computational intelligence in control

Annual Reviews in Control xxx (2014) xxx–xxx

Contents lists available at ScienceDirect

Annual Reviews in Control

journal homepage: www.elsevier .com/ locate /arcontrol

Review

Computational intelligence in control

http://dx.doi.org/10.1016/j.arcontrol.2014.09.0061367-5788/� 2014 Elsevier Ltd. All rights reserved.

⇑ Corresponding author.E-mail addresses: [email protected] (A.E. Ruano), [email protected] (S. Sam Ge),

[email protected] (T.M. Guerra), [email protected] (F.L. Lewis),[email protected] (J.C. Principe), [email protected] (M. Colnaric).

Please cite this article in press as: Ruano, A. E., et al. Computational intelligence in control. Annual Reviews in Control (2014), http://dx.doi.org/1j.arcontrol.2014.09.006

António E. Ruano a,⇑, Shuzhi Sam Ge b, Thierry Marie Guerra c, Frank L. Lewis d, Jose C. Principe e,Matjaz Colnaric f

a Centre for Intelligent Systems, IDMEC, IST and Univ. of Algarve, 8005-139 Faro, Portugalb Social Robotics Laboratory, Inter. Digital Media Inst., Dep. Electrical & Computer Eng., Nat Univ. of Singapore, 117576 Singapore, Singaporec Univ. Valenciennes Hainaut-Cambresis, LAMIH UMR CNRS 8201, Le Mont Houy, Valenciennes, Franced Univ. Texas Arlington, Automat & Robot Res Inst, 7300 Jack Newell Blvd S, Ft Worth, TX 76118, USAe Univ. Florida, Dept. Elect. & Comp. Engn., Gainesville, FL 32611, USAf Univ. of Maribor, Fac. for El. Eng. and Comp. Sci., Smetanova 17, 2000 Maribor, Slovenia

a r t i c l e i n f o

Article history:Received 26 May 2014Accepted 20 June 2014Available online xxxx

a b s t r a c t

This Milestone Report addresses first the position of the areas of computers, computational intelligenceand communications within IFAC. Subsequently, it addresses the role of computational intelligence incontrol. It focuses on four topics within the Computational Intelligence area: neural network control,fuzzy control, reinforcement learning and brain machine interfaces. Within these topics the challengesand the relevant theoretical contributions are highlighted, as well as expected future directions arepointed out.

� 2014 Elsevier Ltd. All rights reserved.

Contents

1. The coordinating committee for computers, cognition and communications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
1.1. The future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
2. Computational intelligence in control. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 003. Neural network control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

3.1. Challenges and future research directions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
4. Fuzzy control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 005. Reinforcement learning in feedback control systems . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
5.1. The future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
6. Brain machine interfaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00
6.1. Example of BMIs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 006.2. The future . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

6.2.1. System identification and control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 006.2.2. Augmented optimal control . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

7. Concluding remarks . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 00

1. The coordinating committee for computers, cognition andcommunications

This paper intends to present the state of the art and the out-look of some important domains of computational intelligence in

control, falling under the domain of IFAC TC 3.2. It will start, how-ever, with a more general view of computers, computational intel-ligence and telecommunications in control, and their global role inthe worlds present and future situation.

The Coordinating Committee for Computers, Cognition andCommunications (CC3) belongs to the group of IFAC’s applicationCCs. It consists of three Technical Committees, namely Computersfor Control (TC3.1), Computational Intelligence in Control (TC3.2),and Telematics: Control via Communication Networks (TC3.3). This

0.1016/

http://dx.doi.org/10.1016/j.arcontrol.2014.09.006

mailto:[email protected]







http://www.sciencedirect.com/science/journal/13675788

http://www.elsevier.com/locate/arcontrol



2 A.E. Ruano et al. / Annual Reviews in Control xxx (2014) xxx–xxx

introduction will give a more general view of CC 3 methodologiesin control and their global role.

Although computers for control, computational intelligence andcommunications do not present the main focus of interest in theInternational Federation for Automatic Control (IFAC), they havetraditionally played an important role, providing enabling technol-ogies: they can hardly be avoided in the implementation of controlmethodologies, techniques, and applications.

In the vision of European ARTEMIS platform and its collabora-tion with ITEA, there is an excellent description of the importanceof embedded computers and related fields within the contempo-rary world society:

‘‘Embedded Systems will be part of all future products and ser-vices providing intelligence on the spot and capabilities to clev-erly connect to the abundance of systems in the environment;either physical or at the cyber space level, in real time or in dis-tributed and heterogeneous systems, in open networks ofembedded systems applications from multiple domains or inthe Internet: everything can, in principle, be connected toeverything else. Networked intelligent embedded systems are,in effect, becoming the Neural System of Society.

[ARTEMIS-ITEA, 2012]

It is hard to imagine any technical field where computers wouldnot be employed as a cost-effective, relatively easy to implement,and flexible means for control. Often, they exhibit intelligence,and, except in trivial cases, they are distributed and connectedthrough communication networks.

In the same document (ARTEMIS-ITEA, 2012; Pétrissans et al.,2012), the seven most important challenges to the future worldsociety are identified: globalization and demographic change,management of scarce resources, climate change, urbanization,mobility, healthcare and nutrition, and digital society. There mayonly be a few exceptions where computers and control wouldnot play a major role.

Looking into the world economy, not many people realize howimportant embedded control systems are to the world market.Today, out of 100 processing devices, 98 will be employed inembedded systems. Over 40 billion devices are expected world-wide by 2020. Embedded systems accounted for almost €852 bil-lion of value in 2010, and €1.100 billion are expected (in 2012)worldwide by 2020 (ARTEMIS SRA, 2011). The overall industry isgrowing at a compound annual growth rate (CAGR) of 12%throughout the forecast period and should reach €1.5 trillion inrevenue by 2015 (Pétrissans et al., 2012).

It is also important to note that the value added by embeddedsystems exceeds the price of components by orders of magnitude,thus presenting an excellent opportunity to sell the built-in knowl-edge. In the automotive industry, for example, 80% of productinnovation includes embedded systems (Pétrissans et al., 2012).

There are many fields in the intersection of embedded com-puter, computer intelligence, communications and controldomains that experience enormous growth of interest. Because ofthe growing complexity, there is, e.g., a need to widen the under-standing of systems-of-systems, intelligent systems consisting ofcommunicating embedded HW (e.g., in traffic, smart grid, etc.).Cloud computing can provide support for much more powerfulubiquitous solutions. Through the internet of things, even smalldevices become a part of global systems; already in 2010, of 7 bil-lion devices connected to the internet, 5 billion were not comput-ers (Pétrissans et al., 2012).

At the more technical level, because of the novel challenges,new methods and tools are needed for faster and cheaper develop-ment of much more complex applications, as well as theirvalidation, (rigorous or formal) verification and certification.State-of-the-art hardware components still do not ensure full

Please cite this article in press as: Ruano, A. E., et al. Computational intelligenj.arcontrol.2014.09.006

temporal predictability even at the lowest layers. Through theubiquity and penetration into extremely safety critical applica-tions, more effort must be stressed on their security and safetyissues. The cyber-physical approach, well known for a decade, isgaining interest. Holistic understanding, modeling and implemen-tation of the increasingly complex controlled and control systemspromise more competent solutions.

