1513 Journal of Intelligent & Fuzzy Systems 31 (2016) 1513–1520 DOI:10.3233/JIFS-151214 IOS Press A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles Luiz F.F. de Oliveira∗ , Fernando B. de Lima, Sergio C. Oliveira and Carmelo J.A. Bastos-Filho University of Pernambuco, Av. Benfica, Madalena, Recife, Pernambuco, Brazil Abstract. In this paper, we propose a hybrid mechanism based on fuzzy logics and swarm intelligence for the coordination of multiple Unmanned Aerial Vehicles (UAVs). In a previous work we have developed a finite state machine to control the swarm of robots by pre-defining expected behaviors. The main goal of our proposal is to replace the finite state machine by a TakagiSugeno fuzzy mechanism that smoothly changes the expected behavior of the UAVs aiming to obtain the dynamic behavior required by real-time critical systems. We deployed three different metrics to compare our proposal to the previous one, such as the target tracking capability, the anti-collision rate and the cohesion rate aiming to catch the most important aspects desired in a group of UAVs tracking targets. We performed simulations varying the number of UAVs in the environment and the number of targets to be tracked. The results indicate that the proposed mechanism diminished the number of collisions and increased the cohesion rate. Keywords: Unmanned aerial vehicles, fuzzy logics, swarm intelligence, Takagi-Sugeno, hybrid algorithms 1. Introduction Unmanned Aerial Vehicles (UAV) have been deployed for many real-world applications, including tasks that may put human being lives in risk. As a consequence of this potentiality, UAVs have attracted the attention in different research fields during the last years. Despite the vast applicability of the UAVs, there are situations in which the usage of a single UAV might not be enough to accomplish a given mission. This may occur in tasks where the coverage area is too large or when the robustness of the entire system is crucial [10]. In these cases, it is quite common to deploy a group of UAVs working together aiming to reach a global goal. This coordinated groups are often called as swarms of robots or swarms of UAVs. In [5], a layered architecture is proposed to control not only low and mid-level tasks but also high-level ∗ Corresponding author. Luiz F.F. de Oliveira, University of Pernambuco, Av. Benfica, 455 Madalena, Recife, Pernambuco, Brazil. E-mail: lffo@ecomp.poli.br. functions such as the distribution of roles in the group of UAVs, coalition formation, and behavior planning, so the UAVs are capable of performing complex tasks in different scenarios. The architecture is divided into different modules, and each module is supposed to solve specific tasks varying in level of complexity. The low-level module is responsible for dealing with trajectory tracking control, which is based on the Riccati equation with state-dependent coefficients. Quintero et al. [14] consider the coordination of two UAVs equipped with cameras responsible for measuring the target’s position in a noisy environment. The UAVs control actions are computed based on noisy measurements of the UAV’s state and the positions of the target. The target must always be at the center of the camera’s field of view and must travel at a constant velocity. The UAV coordination model is based on a combination of model predictive control (MPC) with moving horizon estimation (MHE). In [3], Cichella et al. focus on a solution to the coordinated vision based tracking (CVBT) problem, so a set of n vehicles can track and monitor a ground 1064-1246/16/$35.00 © 2016 – IOS Press and the authors. All rights reserved 1514 L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles moving target. A control algorithm is derived for the angle rate of the UAVs to allow the vehicles to track the target, and rotate around it. Besides, control laws were formulated for the vehicles’ speed so that the UAVs can keep a certain distance from each other while orbiting around the targets. The UAVs are considered to have an autopilot to adjust the yaw rates and ground speed commands. Two other papers [1, 4] tackle communication issues that are essential for the formation of missions containing multiple UAVs. However, we did not consider theses aspects in this paper. Swarms of UAVs require a coordination system in order to guarantee the proper functionality of the robots during the mission. One possible approach to tackle this is the application of swarm intelligence concepts for the coordination. Swarm intelligence seems to fit these requirements since it can provide non-centralized control by exchanging