent topology and protocol

for collaborative system. Optik-International Journal

for Light and Electron Optics, 127(19):7776–7781.

Hsieh, F.-S. (2002). Modeling and control of holonic manu-

facturing systems based on extended contract net pro-

tocol. In American Control Conference, 2002. Pro-

ceedings of the 2002, volume 6, pages 5037–5042.

Ito, T., Hattori, H., and Klein, M. (2007). Multi-issue nego-

tiation protocol for agents: Exploring nonlinear utility

spaces. In IJCAI, volume 7.

Ito, T., Klein, M., and Hattori, H. (2008). A multi-issue ne-

gotiation protocol among agents with nonlinear utility

functions. Multiagent and Grid Systems.

Ito, T. and Shintani, T. (1997). Persuasion among agents:

An approach to implementing a group decision sup-

port system based on multi-agent negotiation. In In-

ternational Joint Conference on Artiﬁcial Intelligence,

volume 15, pages 592–599.

Jennings, N. R., Faratin, P., Lomuscio, A. R., Parsons, S.,

Wooldridge, M., and Sierra, C. (2001). Automated ne-

gotiation: prospects, methods and challenges. Group

Decision and Negotiation, pages 199–215.

Jian, L. (2008). An agent bilateral multi-issue alternate

bidding negotiation protocol based on reinforcement

learning and its application in e-commerce. In Int.

Symposium on Electronic Commerce and Security,,

pages 217–220.

Kanchanasevee, P., Biswas, G., Kawamura, K., and Tamura,

S. (1999). Contract-net based scheduling for holonic

manufacturing systems. PhD thesis.

Kitchenham, B., Brereton, P., Turner, M., Niazi, M.,

Linkman, S., Pretorius, R., and Budgen, D. (2010).

Reﬁning the systematic literature review process-two

participant-observer case studies. Empirical Software

Engineering, 15(6):618–653.

Kitchenham, B., Pearl Brereton, O., Budgen, D., Turner,

M., Bailey, J., and Linkman, S. (2009). Systematic lit-

erature reviews in software engineering - a systematic

literature review. Information and Software Technol-

ogy, 51(1):7–15.

Kraus, S. (1997). Negotiation and cooperation in multi-

agent environments. Artiﬁcial intelligence, 94(1-

2):79–97.

ICAART 2018 - 10th International Conference on Agents and Artiﬁcial Intelligence

234

Krothapalli, N. K. C. and Deshmukh, A. V. (1999). Design

of negotiation protocols for multi-agent manufactur-

ing systems. International journal of production re-

search, 37(7):1601–1624.

Marzougui, B. and Barkaoui, K. (2013). Interaction proto-

cols in multi-agent systems based on agent petri nets

model. Interaction, 4(7):2013.

Matt, P.-A., Toni, F., and Dionysiou, D. (2006). The

distributed negotiation of egalitarian resource alloca-

tions. In Proceedings of the 1st international work-

shop on computational social choice (COMSOC06),

pages 304–316.

Mazouzi, H., Seghrouchni, A. E. F., and Haddad, S. (2002).

Open protocol design for complex interactions in

multi-agent systems. In Proc. of 1st. international

joint conference on Autonomous agents and multia-

gent systems.

Odell, J., Parunak, H. V. D., and Bauer, B. (2000). Extend-

ing uml for agents. Ann Arbor, 1001:48103.

Odell, J. J., Parunak, H. V. D., and Bauer, B. (2001). Repre-

senting agent interaction protocols in uml. In Agent-

oriented software engineering.

Olfati-Saber, R., Fax, J. A., and Murray, R. M. (2007). Con-

sensus and cooperation in networked multi-agent sys-

tems. Proceedings of the IEEE, 95(1):215–233.

Ouelhadj, D., Garibaldi, J., MacLaren, J., Sakellariou, R.,

Krishnakumar, K., and Meisels, A. (2005). A multi-

agent infrastructure and a service level agreement ne-

gotiation protocol for robust scheduling in grid com-

puting. In EGC.