Without doubt, there is a very strong motivation for researchand development in the area of computers, intelligence and com-munications in control. As mentioned before, these areas are notthe main focus of IFAC, which is primarily concerned with auto-matic control. Besides, there are other professional associationswhich cover the basic research in these areas. Many members ofthe Technical Committees are also members of those Institutions,attending also their technical events and publishing in their jour-nals. It is our opinion that, within IFAC, the synergies betweenthe basic domains of control and automation, and computing, cog-nition and communications should be strengthened. During thelast year, this has been taking place, as demonstrated by theincreasing number of IFAC co-sponsored conferences by the threeTCs within CC3, with TCs belonging to other CCs. There is, however,room to improve these synergies, whether by creating commonWorking Groups among TCs, or by proposing common special ses-sions at IFAC events, particularly at the World Congress.

It is interesting that the following opinion from the MilestoneReports prepared for the 2002 and 2005 IFAC World congresses isstill more or less true: ‘‘People are developing most sophisticatedcontrol algorithms but are less interested in the transfer of thesealgorithms to a well-defined real-time control system. Moreover,the application of control algorithms is in practice many times con-fined to the use of existing hard- and software solutions, which arein many cases application dependent or supplier dependent andsometimes developed for other purposes.’’ (Halang, Sanz,Babuska, & Roth, 2006; Verbruggen, Park, Halang, Irwin, &Zalewski, 2002).

1.1. The future

We strongly believe that it would be of mutual benefit to estab-lish better collaboration among all domains in IFAC, and in partic-ular within the areas of CC 3. Automation and control scientistsand professionals could much better utilize the expertise of themembers of CC3, and the latter would get challenging case studiesto solve within their areas of interest.

CC3 hosts several technical events. Currently, the only event inthe IFAC ‘‘master-plan’’ is the Telecommunications Applications(TA) Symposium. Another successful conference is the IntelligentControl and Automation Science (ICONS), which is being proposedto the IFAC Technical Board to also become a master-plan event.

In 2012, a new conference series was established, and had itsfirst successful event in Würzburg, Germany. The triennial Confer-ence for Embedded Systems, Computer intelligence and Telematics(CESCIT) is supposed to unite all events of the CC3 in one place, oneyear after the IFAC World Congress, thus fostering the communica-tion and synergy among those fields. Also, tracks for applications,industrial case studies, education and other related areas are orga-nized, providing an opportunity for scientists and professionalsfrom other fields to make bonds with the computer community.It is to be noted that the first CESCIT conference featured verystrong industrial emphasis, with excellent industrial keynotespeakers and an interesting industrial round table.

2. Computational intelligence in control

The IFAC Technical Committee on Computational Intelligence(CI) in Control (TC 3.2) focuses on all aspects of data fusion and data

ce in control. Annual Reviews in Control (2014), http://dx.doi.org/10.1016/



A.E. Ruano et al. / Annual Reviews in Control xxx (2014) xxx–xxx 3

mining, knowledge-based, fuzzy, neuro-fuzzy, neural (both artifi-cial and biologically plausible) systems, evolutionary algorithmsand swarm intelligence, relevant to control and automation, boththeoretically and application driven. The techniques used havestrong links to other fields, in particular machine-learning algo-rithms (Cortes & Vapnik, 1995; Scholkopf, Smola, Williamson, &Bartlett, 2000).

CI methodologies are currently applied to all areas of control.Recent work of TC members covers a spectrum of application areassuch as:

– transport (Fischer, Tibken, & Fischer, 2010; Statheros, Howells,& McDonald-Maier, 2008),

– medical and biomedical (Azar, El-Said, Balas, & Olariu, 2013;Chen, Luo, & Lin, 2013; Cismondi et al., 2012; Goodband,Haas, & Mills, 2008; Liu, Pang, & Lloyd, 2008; Nomm, Vain,Petlenkov, Miyawaki, & Yoshimitsu, 2009; Wojtowicz & Wajs,2012),

– biology (Chen et al., 2007; Gormley, Li, & Irwin, 2007; Paik, Lee,Park, Kim, & Lee, 2011; Teixeira, Ruano, Ruano, & Pereira, 2008),

– aerospace (Chwa, Choi, & Anavatti, 2006; Yang, Chen, & Li,2012),

– instrumentation (Benoit & Foulloy, 2013; Cerman & Husek,2012; Ferreira, Gomes, Martins, & Ruano, 2012; Omatu, Araki,Fujinaka, Yoshioka, & Nakazumi, 2012),

– biotechnology (Adamy & Schwung, 2010; Heikkinen, Latvala,Juuso, & Hiltunen, 2008; Koprinkova-Hristova & Palm, 2010),

– mechatronics (Garcia, Rolle, Gomez, & Catoira, 2013; Riccardi,Naso, Janocha, & Turchiano, 2012),

– automation and manufacturing (Boukef, Benrejeb, & Borne,2008; Dumitrache & Caramihai, 2008; Guillaume, Grabot, &Thierry, 2013; Tavakolpour, Darus, Tokhi, & Mailah, 2010;Wong, Tikk, Gedeon, & Koczy, 2005; Zribi, Kacem, El Kamel, &Borne, 2007),

– process control (Govindhasamy, McLoone, & Irwin, 2005; Karer,Music, Skrjanc, & Zupancic, 2007; Korosi & Kozak, 2013),

– power systems (Essounbouli, Manamanni, Hamzaoui, &Zaytoon, 2006),

– energy and smart grid (Ferreira, Ruano, Silva, & Conceicao,2012; Malatji, Zhang, & Xia, 2013),

– agriculture (Ferreira & Ruano, 2008),– environmental systems (Petelin, Grancharova, & Kocijan, 2013;

Vousdoukas et al., 2011),– robotics and autonomous systems (Chen, Sekiyama, & Fukuda,

2012; Guelton, Delprat, & Guerra, 2008; Marcos, Machado, &Azevedo-Perdicoulis, 2009; Sakaguchi, Horio, & Yamakawa,2011; Sanz, Hernandez, Hernando, Gomez, & Bermejo, 2009;Schuitema, Busoniu, Babuska, & Jonker, 2010; Shakev, Topalov,Kaynak, & Shiev, 2012; van de Ven, Johansen, Sorensen,Flanagan, & Toal, 2007),

– economics and business systems (Caloba, Caloba, & Contador,2001; Nagesh, Lendek, Khalate, & Babuska, 2012) and

– telematics (Pulkkinen, Laurikkala, Ropponen, & Koivisto, 2008).

This milestone paper, by lack of space, cannot cover all CI meth-odologies relevant to control. It will instead be focused on only fourareas: neural network control, fuzzy control, reinforcement learn-ing and brain machine interfaces, trying to highlight in these areasthe challenges, relevant theoretical contributions, as well as pointout expected future directions.

3. Neural network control

The basic idea of neural network is to study the computationalabilities of networks composed of simple models of neurons(McCulloch & Pitts, 1943). It has been successfully applied to sys-


tem control with its universal approximation and adaptation capa-bilities. The back-propagation method (BP) proposed in Rumelhart,Hinton, and Williams (1986) boosts the development of neural net-work control in a vast number of applications. Besides practicalimplementations, there is parallel research effort in rigorous anal-ysis of neural network control system on stability and robustness(Ge, Hang, & Zhang, 2001).