only local information, thus generating the emergence of a global behavior [8]. As an example of the application of swarm intelligence we can cite the use of an autonomous coordination strategy for the detection of pollution sources. Varela et al. [7] used the collaborative aspects of swarm intelligence to spread information acquired by the agents in order to find new sources of pollution. Another example is presented in [9], in which a multi-robot search algorithm is proposed to overcome the limitations in an environment where the GPS signal might not be available to define the positions of the UAVs. Other applications of swarm intelligence for the coordination of UAVs can be found in [2, 6, 11, 13]. Fuzzy systems have also been widely used for the design of UAV controllers due to the capability to introduce semantics for the control process and to proper treat the uncertainties presented in real environments. In [16], a Mamdani fuzzy controller is used to control a full flight of a UAV in self-pilot mode. In [12], a Takagi-Sugeno fuzzy framework is introduced to tackle the tracking problem in missions deploying a swarm of UAVs with linear and angular velocity constraints. However, this framework can not manage different expected behaviors, such as patrol mode or returning to the base mode in order to charge the energy system. Silva et al. [15] proposed a distributed coordination model for UAVs based on the Particle Swarm Optimization (PSO) algorithm. In this model, the UAVs have the following objectives: (i) establish and maintain the communication with their neighbors in order to provide an ad-hoc network among the UAVs; (ii) avoid obstacles and collisions with other UAVs in the environment; and (iii) track detected targets in the environment. This first model of coordination was structured with a finite state machine (FSM), responsible for defining which acceleration components should be activated and/or deactivated during the missions according to the UAV required behavior. Although this model presents some interesting features, such as tracking targets in the environment in a distributed way and self-organizing capability in order to keep the connection with neighbor UAVs, we observed that some collisions still occur under some circumstances. We believe this occurs since the FSM do not provide flexibility and can not predict dangerous situations. Based on the proposal of Silva et al. [15], this paper proposes to replace the FSM by a simple system that uses Takagi-Sugeno fuzzy controllers to smoothly define the acceleration components of the UAVs. The FSM, previously responsible for completely activating or deactivating the acceleration components, now gives place to controllers capable to set the activation levels of the acceleration components with values in the interval [0,1]. The major benefit of the proposed approach is to avoid the rigid definition of extreme behaviors imposed by the FSM. 2. PSO inspired coordination model for swarms of UAVs The coordination model proposed in [15] is composed by the following features: (i) localization mechanism, from which the UAVs obtain their position in the environment, xuav (t) using the location sensor; (ii) tracking mechanism, responsible for the detection and tracking of moving targets in the environment using tracking sensors; (iii) communication mechanism, responsible for maintaining the connectivity between neighbors using the communication sensor; (iv) and anti-collision mechanism, responsible for the avoidance of collisions with obstacles and other UAVs in the environment using the anticollision sensor. The locomotion mechanism features the aerodynamic aspects of the UAVs in a two-dimensional environment, such as horizontal acceleration (a), horizontal speed (v) and maximum speed (v). The acceleration vector (auav ) is composed by the acceleration vectors originated from the UAV coordination mechanisms: synchronism (asyn ) and tracking (atrk ). 