Conﬂicts Resolution and Situation Awareness in Heterogeneous

Multi-agent Missions using Publish-subscribe Technique and Inferential

Reasoning

Sagir Muhammad Yusuf and Chris Baber

University of Birmingham, B15 2TT, U.K.

Keywords:

Multi-agent Reasoning, Bayesian Networks, Multi-agent Learning, Publish-subscribe, Heterogeneous

Multi-agent Coordination, Cooperative Perception.

Abstract:

In this paper, we propose a priority-based publish-subscribe approach to tackle reasoning in beliefs conﬂicts for

a heterogeneous multi-agent mission. Agents subscribe to other agents’ topics and rank them based on agents’

situation awareness. Bayesian Belief Network (BBN) was used in maintaining agents’ belief and recorded

mission information could be used for the BBN training using conjugate gradient descent or expectation-

maximization algorithms. The output of the training is the learned network for agents’ predictions, estimations,

and conclusions. We also propose an agent’s self presumption inferential reasoning where agents learned

heuristics and used them for future inferences. We test the system by using a team of heterogeneous Unmanned

Aerial Vehicles (UAVs) with different sensor proﬁles and capacities tasked together to perform forest ﬁre

searching. To verify belief and settle conﬂicts, agents follow these steps: sequentially assess the prioritized

publish-subscribe topics, inferential reasoning using the learned network, inferential reasoning using logical

propositions, and learning process. From our experiment, the BBN training and prediction perfection grow

up with the increase in the number of training data. Future work focuses on obtaining the optimal number

of samples needed for effective prediction, effective agents’ beliefs merging, communication protocol, and

bandwidth utilization.

1 INTRODUCTION

The heterogeneous multi-agents mission comprises of

a team of different agents with different beliefs (sen-

sor data), roles, and capacities tasked collectively to

achieve a particular goal. It has a potential advan-

tage of categorizing agents’ based on their special-

ization, backup provision, and robustness (Cort

es and

Egerstedt, 2017; Setter and Egerstedt, 2017; Khan

et al., 2015; Setter and Egerstedt, 2017; Yanmaz et al.,

2017). For example, consider a team of aerial and

grounds robots tasked to conduct forest ﬁre ﬁght-

ing, agents can divide the task into extinguishing,

ﬁre spreading monitoring, rescuing, and so on. Use

of heterogeneous agents provides a wider opportu-

nity for detecting false alarm due to sensor variation

(Merino et al., 2006). For example, agents carrying

camera sensors may have different beliefs from agents

using object sensors in the vehicle overload detection

system. The agents using facial recognition can de-

tect sitting on each other’s lap passengers and count

them as two, while agents using an object sensor may

count it as one passenger. Another agent may use a

weight balance sensor to count passengers based on

their masses. A challenge arises in merging beliefs

derived from different sensors to make an optimal cor-

rect decision that will support agents’ cost functions.

In every multi-agent mission, energy, mission time,

communication link, and other resources need to be

utilized.

We tackled this problem by applying a priority-

based publish-subscribe technique model using

Bayesian inferential reasoning and Distributed Situa-

tion Awareness (Endsley, 1995; Stanton et al., 2006).

The idea is that each agent has its own beliefs. It

will then subscribe to other relevant topics. Agent

belief variation is sorted out using probabilistic prior-

ity value. The probabilistic priority value rises with

the number of right decision and agents’ situation-

awareness (i.e., how agents currently perceive the en-

vironment). It changes from agents to agents based on

the agents’ environmental adaptation because some

agents perform better than other agents in different

scenes. For example, agents using thermal sensors

846

Yusuf, S. and Baber, C.

Conﬂicts Resolution and Situation Awareness in Heterogeneous Multi-agent Missions using Publish-subscribe Technique and Inferential Reasoning.

DOI: 10.5220/0009147408460851

In Proceedings of the 12th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2020) - Volume 2, pages 846-851

ISBN: 978-989-758-395-7; ISSN: 2184-433X

could perform better than the one using camera sensor

in detecting ﬁre during day time because of the pos-

sibility of having ﬁre-like terrains (e.g., dried grass).

The reverse could be the case when the agents are op-

erating on higher altitudes where heat could not be

sensed effectively.