In early work of neural network control theory, much researcheffort was aimed at designing stable adaptive neural network con-trol for single-input-single-output (SISO) continuous-time systemsin affine forms (Ge et al., 2001). Due to the fact that most systemsin practice are of nonlinear and multi-variable characteristics,many recent studies have been conducted on control design forsystems of multi-input-multi-output (MIMO) nature or/and innon-affine forms. The extension of control designs for affine sys-tems to non-affine systems is generally non-trivial, because ofthe lack of corresponding mathematical tools, e.g., the counterpartof the feedback linearization technique for non-affine systems. Dueto couplings among inputs, outputs, and states, the extension ofcontrol designs for SISO systems to MIMO systems is also foundto be difficult. Since most of the control algorithms are realizedin digital form, neural network control design for discrete-timesystems has also attracted significant research interest (Ge, Yang,& Lee, 2008).

On the application side, neural network control has been suc-cessfully implemented in many practical systems (Ge, Lee, &Harris, 1998), such as robots, helicopters and hard disks. Manykinds of nonlinearities in practical systems such as hysteresis,time-delay, dead-zone, have been taken into account in the controldesign, because it has been demonstrated that neural network con-trol is particularly suitable for controlling highly uncertain nonlin-ear complex systems.

3.1. Challenges and future research directions

A broadly applicable methodology to develop a workable con-trol strategy for general systems is not yet available. Most of thecontributions mentioned in the previous section only focus onaddressing a single issue. For example, the discrete-time Nuss-baum gain is introduced to cope with the problem of unknowncontrol direction (Ge et al., 2008). For another example, implicitfunction theorem is used to handle the non-affine issue (Geet al., 2001). While it is reasonable to incrementally generalizethe problem formulation by accumulating different issues into asingle system, it is also essential to consider the control design inthe sense of a systemic framework.

Most of early work on neural network control focuses on a sin-gle objective of stabilization, regulation, and tracking. In manycases, control design of a system can be described as a multi-objec-tive control problem. For example, obstacle avoidance, targetreaching, and control effort minimization are all essential in theapplication of robotic exploration. In this sense, a control systemwhich takes all the requirements into consideration is expected.In recent years, optimization once again has become popular inthe control field, and adaptive dynamic programming (ADP) is atimely response which aims at a guaranteed optimized perfor-mance subject to unknown environments (Lewis, 2009; Werbos,2009). Many open problems such as ADP design for time-varyingand non-affine continuous systems need to be addressed.

Although neural network control has been well recognized byacademic researchers, it has not yet been embraced by engineerswho are very careful about the feasibility of an algorithm. Onemain concern to implement neural network control in practicalsystems is computation burden. To cope with this problem, muchresearch effort has been aimed at developing real-time computa-tional algorithms, in which fewer neurons and simpler networks





are expected. This research direction is believed to be still worthyof further investigation.

The study on the structure and adaptation/learning scheme ofneural network itself also attracts a great deal of research interest.More efficient and accurate approximation can be anticipated withbetter network structures and adaptation/learning schemes. As aresult, fundamental problems like local minimum may beaddressed to some extent and better control performance can beachieved. Recent work in this direction includes deep learning(Hinton & Salakhutdinov, 2006) and extreme learning (Huang,Zhu, & Siew, 2006), of which relatively few results have beenapplied to the control field.

4. Fuzzy control

We can reuse the question of (Zadeh, 2008) ‘‘Is there a need forfuzzy logic’’? to ask ‘‘Is there a need for fuzzy control?’’.

Fuzzy control/observation/diagnosis today is very often relatedto the so-called Takagi–Sugeno (TS) models (Takagi & Sugeno,1985). The approach of this kind of modeling is to interconnectfamilies of models via nonlinear functions having the nice convexsum property. When the outcome are linear models then, mostof the cases, from a point of view of automatic control they concernfamilies of Linear Parameter Variant (LPV) and quasi-LPV models,where the parameters depend on the state of the system (Tanaka& Wang, 2001). There are three main components in finding resultsfor this area: the way the nonlinear functions are taken intoaccount, including the so-called relaxations on co-positivity prob-lems, see (Sala & Arino, 2007) for an asymptotic solution via Polya’smethod; the Lyapunov functions used; and last the solvers andtechniques of solving the problems, generally Linear MatrixInequalities (LMI) or Sum-of-Squares (SOS) constraints imposedto get formulations that are in the pre-described form of LMIand/or SOS. Each of these three steps introduces conservatism inthe results.

Representing in Fig. 1 the whole set (unknown of course) ofproblems with a solution, the goal is twofold. One is to excludepoints that are unfeasible (cross mark out of the set, Fig. 1 left)and correspond to necessary conditions, the other to enlarge theset where a solution can be found using a fuzzy representation ofa nonlinear system (Fig. 1 right). For the first point very few resultsare available (Johansson, Rantzer, & Arzen, 1999; Kruszewski, Sala,Guerra, & Arino, 2009). Most of the results try to solve the secondproblem. Where are we today? Solutions are available for statefeedback and quadratic Lyapunov function under various possibil-ities/performances, H2, H-infinity, robustness with delays (seeDing, Sun, & Yang, 2006; Feng, 2006).

Output feedback is still a subject for some works in progress(Chadli & Guerra, 2012). As for quasi-LPV and nonlinear systems,it is harder to derive strict conditions that guarantee performance

Fig. 1. Representation of reachable solutions using T-S models. How to exclude ill conditisolutions (right)?


(input-to-stability, following trajectories, etc.); these points mustalso be addressed in the future. Significant attention has also beendevoted to non-quadratic Lyapunov functions with real successesin the discrete case (Ding et al., 2006; Kruszewski, Wang, &Guerra, 2008) and more limited ones in the continuous case(Mozelli, Palhares, Souza, & Mendes, 2009; Pan, Guerra, Fei, &Jaadari, 2012; Rhee & Won, 2006).

Another important part of activities in control is related toadaptive control where the property of universal approximationof fuzzy systems is used in order to approximate an ideal controllaw. When there is no explicit identification of the nonlinear sys-tem, the control is called direct, otherwise it is called indirect. Eventhough pioneering work, e.g. (Wang, 1993) is rather ‘‘old’’, it seemsthat these methods still suffer from technical limitations thatapparently are hard to solve. Among them are the necessity of aparticular Brunowski form for the nonlinear model and the draw-back that the states, i.e. generally the derivatives of the output(s),are supposed to be known. Several contributions try to solve theseproblems, see Boulkroune, Tadjine, M’Saad, and Farza (2008), Liu,Tong, and Li (2011), and Tong and Li (2009).

Some new trends arising from the so-called Type-2 Fuzzy Setsare entering the general area of control, TS and adaptive frame-works. These papers claim that Type-2 Sets are an efficient wayto cope with vagueness and imprecision and therefore they aremore suitable than classical Type-1 Fuzzy Sets (Mendel & John,2002). A paper reviews industrial applications related to Type-2Fuzzy Sets (Dereli, Baykasoglu, Altun, Durmusoglu, & Turksen,2011) trying to exhibit which problems do require such kind ofmodeling. Nevertheless, even if some results claim superiority overtype-1 or conventional linear controllers (see for example Atacak &Bay, 2012; Oh, Jang, & Pedrycz, 2011), these results need to beenforced with theoretical arguments. As long as these argumentsare not given – to the best of our knowledge, they are not –type-2 FS in control and observation does not really fill some exist-ing gaps.

The reader interested in applications related to the area of fuzzycontrol can find a good overview in Precup and Hellendoorn,(2011).

Last, several questions arise: how can we come back to the fun-damentals of fuzzy logics? Is there still room for a linguistic way ofdoing this? Or are we in a sense losing the ‘‘essence’’ of what fuzz-iness was created for? Of course, fuzzy control should also go to theareas where automatic control is moving to, among them, largescale systems, interconnected (networks) large systems, hybridsystems, etc.