1515 L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles The synchronism acceleration is given by: (1) (3) where xtar (t) is the target position provided by the tracking sensor. The resultant coordination acceleration for the UAV is given by the sum of the synchronism and the tracking acceleration vectors: auav (t) = γcol acol (t) + γcom acom (t) + γtrk atrk (t) (4) where γcol , γcom and γtrk are the activation levels of each acceleration component. For the proposal introduced by Silva et al. [15], the values for the activation levels are determined by the FSM for each expected behavior (here defined as states of the FSM) as shown in Table 1. Finally, the resultant velocity of the UAV is given by: v(t + 1) = v(t) · ω + auav (t + 1) (5) where v(t + 1) is the new velocity, v(t) is the current velocity and ω is the inertia factor. Table 1 UAV states in the FSM and its values for the activation levels, defined by Silva et al. [15] We introduce a fuzzy input variable for each one of these variables, named: Communication Distance (DCom ), Collision Distance (DCol ) and Tracking Distance (DTrack ). We define a membership function for these three variables. The fuzzy variable DCom is represented by two fuzzy sets: Close and Far, and its membership functions are depicted in Fig. 1. We just used two fuzzy membership functions in this case since we just want to trigger this mechanism to adjust the position if other UAVs are too close (it might generate a collision) or too far (so it may lose the communication with other UAVs) from the reference UAV. The fuzzy variable DCol is represented by three fuzzy sets: Unsafe, Near and Safe, and its membership functions are depicted in Fig. 2. Finally, the fuzzy variable DTrack is represented by two fuzzy sets: TargetClose and TargetFar, and its membership functions are depicted in Fig. 3. 0 State Patrol Tracking Return to base station Power Recharging γcol γcom γtrk 1 1 1 0 1 1 0 0 0 1 0 0 Far Close 1.00 Fitnessuav (t) = |xtar (t) − xuav (t)| 1. The communication distance between UAVs i and j: dcom−ij (t); 2. the collision distance between UAVs i and j: dcol−ij (t); 3. and the tracking distance between UAV i and target j: dtrk−ij (t); 0.75 where acog (t) is the cognitive acceleration and is related to information generated by the sensors of the UAV, while asoc (t) is the social acceleration and is related to information provided by the UAVs in the neighborhood. The cognitive and social accelerations are calculated by the PSO algorithm at each iteration. The fitness function of the PSO is given by the euclidian distance between the UAV and the target being tracked: 0.50 (2) 0.25 atrk (t) = acog (t) + asoc (t) This section specifies the fuzzy controllers designed for the coordination mechanism. The fuzzy controllers are responsible for calculating the level of activation for the three acceleration components of the UAV, represented by the following coefficients γcol , γcom e γtrk . The proposed control mechanism is based on the following continuous variables: Membership degree where acol (t) is the anti-collision acceleration and acom (t) is the communication acceleration. The tracking acceleration is given by: 0.00 asyn (t) = acol (t) + acom (t) 3. Takagi-Sugeno based fuzzy controllers for the acceleration components 4 8 10 16 20 DCom (m) Fig. 1. Representation of the membership functions related to the fuzzy variable Communication Distance (DCom ) and its respective fuzzy sets: Close and Far. 1516 L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles 0.25 0.50 0.75 1.00 Rule 3: IF DCol ∈ Safe, THEN y3 = 0.25 + colnear ; Safe 0.00 Membership degree Near Unsafe 0 3 5 6 7 9 10 DCom (m) Fig. 2. Representation of the membership functions related to the fuzzy variable Collision Distance (DCol ) and its respective fuzzy sets: Unsafe, Near and Safe. 1.00 0.75 0.50 0.25 0 2 3 6 8 where αk corresponds to the firing strengths of Rules 1, 2 and 3 respectively. The controller for the communication coefficient γcom consists of the following rules: Rule 4: IF DCom ∈ Close AND DCol ∈ Unsafe, THEN y4 = 1 − 2 · comclose ; Rule 5: IF DCom ∈ Far, THEN y5 = 1; Rule 6: IF DCom ∈ Close AND DCol ∈ Near, THEN y6 = comclose − 0.25 · colnear ; Rule 7: IF DCom ∈ Close AND DCol ∈ Safe, THEN y7 = comclose − 0.5 · colsafe ; 0.00 Membership degree TargetFar TargetClose where colunsafe represents the membership degree of the variable DCol in the fuzzy set Unsafe, colnear represents the membership degree of the variable DCol for the fuzzy set Near. The collision coefficient γcol is given by (6): 3 αk yk γcol = k=1 (6) 3 k=1 αk 10 DTrack (m) Fig. 3. Representation of the membership functions related to the fuzzy variable Tracking Distance (DTrack ) and its respective fuzzy sets: TargetClose and TargetFar. Figure 1 shows the fuzzy sets Close and Far. For the fuzzy set Close, the membership degree is maximum for DCom < 4 and decays linearly from 1 to 0 as DCom increases from 4 to 10. We defined these values based on the UAVs available in our laboratory and some preliminary simulation experiments. For the fuzzy set Far, the membership degree is maximum for DCom > 16 and decays linearly from 1 to 0 as DCom decreases from 16 to 8. The same analysis can be made regarding the fuzzy sets Unsafe, Near and Safe in Fig. 2 and