We applied Bayesian Belief Network (Pavlin

et al., 2010; Wang and Xu, 2014; Williamson, 2001)

in monitoring the agents’ beliefs and their causal

relationships hard-coded during pre-mission plan-

ning. The BBN’s nodes are categorized into situa-

tion, awareness, and utility nodes in the BBN. The

agents’ situation nodes are the current belief of the en-

vironment. For example, ﬁre present or absenteeism

based on sensor information. Awareness node is the

conﬁrmed situation; that is decided belief that under-

gone veriﬁcation process from different agents using

different sensors that, ﬁre is present or not at a par-

ticular location. The utility node is a measure of

how welcome the agents are with the proposed be-

liefs. We modelled the agents’ utility as Distributed

Constraint Optimization (DCOP). DCOP is the be-

haviour of the agents towards the optimal assignment

of their variables with the aim of minimizing cost

functions(Fioretto et al., 2018; Fransman et al., 2019;

Maheswaran et al., 2004; Zhou et al., 2018).

Every agent has topics to broadcast to the network

(i.e., its sensor data) and subscribe to other agents’

topics in order to be authenticating its beliefs to a gen-

uine knowledge of the environment. The environment

may keep changing, and communication may be lim-

ited, which is a potential challenge of such architec-

ture. However, we propose a knowledge-base infer-

ential reasoning and Bayesian learning to solve this

issue.

Inferential reasoning allows the agents to predict,

estimate, and draw conclusions based on previous

experience (Fransman et al., 2019; Wang and Xu,

2014; Williamson, 2001). Bayesian inference com-

putes predictions based on the probabilities of the

prior variables using conditional probabilities (Wang

and Xu, 2014). Another mode of inferential reasoning

we want to discuss is the agents’ presumption infer-

ence. In this approach, agents build in their knowl-

edge in the form of if-then logical propositions rules,

then use that knowledge to make predictions and con-

clusions. For example, in multi-agent search and res-

cue missions, if an agent sees its co-agents hovering

over a place instead of usual navigation, it can per-

ceive that something interesting is present in that lo-

cation and act to support that agent.

In this paper, we tackle the problem of agents’

belief variation conﬂict using the publish-subscribe

technique. The agents’ beliefs are modelled us-

ing Bayesian Belief Network, and agents’ resource

utilization uses DCOP. We propose Bayesian and

heuristic-based inferential reasoning in monitoring

the agents’ belief conﬂict. We use heterogeneous

multi-agent coordination to conduct wildﬁre monitor-

ing as the use case.

The rest of this paper was organized as follows.

Section 2 describes the basics background of publish-

subscribe, inferential reasoning, and agents’ reason-

ing towards resource utilization (DCOP). Section 3

describes the related work and summary of our con-

tribution. Section 4 describes the ontology of the sys-

tems using forest ﬁre monitoring as the use case and

experiments with their results. Finally, section 5 de-

scribes the conclusions and future works.

2 BACKGROUND

2.1 Publish-subscribe Multi-agent

Interaction

In a multi-agent system, the publish-subscribe tech-

nique categorizes the agents into senders (publishers)

and receivers (subscribers). Agents subscribe to other

co-agents’ topics (information) if it is helpful and in-

teresting to them (Hackney and Clayton, 2015; Rivera

et al., 2016). Agents might have a maximum number

of subscriptions and topics to be broadcast into the

network. As such, agents prioritize their publishers

based on information saliency and context-awareness

(i.e., current environmental situation). Agents use

sends and acknowledge protocols to monitor mes-

sages delivery and update a list of subscribers (Rivera

et al., 2016). That is, agents may change their pub-

lishers or subscribers over time based on the environ-

mental changes.

Moreover, the agents’ coordination architecture

can be centralized or decentralized. In a centralized

approach, agents are controlled by a server, as such

publishing and subscribing management became eas-

ier. The challenges of this approach is that the cen-

tral server needs much computational power, mem-

ory, communication bandwidth, and other resources

if the number of agents increases (Cort

es and Egerst-

edt, 2017; Saicharan et al., 2016; Turpin et al., 2014;

Vasile and Zuiani, 2011). In a decentralized approach,

agents have the full right to control their values and

use two types of communications approaches, that is,

explicit and implicit messaging (Gerkey and Mataric,

2002). In explicit messaging, agents share informa-

tion when they are close to each other. In implicit

communication, agents do not send messages; rather,

Conﬂicts Resolution and Situation Awareness in Heterogeneous Multi-agent Missions using Publish-subscribe Technique and Inferential

Reasoning

847

they sense other co-agents within a few inches as

in swarm of ants, birds, etc. (Ferrante et al., 2015;

Reynolds, 1987; Saicharan et al., 2016).