It seems that, for the theoretical aspects of fuzzy control, we areat a crossroads where some new interesting idea has to emerge, inorder to go further than just some adjustments in various knowntechniques, so that we may say ‘‘yes there is a need for fuzzycontrol’’.

oned problems (left, cross mark outside the set)? How to enlarge the set of reachable





5. Reinforcement learning in feedback control systems

Adaptive control (Astrom & Wittenmark, 1995; Ioannou &Fidan, 2006) and optimal control (Lewis, Vrabie, & Syrmos, 2012)represent different philosophies for designing feedback controllers.Optimal controllers are normally designed offline by solving Ham-ilton–Jacobi–Bellman (HJB) equations, for example, the Riccatiequation, using complete knowledge of the system dynamics.Determining optimal control policies for nonlinear systemsrequires the offline solution of nonlinear HJB equations, whichare often difficult or impossible to solve. By contrast, adaptive con-trollers learn online to control unknown systems using data mea-sured in real time along the system trajectories. Adaptivecontrollers are not usually designed to be optimal in the sense ofminimizing user-prescribed performance functions. Indirect adap-tive controllers use system identification techniques to first iden-tify the system parameters, then use the obtained model to solveoptimal design equations (Astrom & Wittenmark, 1995). Adaptivecontrollers may satisfy certain inverse optimality conditions.

The computational intelligence technique known as reinforce-ment learning allows for the design of a class of adaptive control-lers with actor-critic structure that learn optimal control solutionsby solving HJB design equations online, forward in time, and with-out knowing the full system dynamics. In the linear quadratic case,these methods determine the solution to the algebraic Riccatiequation online, without specifically solving the Riccati equationand without knowing the system state matrix A. As such, thesecontrollers can be considered as being optimal adaptive control-lers. Chapter 11 of Lewis et al. (2012) places these controllers inthe context of optimal control systems.

Reinforcement learning is a type of machine learning developedin the Computational Intelligence Community in computer scienceengineering. It has close connections to both optimal control andadaptive control. More specifically, reinforcement learning refersto a class of methods that enable the design of adaptive controllersthat learn online, in real time, the solutions to user-prescribedoptimal control problems. Reinforcement learning methods wereused by Ivan Pavlov in the 1860s to train his dogs. In machinelearning, reinforcement learning (Lewis, Lendaris, & Liu, 2008;Powell, 2007; Sutton & Barto, 1998; Werbos, 1991) is a methodfor solving optimization problems that involves an actor or agentthat interacts with its environment and modifies its actions, orcontrol policies, based on stimuli received in response to itsactions. Reinforcement learning is inspired by natural learningmechanisms, where animals adjust their actions based on rewardand punishment stimuli received from the environment(Busoniu, Babuska, De Schutter, & Ernst, 2009; Mendel &MacLaren, 1970). Other reinforcement learning mechanisms oper-ate in the human brain, where the dopamine neurotransmitter inthe basal ganglia acts as a reinforcement informational signal thatfavors learning at the level of the neuron (Doya, Kimura, & Kawato,2001; Schultz, 2004; Vrabie & Lewis, 2009; Werbos, 1992).

Reinforcement learning implies a cause-and-effect relationshipbetween actions and reward or punishment. It implies goal-direc-ted behavior, at least insofar as the agent has an understanding ofreward versus lack of reward or punishment. The reinforcementlearning algorithms are constructed on the idea that effective con-trol decisions must be remembered, by means of a reinforcementsignal, such that they become more likely to be used a second time.Reinforcement learning is based on real-time evaluative informa-tion from the environment and could be called action-based learn-ing. Reinforcement learning is connected from a theoretical pointof view with both adaptive control and optimal control methods.

One type of reinforcement learning algorithms employs theactor-critic structure (Barto, 1984). This structure produces


forward-in-time algorithms that are implemented in real timewherein an actor component applies an action, or control policy,to the environment, and a critic component assesses the value ofthat action. The learning mechanism supported by the actor-criticstructure has two steps, namely, policy evaluation, executed by thecritic, followed by policy improvement, performed by the actor.The policy evaluation step is performed by observing from theenvironment the results of applying current actions. These resultsare evaluated using a performance index, or value function(Bellman, 1957; Bertsekas & Tsitsiklis, 1996; Busoniu et al., 2009;Powell, 2007; Sutton & Barto, 1998) that quantifies how close thecurrent action is to optimal. Performance or value can be definedin terms of optimality objectives such as minimum fuel, minimumenergy, minimum risk, or maximum reward. Based on the assess-ment of the performance, one of several schemes can then be usedto modify or improve the control policy in the sense that the newpolicy yields a value that is improved relative to the previous value.In this scheme, reinforcement learning is a means of learning opti-mal behaviors by observing the real-time responses from the envi-ronment to non-optimal control policies.

Werbos (1989), Werbos (1991, Werbos (1992) developed actor-critic techniques for feedback control of discrete-time dynamicalsystems that learn optimal policies online in real time using datameasured along the system trajectories. These methods, knownas approximate dynamic programming (ADP) or adaptive dynamicprogramming, comprise a family of four basic learning methods.The ADP controllers are actor-critic structures with one learningnetwork for the control action and one learning network for thecritic. Many surveys of ADP are available (Balakrishnan, Ding, &Lewis, 2008; Wang, Zhang, & Liu, 2009; Lewis, 2009; Prokhorov& Wunsch, 1997; Si, Barto, Powell, & Wunsch, 2004). Bertsekasand Tsitsiklis developed reinforcement learning methods for thecontrol of discrete-time dynamical systems (Bertsekas &Tsitsiklis, 1996). This approach, known as neuro-dynamic pro-gramming, used offline solution methods.

ADP has been extensively used in feedback control applications.Applications have been reported for missile control, automotivecontrol, aircraft control over a flight envelope, aircraft landing con-trol (Prokhorov & Wunsch, 1997), helicopter reconfiguration afterrotor failure, power system control, and vehicle steering and speedcontrol. Convergence analyses of ADP are provided in (Al-Tamimi,Lewis, & Abu-Khalaf, 2008; Liu & Balakrishnan, 2000).

A key object in applying reinforcement learning to the design offeedback control systems is the Bellman equation. The Bellmanequation results on setting the system Hamiltonian function equalto zero, and it captures the optimality properties with respect to agiven performance index of the system dynamics evolving throughtime. The temporal difference method (Sutton & Barto, 1998) forsolving Bellman equations leads to a family of optimal adaptivecontrollers, that is, adaptive controllers that learn online the solu-tions to optimal control problems without knowing the full systemdynamics. Temporal difference learning is true online reinforce-ment learning, wherein control actions are improved in real timebased on estimating their value functions by observing data mea-sured along the system trajectories. The basic families of algo-rithms provided by RL are Policy Iteration, Value Iteration, and Qlearning.

Most research over the years in reinforcement learning for feed-back control has been conducted for systems that operate in dis-crete time (Si et al., 2004; Sutton & Barto, 1998; Werbos, 1992).This is because the Bellman equation for discrete-time systemsdoes not involve the system dynamics, and has two occurrencesof the value at different times. This special form immediately lendsitself to both Policy Iteration and Value Iteration solution methodsthat do not require complete knowledge of the system dynamics.