the fuzzy sets TargetClose and TargetFar in Fig. 3. For our proposal, the values for the coefficients γcol , γcom e γtrack are determined by the outputs of the fuzzy controllers inspired in the Takagi-Sugeno defuzzyfication process. The controller for the collision coefficient γcol consists of the following rules: Rule 1: IF DCol ∈ Unsafe, THEN y1 = 1; Rule 2: IF DCol ∈ Near, THEN y2 = 2 + colunsafe ; where comclose represents the membership degree of the variable DCom in the fuzzy set Close, colnear represents the membership degree of the variable DCol for the fuzzy set Near and colsafe represents the membership degree of the variable DCol for the fuzzy set Safe. The communication coefficient γcom is given by (7): 7 αk y k (7) γcom = k=4 7 k=4 αk where αk corresponds to the firing strengths of the Rules 4, 5, 6, and 7, respectively. Finally, the controller for the tracking coefficient γtrk consists of the following rules: Rule 8: IF DTrk ∈ TargetClose, THEN y8 = 1 − 0.5 · targetclose ; Rule 9: IF DTrk ∈ TargetFar, THEN y9 = 1; where targetclose represents the membership degree of the variable DTrk in the fuzzy set TargetClose. The tracking coefficient γtrk is given by (8): 9 αk yk (8) γtrk = k=8 9 k=8 αk where αk corresponds to the firing strengths of the Rules 8 and 9, respectively. One has to observe that new values for all coefficients are obtained from the controllers at each L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles iteration of the process in real time. Therefore, the accelerations coefficients acol−i (t), acom−i (t) e atrk−i (t) are now weighted by different degrees of activation depending on the environment circumstances for each one of the UAVs. This fuzzy-based coordination mechanism is used alongside the PSO-inspired mechanism described in Section 2, denoting a hybrid coordination system for the swarm of UAVs. 4. Simulation setup and results The UAVs were modeled as agents in the simulation environment. The UAVs present a coordination model composed of localization sensors, anti-collision sensors, a wireless communication system with a pre-defined reach and a target monitoring system. In the UAV model, we define the size of the UAV in 2D space, the maximum reachable speed, and the maximum acceleration. The environment is described in squared meters and contains the UAVs, the targets and obstacles artificially defined. The entire environment is modeled in 2D for the experiments presented in this paper. The simulator was developed in JAVA language. We performed the simulations in a computer with an Intel Core i5 4570 Quad Core 3.20 GHz processor, with 8 GB of RAM memory and a ASUS Nvidia GeForce GT 640 2 GB DDR3 video card. We deployed three different metrics to assess the performance of the Swarms of UAVs: collision rate (CL), target tracking rate (TT) and cohesion rate (CO). These metrics were proposed in a previous work [15]. For each metrics, we performed simulations considering the finite state machine (FSM) approach introduced by Silva et al. [15] and the hybrid fuzzy-swarm (SF) approach proposed in this paper. We performed some simulations with a low number of UAVs to validate our proposal. The complexity to simultaneously control multiple UAVs increases as the number of UAVs also increase. Therefore, we would like to show that our approach can tackle this tough problem with many UAVs tracking many targets simultaneously. A large quantity of UAVs can be required for surveillance of large areas, such as plantations or forests, or monitoring electronically supervised prisoners. We chose to test 10, 20 and 40 UAVs to asses the scalability of the model. The number of UAVs used for each simulation configuration for both approaches were 10, 20 and 40, in an environment containing 1, 3 and 5 moving targets. 1517 Table 2 Parameters and the default values used in the simulations Parameter Description tmax env vuav−max auav−max euav w acog asoc nnei−max rcol rtrack rcom vtgt−max atgt−max Maximum number of iterations t Environment area UAV maximum speed UAV maximum acceleration UAV extension range Inertial weight Cognitive acceleration Social acceleration Maximum number of neighbors Anti-collision sensor range Tracking sensor range Communication sensor range Target maximum speed Target maximum acceleration Value 3600 10,000 m2 0.5 m/s 0.3 m/s2 0.1 m 0.9 1.0 2.0 2 10 m 10 m 20 m 0.5 m/s 0.3 m/s2 For each simulation configuration we performed 30 independent trials. The default values for the parameters used in all simulations are presented in Table 2. 