We applied a priority-based multi-agents publish-

subscribe model of interactions. Agents subscribe

to other co-agents’ topics and prioritise its belief on

that publisher based on its specialisation, situation-

awareness, or data saliency. For example, in ﬁre

searching missions by a team of heterogeneous UAVs

carrying different sensor proﬁles. Let us assume the

agents are carrying an infrared, thermal, and visual

camera to detect the ﬁre. If the agents with thermal

sensors subscribe to the agents with infrared and cam-

era sensors, it can be updating their priorities based

on the situation awareness (Endsley, 1995; Stanton

et al., 2006). During day time (sunny time), informa-

tion from the agents with visual sensors could have

higher priority than the agent with an infrared sen-

sor. However, the reverse is the case during night

time. Therefore, Distributed Situation Awareness -

DSA (Stanton et al., 2006) changes the agents’ data

saliency (i.e., their topics priority). The overall archi-

tecture was represented in ﬁgure 1.

Figure 1: Agents Subscription Layer.

Figure 1 describes the subscription protocol in a

top-down passion. We assumed that the agents could

access topics and the current environmental condition

(situation awareness of the environment). Based on

this situation, agents reason and assign priority to top-

ics from their publishers against a hard-corded rule.

It means topics priority changes with environmental

change. When an agent needs topics from different

publishers to verify its belief, priority would be used

in making a decision (i.e., making conclusion with the

highest priority topics).

2.2 Inferential Reasoning

We assume that the agents are accessible to the global

belief based on the Bayesian Belief Network (BBN)

of ﬁgure 2. They need to verify their beliefs based

on that network. We use forest ﬁre detection using

heterogeneous UAVs mounted with different sensors

as the use case.

Figure 2: Bayesian Belief Network for Multi-agent Mis-

sion.

Figure 2 describes the agents’ Bayesian Belief

Network for forest ﬁre detection. The system has

four different sensors (i.e., dry material, temperature,

colour, and wind sensors); agents carrying these sen-

sors broadcast their sensor information (as topics) of

which other co-agents can subscribe to for their belief

veriﬁcation. For example, agent using colour (cam-

era) sensor can subscribe to other agents using dry

material, wind, and temperature sensors to be verify-

ing its beliefs against false detection e.g., detecting

object with the same ﬁre colour palettes. It could be

verify simply by using votes technique. Sensor nodes

are the situation node, while other nodes (decision

nodes) are the awareness nodes. Utility nodes could

be attached for DCOP cost function optimization (i.e.,

nodes to be grading how decision nodes favour cost

functions). From ﬁgure 2, four types of sensors were

used in the system. The ﬁre has two sensors, that

is, heat sensors and colour (visual sensors). Different

agents are carrying these sensors, and each can raise

a false detection based on the operating environment.

For example, yellowish dried grass can confuse the

agents using a visual sensor; agents can conﬁrm their

belief from topics broadcast by other co-agents using

heat or dried material sensors and prioritise the top-

ics of those agents during daytime (i.e., the concept

of situation awareness). All agents could subscribe to

the services of the agent providing wind direction in

order to know the direction of the spread of the ﬁre.

In centralized systems, agents could easily access

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

848

and update their topics on time to time basis. In a

decentralized approach, the agent sends request con-

ﬁrmation to their publishers or a medium agent (bro-

ker) to extend their messages. A variation of this ap-

proach is in (Merino et al., 2006). Cases from pre-

vious missions could be gathered and stored as text,

cas, or csv ﬁle for network training using machine

learning algorithms. The well-known machine learn-

ing algorithms to be used for the BBN training are

expectation-maximization and gradient descent algo-

rithms because they handle missing and uncertain

data (Romanycia, 2019). The output network is a

well-trained BBN for making predictions, estimation,

avoiding redundant veriﬁcation (request for verifying

the same belief within a short time), and conclusion.

It reduces the communication cost of the interaction

and improves context-based reasoning and optimiza-

tion. Another approach is to use the agents’ self-

presumption inference technique. That is, build the

knowledge of the environment as if-then rules. The

set of conditions are mapped to their various actions.