Reinforcement learning is considerably more difficult for con-tinuous-time systems than for discrete-time systems, and fewerresults are available. This is due to the fact that, unfortunately,for continuous-time dynamical systems, the Bellman equationhas an inconvenient form that does involve the system dynamics.This hindered the development of RL for continuous-time systemsfor many years. The development of an offline policy iterationmethod for continuous-time systems is described in Abu-Khalaf,Lewis, and Huang (2006). The method known as integral reinforce-ment learning (IRL) developed in Vrabie and Lewis (2009) andVrabie, Pastravanu, Abu-Khalaf, and Lewis (2009) allows the appli-cation of reinforcement learning to formulate online optimal adap-tive control methods for continuous-time systems. These methodsfind solutions to optimal HJ design equations and Riccati equationsonline in real time without knowing the system drift dynamics, orin the LQR case, without knowing the A matrix. Extensions to IRLhave been made to solve optimal control problems online for sys-tems with completely unknown dynamics in (Jiang & Jiang, 2012;Lee, Park, & Choi, 2012). A method of RL on time scales (Seiffert,Sanyal, & Wunsch, 2008) allows the treatment of discrete-timeand continuous-time systems in the same framework.

5.1. The future

Reinforcement learning has been applied in the ComputationalIntelligence Community to solve complex decision problems withvery general performance indices and using various learning tech-niques including episodic Monte-Carlo Methods. Problems solvedinclude finding optimal solutions to games such as Backgammon,and solution of multiple-degrees-of-freedom systems such as thetruck back-up problem. Applications of RL to feedback control havegenerally used performance indices that are summations or inte-grals over time of basic utility functions. Such performance indicesmay not be suitable for optimal decision and control problemssuch as adaptive human–robot interfaces, discrete event schedul-ing and decision problems, and task learning skills for autonomoussystems. More general performance indices could capture theessence of such decision problems. More general learning schemescan be developed rather than the standard temporal differencelearning that relies on information taken at every time instance.It is not known how episodic learning can be used in feedback con-trol, though it seems to be related to iterative learning control.

Due to the vast amounts of data available in networked controlsystems, cooperative control, and internet systems, and real-timeresponse scenarios, new methods are needed for fast simplifieddecision based on emotional cues in uncertain non-stationary envi-ronments. Such methods are used in the human brain in the inter-play between structures such as the amygdala, orbitofrontalcortex, and hippocampus (Levine, 2012). Decision topologies basedon these neuro-cognitive interplays may have more levels andmore complex interconnections than the two-level structure ofthe actor-critic paradigm. The challenge will be developing suchnew decision and control schema and providing rigorous guaran-tees of stability, robustness, and performance for human engi-neered systems.

6. Brain machine interfaces

Throughout our evolution as a species, brains have used bodiesto interact with the external world. However, the combination ofmany emerging technologies is poised to change this status quoand create direct ways of linking the external world directly withthe brain, through what has been called brain machine interfaces(BMIs). BMI research may become both a launch pad for technol-ogy revolution in medical disciplines, as well as a paradigm shift


for research and development in the centuries to come because itrepresents a new level of integration between biologic and ‘‘manmade’’ systems. The general concept of a BMI is to either:

(1) Create artificial sensory systems (retina, cochlea) by deliver-ing the external world stimuli to the appropriate cortices.

(2) Allow the motor cortex to command directly externalrobotic devices.

(3) Repair brain subsystems, and regain cognitive function (e.g.memory).

(4) Link remotely brains in social networks.

In view of the amazing plasticity of the brain these externaldevices may, with appropriate sensory feedback, be assimilatedand become a part of the user’s cognitive space, on par with ourbody. Potentially, BMIs can improve human’s natural reaction timeand force through engineered devices that are much faster andmore powerful than biologic tissue. In particular BMIs may enablea higher communication bandwidth between the human brain andthe digital computer, which has conditioned the present metaphorand research direction for the use of computers in our society.

Work in BMIs is imminently multidisciplinary. In one side wefind neuroscientists and the disciplines of systems, computationaland cognitive neurosciences, while the engineering side is shoredup in advanced signal processing, control theory, ultra low powerVLSI, electrode design, optics, materials, computers, and communi-cations. None of the disciplines can alone solve the challenge ofinterfacing brains to machines, so this is truly a symbioticcollaboration.

A productive taxonomy is to divide BMIs in invasive and non-invasive, which means that the latter collects brain activity by fit-ting electrodes or other apparatus externally to the brain and body.Normally these systems are called brain computer interfaces (BCIs)and their current development has focused on communication andcontrol applications using menus in computer screens. The brain isa multi-scale spatio-temporal system, so when we decide on asensing methodology, we only measure some aspects of brainfunction. For instance when cup electrodes are placed over thescalp, the measurement is the electroencephalogram (EEG). In ananalogy, the EEG quantifies the single neuron activity as a micro-phone mounted on a helicopter hovering over a football stadiumwould quantify the voice of a spectator watching a game. Onlymajor neural events are quantified with the EEG, but surprisinglythe EEG is still today the best indicator of brain activity to diagnoseepilepsy, quantify depth of anesthesia and sleep staging. On theother extreme of the spatial resolution, the invasive microelectrodeinserted in the cortical tissue can quantify exactly the firing of aneuron in the vicinity of the electrode. However, it tells nothingabout the other 1011 neurons in the brain. Normally, electrodearrays with 100 microelectrodes or more are inserted in specificbrain areas to provide a grossly sub- sampled quantification ofthe neural system under analysis. These are the extremes of avail-able sensing methods in BMIs.

6.1. Example of BMIs

BMIs can be divided in two basic types depending upon theapplication: sensory or motor. Sensory BMIs stimulate sensory cor-tex areas with artificially generated signals that translate physicalquantities. The most common type of sensory BMI, with over200,000 implants, are cochlear implants that use tiny microphonesand signal processing to translate wave pressure near the ear intospike firings applied to the auditory nerve, allowing deft people tolisten (see, for instance, McDonnell, Burkitt, Grayden, Meffin, &Grant, 2010). The same basic concept is being developed for retinaprosthesis, which can deliver to the visual cortex the appropriate





stimulation that indicates the outline of objects (Abramian et al.,2014). Motor BMIs on the other hand seek to translate brain elec-trical activity into useful commands to external devices, mimickingthe natural motor control of muscles. There are two basic types ofmotor BMIs: command BCIs, and trajectory BMIs. Research in com-mand BCIs started in the late 80’s by collecting brain electricalactivity over the scalp (EEG) with a set of electrodes. The EEG tech-nology is non-invasive and requires only signal amplification andconditioning. By practicing, subjects learn to control their regionalbrain activity in a predetermined fashion that can be robustly clas-sified by a pattern recognition algorithm, and converted into a setof discrete commands. The paradigm for communication is mostlybased on selection of cursor actions (up/down, left/right) on a com-puter display. The computer presents to the users a set of possibil-ities and they choose one of them through the cursor movement,until an action is completed. BCIs require subject training throughbiofeedback and they display a low bandwidth for effective com-munication (15–25 bits per min) (Wolpaw & Wolpaw, 2012),which hinders the speed at which tasks can be accomplished. How-ever, due to their non-invasiveness, BCIs have already been testedwith success in paraplegics and locked-in patients. Several groupsall over the world have demonstrated working versions of BCIs(Wolpaw & Wolpaw, 2012), and a system’s software standard hasbeen proposed. There is a biannual BCI meeting with a BCI compe-tition and a BCI journal is about to be launched to evaluate the pro-gress of algorithms in this area.