4.1. Collision rate The collision rate (CL) represents the percentage of collided UAVs during the simulation. This metrics is given by Equation (9): CL = imax  1 nuav−col (t), imax · nuav (9) t=1 where imax is the maximum number of iterations, nuav is the total number of UAVs at the beginning of the simulation and nuav−col is the total number of UAVs that have suffered a collision until iteration t. The simulation results for the collision rate (CL) are presented in Fig. 4 for the FSM-based (FSM) and the swarm-fuzzy (SF) approaches. From Fig. 4 considering the cases with 10 or 20 UAVs, it is possible to observe a drastic reduction in the CL. For the FSM approach, most of the experiments shows a lot of collisions (see Fig. 4(a), (b) and (c)). On the other hand, our proposal presented CL equal to 0%, with some outliers. For the cases with 40 UAVs, one can observe a slight increase in the collision rate even for our proposed model. It is important to note that the environment area was not increased to bear the increased amount of UAVs. Even though, while the FSM approach presents an increase on the collision rate from 10% to 20% when increasing the amount of UAVs from 20 to 40, the swarm-fuzzy model presents an increase of 5% on average. 1518 L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles Fig. 4. Collision rates for 10, 20 and 40 UAVs in environments with 1, 3 and 5 targets. Fig. 5. Target tracking rates for 10, 20 and 40 UAVs in environments with 1, 3 and 5 targets. 4.2. Target tracking rate to the fact that this approach presents higher collision rates, which leaves less UAVs in the environment to track the targets. The target tracking rate represents the percentage of targets tracked by the UAVs during the simulation. A target j is tracked by the UAV i when the distance between the position of the target and the position of the UAV is smaller than the tracking sensor range, rtrack . This metrics is given by Equation (10): ntgt imax   1 · tt tgt−j (t), TT = imax · ntgt 4.3. Cohesion rate The cohesion rate (COE) represents the proportion of connected UAVs in the swarm. This metrics is given by (11): (10) t=1 j=1 where imax is the maximum number of iterations, ntgt is the total number of targets and tt tgt−j (t) indicates whether the target j in iteration t is being tracked, assuming the value of 1 if the target is being tracked, and 0 otherwise. The simulation results for the target tracking rate (TT) are presented in Fig. 5 for the FSM-based (FSM) and the swarm-fuzzy (SF) approaches. From the results for TT in Fig. 5, it is possible to observe that the tracking rate are quite similar for both approaches. However, it is important to notice that the FSM approach may undergo a decrease in performance in an environment with a larger number of targets due COE = imax  1 nuav−con (t), imax · nuav (11) t=1 where imax is the maximum number of iterations, nuav is the total number of UAVs and nuav−con (t) is the number of UAVs that have the maximum number of connected neighbors in the iteration t. The simulation results for the cohesion rate (COE) are presented in Fig. 6 for the FSM-based (FSM) and the swarm-fuzzy (SF) approaches. It is possible to observe in the results for COE in Fig. 6 that the swarm-fuzzy approach presents an increase of 10% in average in comparison with the finite state machine approach, reaching a maximum of 25% of increase in an environment with 10 UAVs. L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles 1519 It is important to note that the proposed approach is capable of maintaining a higher connectivity rate among the UAVs because it also presents a smaller collision rate even when the number of UAVs is increased in the environment, which indicates a more stable behavior than the FSM approach. We performed some simulations with a lower number of UAVs to validate our proposal. We performed simulations with 2 UAVs. In this case, collisions were not detected, and the target tracking is one hundred percent. For three UAVs, collisions were not detected, and the target tracking and cohesion are much high for our proposal, as can be seen in Fig. 7. We also compared our proposal with the FSM approach using the Wilcoxon signed-rank nonparametric statistical test considering p = 5%. We assessed 10, 20 and 40 UAVs with 1, 3 and 5 targets. Regarding COE, our model achieved better results in all cases. Regarding TT, our model achieved better results for 20 UAVs with 1 and 3 targets and 40 UAVs in all cases. Regarding COL, our model achieved better results in all cases. Fig. 6. Cohesion rates for 10, 20 and 40 UAVs in environments with 1, 3 and 5 targets. 