3 RELATED WORK

Agents’ belief veriﬁcation is a well-known problem in

multi-agent system. Different approaches and strate-

gies were used in solving the problem. Merino et

al (Merino et al., 2006) describe a ﬁve steps (i.e.,

detection, alarming, conﬁrmation, localization, and

evaluation) approach for multi-agent belief veriﬁca-

tion. They used a team of heterogeneous Unmanned

Aerial Vehicles (UAVs) mounted with different sen-

sors to conduct forest ﬁre monitoring in order to

test the model. Agents raise an alarm, then other

agents verify the claim. The agents will mark the

area and evaluate the ﬁre risk. They implemented a

blackboard communication approach for belief veri-

ﬁcation. Yanguas-Rojas and Mojica-Nava (Yanguas-

Rojas and Mojica-Nava, 2017) describe a multi-agent

searching problem in which space was segmented into

a set of Voronoi cells. Agents Voronoi size allocation

is based on the agent’s capacity (heterogeneity of the

agents).

Moreover, Lumelsky and Harinarayan (Lumelsky

and Harinarayan, 1997) solve the problem using the

cock-tail party model. In this approach, agents navi-

gate to other agents they want to consult and share in-

formation, a similar approach to the publish-subscribe

technique. In (Gerkey and Mataric, 2002) and (Yan

et al., 2011), a similar approach was used, in which

agent was categorized as buyers and sellers or auc-

tioneers and bidders, respectively. Agents will be

broadcasting their knowledge and beliefs for other

agents to bid (buy) based on assigned protocols.

In this paper, we propose a priority-based publish-

subscribe approach for agents’ belief variation han-

dling. It uses a changing priority technique to

make decisions in order to tackle the problem in

dynamic environment. We also suggested Bayesian

and agents’ self-presumption inferential reasoning for

making predictions, learning, and adaption purposes.

We claim that our approach reduces communication

cost (as agents can make optimal predictions with-

out contacting other agents), can work in a dynamic

environment, and improve belief authentication and

agents adaptation. A potential challenge of this ap-

proach arises in belief fusion and communication fail-

ure management.

4 THE PROPOSE MODEL

Our model tackles the problem of agents’ belief vari-

ation due to sensor differences using priority-based

publish-subscribe, Bayesian inference, and learning

technique. It uses four steps:

• Priority-base Publish-Subscribe Technique.

• Bayesian Learning

• Bayesian Inference, and

• Agents self presumption Inference

The task for the agents is to understand differ-

ent knowledge, current context (situation-awareness),

and topics to subscribe to in order to have genuine

knowledge of the environment. Figure 1 describes the

structure of generating the priority value of the topics.

For example, in heterogeneous multi-agent missions

for forest ﬁre searching, during day time, agents with

sensors not affected by sunlight or the nature of the

scene will have higher priority such as those carrying

cameras or heat sensors if the terrain has confusing

scene such as dried grasses with coloured palettes like

a ﬁre. The agents with the camera will have a lower

priority. The priority increase with an increase in the

right decisions made and decrease with an increase in

false information (false data). If agent detect a scene,

it will listen to all its publisher’s interesting topics,

prioritize them, and make a decision by selecting the

one with the highest priority. Equivalence in priority

means both options are the same (have the same effect

on cost function utilization). It made the approach

to be adaptable for the agents. Bayesian Belief Net-

work could be constructed (without ﬁlling in the con-

ditional probabilities table) by a human expert e.g.,

using Netica (Romanycia, 2019) and put as an in-built

add-in for the agents’ memory. Simple if then rules

Conﬂicts Resolution and Situation Awareness in Heterogeneous Multi-agent Missions using Publish-subscribe Technique and Inferential

Reasoning

849

will be used in monitoring the agents’ behaviours dur-

ing the mission. For example, if an agent detects a

ﬁre, it will update its BBN to take the effect. This al-

low an autonomous approach for ﬁlling in the BBN

conditional probability tables during the learning pro-

cess. Bayesian learning could be used in obtaining a

well-trained network for making predictions, estima-

tions, and conclusions in the absence of reliable top-

ics or communication. That is, if agents have no up-

dated topics from their publishers, that will be used to

verify their beliefs, they can use the network for pre-

dictions, estimations, and decision-making purposes.

We conducted and experiments using wildﬁre moni-

toring to monitor agents data on simulation platform.

From the results, we claimed that the network predic-

tion accuracy grows up with the increase in training

data for both expectation-maximization and conjugate

gradient descent algorithms used, as shown in ﬁgure

3, though we are still working on understanding the

nature of the growth of the prediction error perfection

and data utilization.

Figure 3: Prediction error rate reduces by the number of

samples.

Figure 3 describes the network training perfection

using 1, 10, 100, 1000, 10000, 100000, and 1000000

number of samples (mission data). From ﬁgure 3, us-

ing 1 sample data, the network gives prediction per-

fection fully (therefore should not be used). The pre-

diction error rate is the number of samples predicted

wrong (Romanycia, 2019).