Trajectory BMIs use brain activity to control directly the path ofan external actuator in 3D space, mimicking the role of the motorsystem when it controls, for instance, the trajectory of a handmovement. Alternatively, the commands can be sent to peripheralnerves that have been rerouted to activate the required muscles.The technical problem here cannot be solved by classifying brainsignals as in BCIs, but requires the BMI to generate a trajectorydirectly from brain activity. It has been difficult so far to extractfrom the EEG the motor command signatures with sufficient spa-tio-temporal resolution. Therefore, invasive techniques utilizingdirectly neuronal firings or local field potentials have been utilized.In the last decade, the introduction of new methods for recordingand analyzing large-scale brain activity and emerging develop-ments in microchip design, signal processing algorithms, sensorsand robotics are coalescing into a new technology devoted to cre-ating BMIs for controls. Trajectory control BMIs offer the possibilityto overcome the bandwidth limitation of BCIs and have been dem-onstrated with very encouraging results. The first demonstrationinvolved rats that used their brain signals to press a lever andobtain a water reward (Chapin, Moxon, Markowitz, & Nicolelis,1999). Later, similar BMIs were demonstrated in monkeys in foodreaching tasks (Wessberg et al., 2000), and more recently inhumans to control cursor movement in a computer screen(Chadwick et al., 2011).

6.2. The future

These are exciting times in BMI research. Although the prob-lems faced are exceedingly difficult there is a sense of optimismand a road map of intermediate solutions that will some day deli-ver general purpose and efficient BMI. For this audience the chal-lenges in system science are many and I synthetize some below.

6.2.1. System identification and controlBMIs include two complex systems (the user and the computer

agent) in closed-loop feedback through a non-stationary environ-ment. This interaction is online and because the brain is plasticand the environment non-stationary, adaptive online system iden-tification and adaptive optimal control are needed for good perfor-mance, which includes guaranteed stability. These issues are only


now starting to be addressed, because the work to date hasaddressed proof of concept in subsystems. For instance, the tradi-tional way of designing a trajectory BMI is to identify the systemfrom the collected brain activity to the kinematics, which requiresoff line training, but system parameters cannot be easily updatedduring operation, which requires daily retraining (Sanchez &Principe, 2007). Work using reinforcement learning is emerging(DiGiovanna, Mahmoudi, Fortes, Principe, & Sanchez, 2009) butstill requires a lot of further work to make training fast, accurateand reliable.

6.2.2. Augmented optimal controlAnother potential application area of system science is how to

integrate optimally spiking models of neural tissue of particularbrain areas with the data collected on-line by electrodes of dis-abled users to help them compensate for their disability and solvetasks in the real world. A simplified solution that has been tested isto control an electrical stimulator applied to one part of the brainto generate a desired spike train in another part of the brain. Aninverse adaptive control framework was implemented (Li &Principe, 2013), but other approaches are possible and potentiallybetter for this application. One of the lingering difficulties in allthese applications is that neurons communicate through discreteevents in time called spikes, where the information is containedsolely in the time of the event. Therefore all the signal analysismethods described should accept point processes as their inputsand produce also spikes as outputs. The first steps to accomplishthis are underway (Park, Seth, Paiva, Li, & Principe, 2013).

7. Concluding remarks

This milestone paper addressed only four domains within CImethodologies: neural networks control, fuzzy control, reinforce-ment learning and brain machine interfaces. Additional topics,such as knowledge-based systems, evolutionary algorithms andswarm intelligence systems will be covered in subsequent mile-stone papers.

CI methodologies are currently applied to all areas of control. Itis our view that the Technical Committee on Computational Intel-ligence in Control should strengthen the collaboration among otherIFAC TCs and in particularly with the other TCs within CC3, TC 3.1and TC3.3. The latter can be achieved by promoting the triennialConference for Embedded Systems, Computer intelligence and Tel-ematics (CESCIT), while the former can be achieved by maintain-ing, or increasing the number of co-sponsored conferences withother TCs, by creating multi-TC working groups and by proposingjoint special sessions.

IFAC and in particularly TC 3.2 has everything to gain in pro-moting collaboration with external Institutions, such as IEEE andIFSA. This has happened in previous occurrences of our TC 3.2ICONS Conference (co-sponsored by these Institutions), and shouldbe also maintained in the future.

References

Abramian, M., Lovell, N. H., Habib, A., Morley, J. W., Suaning, G. J., & Dokos, S. (2014).Quasi-monopolar electrical stimulation of the retina: A computationalmodelling study. Journal of Neural Engineering, 11, 025002.

Abu-Khalaf, M., Lewis, F. L., & Huang, J. (2006). Policy iterations on the Hamilton–Jacobi–Isaacs equation for state feedback control with input saturation. IEEETransactions on Automatic Control, 51, 1989–1995.

Adamy, J., & Schwung, A. (2010). Qualitative modeling of dynamical systemsemploying continuous-time recurrent fuzzy systems. Fuzzy Sets and Systems,161, 3026–3043.

Al-Tamimi, A., Lewis, F. L., & Abu-Khalaf, M. (2008). Discrete-time nonlinear HJBsolution using approximate dynamic programming: Convergence proof. IEEETransactions on Systems, Man, Cybernetics, Part B, 38, 943–949.


http://refhub.elsevier.com/S1367-5788(14)00040-6/h0005















ARTEMIS Industry Association: ARTEMIS Strategic Research Agenda (2011). Firstedition, April 2011 <http://www.artemis-ia.eu/publications>, DownloadedSeptember 2013.

ARTEMIS ITEA Cooperation Committee (2012). High-level vision ITEA – ARTEMIS2030, version 2012, 19 September 2012 <http://www.artemis-ia.eu/sra_main>,Downloaded September 2013.

Astrom, K. J., & Wittenmark, B. (1995). Adaptive control. Addison-Wesley.Atacak, I., & Bay, O. F. (2012). A type-2 fuzzy logic controller design for buck and

boost DC–DC converters. Journal of Intelligent Manufacturing, 23, 1023–1034.Azar, A.T., El-Said, S.A., Balas, V.E., & Olariu, T. (2013). Linguistic hedges fuzzy

feature selection for differential diagnosis of erythemato-squamous diseases. InV.E. Balas, J. Fodor, A.R. VarkonyiKoczy, J. Dombi, & L.C. Jain (Eds.), Softcomputing applications (pp. 487–500).

Balakrishnan, S. N., Ding, J., & Lewis, F. L. (2008). Issues on stability of ADP feedbackcontrollers for dynamical systems. IEEE Transactions on Systems, Man,Cybernetics, Part B, 38(4), 913–917.

Barto, A. G. (1984). Neuron-like adaptive elements that can solve difficult learningcontrol problems. Behavioural Processes, 9, 89.

Bellman, R. E. (1957). Dynamic programming. Princeton, NJ: Princeton UniversityPress.

Benoit, E., & Foulloy, L. (2013). The role of fuzzy scales in measurement theory.Measurement, 46, 2921–2926.

Bertsekas, D. P., & Tsitsiklis, J. N. (1996). Neuro-dynamic programming. MA: AthenaScientific.

Boukef, H., Benrejeb, M., & Borne, P. (2008). Flexible job-shop scheduling problemsresolution inspired from particle swarm optimization. Studies in Informatics andControl, 17, 241–252.

Boulkroune, A., Tadjine, M., M’Saad, M., & Farza, M. (2008). How to design a fuzzyadaptive controller based on observers for uncertain affine nonlinear systems.Fuzzy Sets and Systems, 159(8), 926–948.