Purvis, M., Nowostawski, M., Oliveira, M., and Crane-

ﬁeld, S. (2003). Multi-agent interaction protocols of

e-business. In International Conference on Intelligent

Agent Technology. IEEE/WIC, pages 318–324. IEEE.

Rajkumar, R. R., Lee, I., Sha, L., and Stankovic, J. (2010).

Cyber-physical systems: the next computing revolu-

tion. In Proceedings of the 47th Design Automation

Conference, pages 731–736. ACM.

Ramchurn, S. D., Huynh, D., and Jennings, N. R. (2004).

Trust in multi-agent systems. The Knowledge Engi-

neering Review, 19(1):1–25.

Russell, S., Norvig, P., and Intelligence, A. (1995). A mod-

ern approach. Artiﬁcial Intelligence. Prentice-Hall,

Egnlewood Cliffs, 25:27.

Sandholm, T. (1993). An implementation of the contract

net protocol based on marginal cost calculations. In

AAAI, volume 93, pages 256–262.

Shen, W. and Norrie, D. H. (1998). An agent-based ap-

proach for dynamic manufacturing scheduling. In

Proc. of Workshop on Agent-Based Manufacturing.

Singh, A., Juneja, D., and Sharma, A. (2010). Introducing

trust establishment protocol in contract net protocol.

In Advances in Computer Engineering (ACE), 2010

International Conference on, pages 59–63. IEEE.

Smith, R. G. (1980). The contract net protocol: High-level

communication and control in a distributed problem

solver. IEEE Transactions on computers.

Smith, R. G. and Davis, R. (1981). Frameworks for coop-

eration in distributed problem solving. IEEE Transac-

tions on systems, man, and cybernetics.

Sun, D. and Wu, J. (2009). Multi-agent coordination based

on contract net protocol. In Int. Symposium on Intelli-

gent Ubiquitous Computing and Education.

Vok

ınek, J., B

ıba, J., Hod

ık, J., Vyb

ıhal, J., and P

echou

cek,

M. (2007). Competitive contract net protocol. Theory

and Practice of Computer Science.

Vulkan, N. and Jennings, N. R. (2000). Efﬁcient mecha-

nisms for the supply of services in multi-agent envi-

ronments. Decision Support Systems, 28(1):5–19.

Wan, W.-n., Zhang, J.-q., and Wang, M. (2007). A multi-

agent negotiation protocol based on extended case

based reasoning. In IV International Conference on

Fuzzy Systems and Knowledge Discovery, volume 4,

pages 462–467. IEEE.

Wang, G., Wong, T., and Wang, X. (2014). A hybrid multi-

agent negotiation protocol supporting agent mobility

in virtual enterprises. Information Sciences, 282.

Wanyama, T. and Far, B. H. (2007). A protocol for multi-

agent negotiation in a group-choice decision making

process. Journal of Network and Computer Applica-

tions, 30(3):1173–1195.

Wellman, M. P. and Wurman, P. R. (1998). Market-aware

agents for a multiagent world. Robotics and Au-

tonomous Systems, 24(3-4):115–125.

Wu, J. (2008). Contract net protocol for coordination in

multi-agent system. In Intelligent Information Tech-

nology Application, 2008. IITA’08. Second Interna-

tional Symposium on, volume 2, pages 1052–1058.

Xueguang, C. and Haigang, S. (2004). Further extensions

of ﬁpa contract net protocol: threshold plus doa. In

Proceedings of the 2004 ACM symposium on Applied

computing, pages 45–51. ACM.

Zhou, P.-C., Hong, B.-R., Wang, Y.-H., and Zhou, T. (2004).

Multi-agent cooperative pursuit based on extended

contract net protocol. In Machine Learning and Cy-

bernetics, 2004. Proceedings of 2004 International

Conference on, volume 1, pages 169–173. IEEE.

Zlotkin, G. and Rosenschein, J. S. (1991). Incomplete in-

formation and deception in multi-agent negotiation. In

IJCAI, volume 91, pages 225–231.

Multi-Agent Systems’ Negotiation Protocols for Cyber-Physical Systems: Results from a Systematic Literature Review

235