Bayesian inference: agents could use the network

to predict less important matter or in the case where

communication within the system is limited, or no up-

dated topics. We claimed that for a static environ-

ment mission, the network could make good predic-

tions that are as nice as received information from the

publisher (from our Aerospace Multi-agent Simula-

tion Environment results and netica). In the case of a

dynamic system, the agents use the time-base learn-

ing algorithm such a context-based gradient descent

algorithm (Bottou, 2010; Romanycia, 2019). It con-

siders recent cases with higher priority, as such made

it to handle environment dynamism.

Agents’ self presumption inference is the learned

or in-built set of knowledge (in form of if-then rules)

for agents’ reference and decision making. In order

word it is a tuple P:

P = {A, C, β, α} (1)

Where A is the set of agents, C is the set of condi-

tions or logical proportions, β is the function for map-

ping conditions to actions, and α is the corresponding

actions. Agents update their self presumption on time

bases and learned new rules based on the environmen-

tal interaction and sensor data. Therefore, the agents’

option starts from topic information, inferred value,

and agents’ self presumption. Human expert and sen-

sor information will be responsible for updating the

rules of the system.

5 CONCLUSIONS AND FUTURE

WORK

We propose a multi-agent belief variation handling

architecture using four main steps, priority-based

publish-subscribe, Bayesian learning, Bayesian infer-

ence, and agents’ self-presumption inference based on

in-built or learned parameter. We claimed that (from

our experiment) this approach reduces the communi-

cation cost, system failure, and improve adaptability

based on environmental changes. Because the learned

network could be used to make an optimal predic-

tions, estimations, and conclusions on current and fu-

ture events during the agents mission. This could

remove too much communication and computation

process in other call-and-conﬁrm approaches such as

in (Gerkey and Mataric, 2002; Merino et al., 2010;

Merino et al., 2006; Yan et al., 2011) etc. The learn-

ing process also handles environment dynamism by

prioritising recent cases in learning process over older

cases and instant update of the BBN policies.

Future work focused on investigating the optimal

number of samples for BBN training and nature of

growing of the prediction accuracy of the learned net-

work. Information fusion management also remain

a challenge for our proposed architecture which is

to be solved later. Other challenges are incomplete

information due to communication failure or insuf-

ﬁcient resources, communication protocol manage-

ments, etc.

We are also intended to innovate a way of solving

redundant belief veriﬁcation. That is, agents to avoid

double belief authentication and deadlock occurrence

solution more especially in a dynamic and uncertain

situation. Finally, we will look at the effective tech-

nique for making decision more especially in a dy-

namic and uncertain situation. This aspect refers to

ICAART 2020 - 12th International Conference on Agents and Artiﬁcial Intelligence

850

effective application of game theory in agents belief

veriﬁcation.

ACKNOWLEDGEMENTS

We appreciate any comments that help in making this

paper a great one. Our thanks also goes to Petroleum

Technology Development Fund (PTDF) of Nigeria

for sponsoring this project.

REFERENCES

Bottou, L. (2010). Large-scale machine learning with

stochastic gradient descent. page 10.

Cort

es, J. and Egerstedt, M. (2017). Coordinated con-

trol of multi-robot systems: A survey. SICE Jour-

nal of Control, Measurement, and System Integration,

10(6):495–503.

Endsley, M. R. (1995). Toward a theory of situation aware-

ness in dynamic systems. [Online; accessed 2019-11-

14].

Ferrante, E., Turgut, A. E., Du

nez-Guzm

an, E., Dorigo,

M., and Wenseleers, T. (2015). Evolution of self-

organized task specialization in robot swarms. PLOS

Computational Biology, 11(8):e1004273.

Fioretto, F., Pontelli, E., and Yeoh, W. (2018). Distributed

constraint optimization problems and applications: A

survey. Journal of Artiﬁcial Intelligence Research,

61:623–698. arXiv: 1602.06347.

Fransman, J., Sijs, J., Dol, H., Theunissen, E., and

De Schutter, B. (2019). Bayesian-dpop for continu-

ous distributed constraint optimization problems. AA-

MAS ’19, page 1961–1963, Richland, SC. Interna-

tional Foundation for Autonomous Agents and Mul-

tiagent Systems. event-place: Montreal QC, Canada.

Gerkey, B. and Mataric, M. (2002). Sold!: auction meth-

ods for multirobot coordination. IEEE Transactions

on Robotics and Automation, 18(5):758–768.