Busoniu, L., Babuska, R., De Schutter, B., & Ernst, D. (2009). Reinforcement learningand dynamic programming using function approximators. Boca Raton: CRC Press.

Caloba, L. O. M., Caloba, L. P., & Contador, C. R. (2001). Delta-neutral volatilitytrading using neural networks. Engineering Intelligent Systems for ElectricalEngineering and Communications, 9, 243–249.

Cerman, O., & Husek, P. (2012). Adaptive fuzzy sliding mode control for electro-hydraulic servo mechanism. Expert Systems with Applications, 39, 10269–10277.

Chadli, M., & Guerra, T. M. (2012). LMI solution for robust static output feedbackcontrol of discrete Takagi–Sugeno fuzzy models. IEEE Transactions on FuzzySystems, 20(6), 1160–1165.

Chadwick, E. K., Blana, D., Simeral, J. D., Lambrecht, J., Kim, S. P., Cornwell, A. S., et al.(2011). Continuous neuronal ensemble control of simulated arm reaching by ahuman with tetraplegia. Journal of Neural Engineering, 8(3).

Chapin, J. K., Moxon, K. A., Markowitz, R. S., & Nicolelis, M. A. (1999). Real-timecontrol of a robot arm using simultaneously recorded neurons in the motorcortex. Nature Neuroscience, 2(7), 664–670.

Chen, F., Sekiyama, K., & Fukuda, T. (2012). A genetic algorithm for subtaskallocation within human and robot coordinated assembly.

Chen, B., Harrison, R. F., Papadatos, G., Willett, P., Wood, D. J., Lewell, X. Q., et al.(2007). Evaluation of machine-learning methods for ligand-based virtualscreening. Journal of Computer-Aided Molecular Design, 21, 53–62.

Chen, S. L., Luo, G. A., & Lin, T. L. (2013). Efficient fuzzy-controlled and hybridentropy coding strategy lossless ECG encoder VLSI design for wireless bodysensor networks. Electronics Letters, 49, 1058–1059.

Chwa, D., Choi, J. Y., & Anavatti, S. G. (2006). Observer-based adaptive guidance lawconsidering target uncertainties and control loop dynamics. IEEE Transactions onControl Systems Technology, 14, 112–123.

Cismondi, F., Horn, A. L., Fialho, A. S., Vieira, S. M., Reti, S. R., Sousa, J. M. C., &Finkelstein, S. (2012). Multi-stage modeling using fuzzy multi-criteria featureselection to improve survival prediction of ICU septic shock patients. ExpertSystems with Applications, 39, 12332–12339.

Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20,273–297.

Dereli, T., Baykasoglu, A., Altun, K., Durmusoglu, A., & Turksen, I. B. (2011). Industrialapplications of type-2 fuzzy sets and systems: a concise review. Computers inIndustry, 62, 125–137.

DiGiovanna, J., Mahmoudi, B., Fortes, J., Principe, J., & Sanchez, J. (2009). Co-adaptivebrain machine interface via reinforcement learning. IEEE Transactions onBiomedical Engineering, 56(1), 54–64.

Ding, B. C., Sun, H. X., & Yang, P. (2006). Further studies on LMI-based relaxedstabilization conditions for nonlinear systems in Takagi–Sugeno’s form.Automatica, 42(3), 503–508.

Doya, K., Kimura, H., & Kawato, M. (2001). Neural mechanisms of learning andcontrol. IEEE Control Systems Magazine, 21, 42–54.

Dumitrache, I., & Caramihai, S. I. (2008). Towards a new generation of intelligentmanufacturing systems.

Essounbouli, N., Manamanni, N., Hamzaoui, A., & Zaytoon, J. (2006). Synthesis ofswitching controllers: A fuzzy supervisor approach. Nonlinear Analysis-TheoryMethods & Applications, 65, 1689–1704.

Feng, G. (2006). A survey on analysis and design of model-based fuzzy controlsystems. IEEE Transactions on Fuzzy Systems, 14(5), 676–697.

Ferreira, P.M., & Ruano, A.E. (2008). Discrete model based greenhouseenvironmental control using the branch & bound algorithm. In 17th IFACworld congress. Seoul, Korea.


Ferreira, P. M., Gomes, J. M., Martins, I. A. C., & Ruano, A. E. (2012). A neural networkbased intelligent predictive sensor for cloudiness, solar radiation, and airtemperature. Sensors, 12, 15750–15777.

Ferreira, P. M., Ruano, A. E., Silva, S., & Conceicao, E. Z. E. (2012). Neural networksbased predictive control for thermal comfort and energy savings in publicbuildings. Energy Building, 55, 238–251.

Fischer, C., Tibken, B., & Fischer, T. (2010). Left behind occupant recognition in parkedcars based on acceleration and pressure information using k-nearest-neighborclassification. In 2010 IEEE intelligent vehicles symposium (pp. 1242–1247).

Garcia, R. F., Rolle, J. L. C., Gomez, M. R., & Catoira, A. D. (2013). Expert conditionmonitoring on hydrostatic self-levitating bearings. Expert Systems withApplications, 40, 2975–2984.

Ge, S. S., Hang, C. C., & Zhang, T. (2001). Stable adaptive neural network control.Norwell, USA: Kluwer Academic.

Ge, S. S., Lee, T. H., & Harris, C. J. (1998). Adaptive neural network control of roboticmanipulators. London: World Scientific.

Ge, S. S., Yang, C., & Lee, T. H. (2008). Adaptive predictive control using neuralnetwork for a class of pure-feedback systems in discrete-time. IEEE Transactionson Neural Networks, 19(9), 1599–1614.

Goodband, J. H., Haas, O. C. L., & Mills, J. A. (2008). A comparison of neural networkapproaches for on-line prediction in IGRT. Medical Physics, 35, 1113–1122.

Gormley, P., Li, K., & Irwin, G. W. (2007). Modelling molecular interaction pathwaysusing a two-stage identification algorithm. Systems and Synthetic Biology, 1,145–160.

Govindhasamy, J. J., McLoone, S. F., & Irwin, G. W. (2005). Second-order training ofadaptive critics for online process control. IEEE Transactions on Systems Man andCybernetics Part B-Cybernetics, 35, 381–385.

Guelton, K., Delprat, S., & Guerra, T. M. (2008). An alternative to inverse dynamicsjoint torques estimation in human stance based on a Takagi–Sugeno unknown-inputs observer in the descriptor form. Control Engineering Practice, 16,1414–1426.

Guillaume, R., Grabot, B., & Thierry, C. (2013). Management of the risk of backordersin a MTO-ATO/MTS context under imperfect requirements. AppliedMathematical Modelling, 37, 8060–8078.

Halang, W. A., Sanz, R., Babuska, R., & Roth, H. (2006). Information andcommunication technology embraces control: Status report prepared by theIFAC coordinating committee on computers, cognition and communication.Annual Reviews in Control, 30, 31–40.

Heikkinen, M., Latvala, T., Juuso, E., & Hiltunen, Y. (2008). SOM-based modelling foran activated sludge treatment process.

Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of datawith neural networks. Science, 313(5786), 504–507.

Huang, G.-B., Zhu, Q.-Y., & Siew, C.-K. (2006). Extreme learning machine: Theory andapplications. Neurocomputing, 70(1–3), 489–501.