5. Conclusions Fig. 7. Results for 3 UAVs in environments with 1, 3 and 5 targets. This work proposes a hybrid mechanism using fuzzy logics and swarm intelligence for the coordination of a swarm of UAVs. The controllers are inspired by the Takagi-Sugeno fuzzy controllers and aim to replace a finite state machine previously responsible for determining the activation levels of the acceleration components of the UAVs. The use of fuzzy controllers decouples the static influence of a state machine from the system. This is possible due to the continuous nature of Takagi-Sugeno fuzzy controllers, making the UAVs to quickly react to dangerous situations. We performed simulations considering 3, 10, 20 and 40 UAVs aiming to track 1, 3 and five targets simultaneously. The proposed model presented a considerable reduction in the collision rates and an increase in the cohesion rates, proving to be more robust and stable. Although no significant improvements in the tracking rates were achieved for some cases, the proposed model is less prone to a decrease in performance with a higher number of UAVs in the environment. We also compared our proposal with the FSM approach using the Wilcoxon signed-rank non-parametric statistical test, and we observed that proposal is always better regarding avoiding collisions and maintaining the cohesion of the swarm. 1520 L.F.F. de Oliveira et al. / A fuzzy-swarm based approach for the coordination of unmanned aerial vehicles References [1] M. Balduccini, D.N. Nguyen and W.C. Regli, Coordinating uavs in dynamic environments by network-aware mission planning, In Military Communications Conference (MILCOM), 2014, pp. 983–988. [2] A.A. Bandala, R.R.P. Vicerra and E.P. Dadios, Formation stabilization algorithm for swarm tracking in unmanned aerial vehicle (uav) quadrotors, TENCON 2014 - 2014 IEEE Region 10 Conference, 2014, pp. 1–6. [3] V. Cichella, I. Kaminer, V. Dobrokhodov and N. Hovakimyan, Coordinated vision-based tracking for multiple uavs, IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2015, pp. 656–661. [4] B.J.O. de Souza and M. Endler, Coordinating movement within swarms of uavs through mobile networks, 2015 IEEE International Conference on Pervasive Computing and Communication Workshops (PerCom Workshops), 2015, pp. 154–159. [5] S. Emel’yanov, D. Makarov, A.I. Panov and K. Yakovlev, Multilayer cognitive architecture for uav control, Cognitive Systems Research (2015). [6] A. Jevtic, et al., Unmanned aerial vehicle route optimization using ant system algorithm, 2010 5th International Conference on System of Systems Engineering (SoSE), 2010, pp. 1–6. [7] G. Varela, et al., Swarm intelligence based approach for real time uav team coordination in search operations, 2011 Third World Congress on Nature and Biologically Inspired Computing (NaBIC), 2011, pp. 365–370. [8] M. Dorigo, et al., Swarmanoid: A novel concept for the study of heterogeneous robotic swarms, Robotics Automation Magazine, IEEE 20(4) (2013), 60–71. [9] Q. Zhu, et al., A pso-inspired multi-robot search algorithm independent of global information, 2011 IEEE Symposium on Swarm Intelligence (SIS), 2011, pp. 1–7. [10] S. Gade and A. Joshi, Heterogeneous uav swarm system for target search in adversarial environment, 2013 International Conference on Control Communication and Computing (ICCC), 2013, pp. 358–363. [11] R. Hao, D. Luo and H. Duan, Multiple uavs mission assignment based on modified pigeon-inspired optimization algorithm, 2014 IEEE Chinese Guidance, Navigation and Control Conference (CGNCC), 2014, pp. 2692–2697. [12] G.P. Kladis, P.P. Menon and C. Edwards, Cooperative tracking for a swarm of unmanned aerial vehicles: A distributed tagaki-sugeno fuzzy framework design, 2011 50th IEEE Conference on Decision and Control and European Control Conference (CDC-ECC), 2011, pp. 4114–4119. [13] J. Leonard, A. Savvaris and A. Tsourdos, Towards a fully autonomous swarm of unmanned aerial vehicles, 2012 UKACC International Conference on Control (CONTROL), 2012, pp. 286–291. [14] S.A.P. Quintero, D.A. Copp and J.A.P. Hespanha, Robust uav coordination for target tracking using output-feedback model predictive control with moving horizon estimation, American Control Conference (ACC), 2015, pp. 3758–3764. [15] D.M.P.F. Silva, L.F.F. Oliveira, M.G.M. Macedo and C.J.A. Bastos-Filho, On the analysis of a swarm intelligence based coordination model for multiple unmanned aerial vehicles, 2012 Brazilian Robotics Symposium and Latin American Robotics Symposium (SBR-LARS), 2012, pp. 208–213. [16] D. Stojcsics, Fuzzy controller for small size unmanned aerial vehicles, 2012 IEEE 10th International Symposium on Applied Machine Intelligence and Informatics (SAMI), 2012, pp. 91–95.