Hackney, C. R. and Clayton, A. I. (2015). Unmanned aerial

vehicles (uavs) and their application in geomorphic

mapping.

Khan, A., Yanmaz, E., and Rinner, B. (2015). Information

exchange and decision making in micro aerial vehicle

networks for cooperative search. IEEE Transactions

on Control of Network Systems, 2(4):335–347.

Lumelsky, V. and Harinarayan, K. (1997). Decentral-

ized motion planning for multiple mobile robots: The

cocktail party model. Autonomous Robots, 4(1):121–

135.

Maheswaran, R. T., Pearce, J. P., and Tambe, M. (2004).

Distributed algorithms for dcop: A graphical-game-

based approach. page 8.

Merino, L., Caballero, F., Dios, J. R. M.-d., Ferruz, J., and

Ollero, A. (2006). A cooperative perception system

for multiple uavs: Application to automatic detection

of forest ﬁres. Journal of Field Robotics, 23(3-4):165–

184.

Merino, L., Caballero, F., Mart

ınez-de Dios, J. R., Maza, I.,

and Ollero, A. (2010). Automatic forest ﬁre monitor-

ing and measurement using unmanned aerial vehicles.

page 15.

Pavlin, G., de Oude, P., Maris, M., Nunnink, J., and Hood,

T. (2010). A multi-agent systems approach to dis-

tributed bayesian information fusion. Information Fu-

sion, 11(3):267–282.

Reynolds, C. W. (1987). Flocks, herds and schools: A

distributed behavioral model. SIGGRAPH ’87, page

25–34, New York, NY, USA. ACM. [Online; accessed

2019-04-21].

Rivera, d. D. S., Alcarria, R., Andr

es, d. D. M., S

anchez-

Picot, l., S

anchez, B. B., and Robles, T. (2016).

Distributed query results and iot data in a publish-

subscribe network implementing user notiﬁcations.

2016 30th International Conference on Advanced In-

formation Networking and Applications Workshops

(WAINA), pages 778–783.

Romanycia, M. (2019). Netica-j reference manual. page

119.

Saicharan, B., Tiwari, R., and Roberts, N. (2016). Multi

objective optimization based path planning in robotics

using nature inspired algorithms: A survey. pages 1–

6. 2016 IEEE 1st International Conference on Power

Electronics, Intelligent Control and Energy Systems

(ICPEICES).

Setter, T. and Egerstedt, M. (2017). Energy-constrained co-

ordination of multi-robot teams. IEEE Transactions

on Control Systems Technology, 25(4):1257–1263.

Stanton, N. A., Stewart, R., Harris, D., Houghton, R. J.,

Baber, C., McMaster, R., Salmon, P., Hoyle, G.,

Walker, G., Young, M. S., Linsell, M., Dymott, R.,

and Green, D. (2006). Distributed situation awareness

in dynamic systems: theoretical development and ap-

plication of an ergonomics methodology. Ergonomics,

49(12-13):1288–1311.

Turpin, M., Michael, N., and Kumar, V. (2014). Capt: Con-

current assignment and planning of trajectories for

multiple robots. The International Journal of Robotics

Research, 33(1):98–112.

Vasile, M. and Zuiani, F. (2011). Multi-agent collabora-

tive search : an agent-based memetic multi-objective

optimization algorithm applied to space trajectory de-

sign. Proceedings of the Institution of Mechanical En-

gineers, Part G: Journal of Aerospace Engineering,

225:1211–1227.

Wang, J. and Xu, Z. (2014). Bayesian inferential reasoning

model for crime investigation. page 11.

Williamson, J. (2001). Bayesian networks for logical rea-

soning. page 19.

Yan, Z., Jouandeau, N., and Ch

erif, A. A. (2011). Multi-

robot decentralized exploration using a trade-based

approach.

Yanguas-Rojas, D. and Mojica-Nava, E. (2017). Explo-

ration with heterogeneous robots networks for search

and rescue.

Yanmaz, E., Yahayanajeed, S., and Bernerd, R. (2017).

Drone networks: Communications, coordination, and

sensing. Elsevier.

Zhou, X., Wang, W., Tao, W., Xiaboo, L., and Tian, J.

(2018). Continuous patrolling in uncertain environ-

ment with the uav swarm. [Online; accessed 2019-08-

18].

Conﬂicts Resolution and Situation Awareness in Heterogeneous Multi-agent Missions using Publish-subscribe Technique and Inferential

Reasoning

851