Ioannou, P., & Fidan, B. (2006). Adaptive control tutorial. Philadelphia: SIAM Press.Jiang, Y., & Jiang, Z. P. (2012). Computational adaptive optimal control for

continuous-time linear systems with completely unknown dynamics.Automatica, 48, 2699–2704.

Johansson, M., Rantzer, A., & Arzen, K. E. (1999). Piecewise quadratic stability offuzzy systems. IEEE Transactions on Fuzzy Systems, 7(6), 713–722.

Karer, G., Music, G., Skrjanc, I., & Zupancic, B. (2007). Hybrid fuzzy model-basedpredictive control of temperature in a batch reactor. Computers & ChemicalEngineering, 31, 1552–1564.

Koprinkova-Hristova, P., & Palm, G. (2010). Adaptive critic design with ESN critic forbioprocess optimization. In K. Diamantaras, W. Duch, & L. S. Iliadis (Eds.),Artificial Neural Networks-ICANN 2010 (pp. 438–447). Pt Ii.

Korosi, L., & Kozak, S. (2013). An effective application of soft computing methods forhydraulic process control. In V. Snasel, A. Abraham, & E.S. Corchado (Eds.), SoftComputing Models in Industrial and Environmental Applications (pp. 379–388).

Kruszewski, A., Sala, A., Guerra, T. M., & Arino, C. (2009). A triangulation approach toasymptotically exact conditions for fuzzy summations. IEEE Transactions onFuzzy Systems, 17(5), 985–994.

Kruszewski, A., Wang, R., & Guerra, T. M. (2008). Nonquadratic stabilizationconditions for a class of uncertain nonlinear discrete time TS fuzzy models: Anew approach. IEEE Transactions on Automatic Control, 53(2), 606–611.

Lee, J. Y., Park, J. B., & Choi, Y. H. (2012). Integral Q-learning and explorized policyiteration for adaptive optimal control of continuous-time linear systems.Automatica, 48, 2850–2859.

Levine, D. S. (2012). Neural dynamics of affect, gist, probability, and choice.Cognitive Systems Research, 15–16, 57–72.

Lewis, F. L. (2009). Reinforcement learning and adaptive dynamic programming forfeedback control. Circuits and Systems Magazine, 9(3), 32–50.

Lewis, F. L., Lendaris, G., & Liu, D. (2008). Special issue on approximate dynamicprogramming and reinforcement learning for feedback control. IEEETransactions on Systems, Man, and Cybernetics, Part B, 38(4).

Lewis, F. L., Vrabie, D., & Syrmos, V. (2012). Optimal control (3rd ed.). New York: JohnWiley.

Li, L., & Principe, J. (2013). Adaptive inverse control of neural spatiotemporal spikepatterns with a reproducing kernel hilbert space (RKHS) framework. IEEETransactions on Neural Systems and Rehabiliation Engineering, 21(4), 532–543.

Liu, X., & Balakrishnan, S. N. (2000). Convergence analysis of adaptive critic basedoptimal control. In American control conference (pp. 1929–1933).

Liu, D. R., Pang, Z. Y., & Lloyd, S. R. (2008). A neural network method for detection ofobstructive sleep apnea and narcolepsy based on pupil size and EEG. IEEETransactions on Neural Networks, 19, 308–318.


http://www.artemis-ia.eu/sra_main


















































































































































Liu, Y.-J., Tong, S.-C., & Li, T.-S. (2011). Observer-based adaptive fuzzy trackingcontrol for a class of uncertain nonlinear MIMO systems. Fuzzy Sets and Systems,164(1), 25–44.

Malatji, E. M., Zhang, J. F., & Xia, X. H. (2013). A multiple objective optimisationmodel for building energy efficiency investment decision. Energy and Buildings,61, 81–87.

Marcos, M. D., Machado, J. A. T., & Azevedo-Perdicoulis, T. P. (2009). Trajectoryplanning of redundant manipulators using genetic algorithms. Communicationsin Nonlinear Science and Numerical Simulation, 14, 2858–2869.

McCulloch, W. S., & Pitts, W. (1943). A logical calculus of the ideas immanent innervous activity. The Bulletin of Mathematical Biophysics, 5, 115–133.

McDonnell, M. D., Burkitt, A. N., Grayden, D. B., Meffin, H., & Grant, A. J. (2010). Achannel model for inferring the optimal number of electrodes for futurecochlear implants. IEEE Transactions on Information Theory, 56, 928–940.

Mendel, J. M., & John, R. I. (2002). Type-2 fuzzy sets made simple. IEEE Transactionson Fuzzy Systems, 10(2), 117–127.

Mendel, J. M., & MacLaren, R. W. (1970). Reinforcement learning control and patternrecognition systems. In J. M. Mendel & K. S. Fu (Eds.), Adaptive, learning, andpattern recognition systems: theory and applications (pp. 287–318). New York:Academic Press.

Mozelli, L. A., Palhares, R. M., Souza, F. O., & Mendes, E. M. A. M. (2009). Reducingconservativeness in recent stability conditions of TS fuzzy systems. Automatica,45(6), 1580–1583.

Nagesh, S.B., Lendek, Z., Khalate, A.A., & Babuska, R. (2012). Adaptive fuzzy observerand robust controller for a 2-DOF robot arm. In 2012 IEEE internationalconference on fuzzy systems.

Nomm, S., Vain, J., Petlenkov, E., Miyawaki, F., & Yoshimitsu, K. (2009). Hybridapproach to detection of the surgeon’s hand motions during endoscope surgery.

Oh, S.-K., Jang, H.-J., & Pedrycz, W. (2011). A comparative experimental study oftype-1/type-2 fuzzy cascade controller based on genetic algorithms and particleswarm optimization. Expert Systems with Applications, 38, 11217–11229.

Omatu, S., Araki, H., Fujinaka, T., Yoshioka, M., & Nakazumi, H. (2012). Mixed odorclassification for QCM sensor data by neural networks. In S. Omatu, J.F.D.Santana, S.R. Gonzalez, J.M. Molina, A.M. Bernardos, & J.M.C. Rodriguez (Eds.),Distributed computing and artificial intelligence (pp. 1–8).

Paik, H., Lee, E., Park, I., Kim, J., & Lee, D. (2011). Prediction of cancer prognosis withthe genetic basis of transcriptional variations. Genomics, 97, 350–357.

Pan, J.-T., Guerra, T. M., Fei, S.-M., & Jaadari, A. (2012). Nonquadratic stabilization ofcontinuous T–S fuzzy models: LMI solution for a local approach. IEEETransactions on Fuzzy Systems, 20(3), 594–602.

Park, I., Seth, S., Paiva, A., Li, L., & Principe, J. (2013). Kernel methods on spike trainspace for neuroscience: a tutorial. IEEE Signal Processing Magazine, 30(4),149–160.

Petelin, D., Grancharova, A., & Kocijan, J. (2013). Evolving Gaussian process modelsfor prediction of ozone concentration in the air. Simulation Modelling Practiceand Theory, 33, 68–80.

Pétrissans, A., Krawczyk, S., Cattaneo, G., Feeney, N., Veronesi, L., & Meunier, C.(2012). Design of future embedded systems (SMART 2009/0063). Paris.

Powell, W. B. (2007). Approximate dynamic programming. Hoboken: Wiley.Precup, R.-E., & Hellendoorn, H. (2011). A survey on industrial applications of fuzzy

control. Computers in Industry, 62, 213–226.Prokhorov, D., & Wunsch, D. (1997). Adaptive critic designs. IEEE Transactions on

Neural Networks, 8(5), 997–1007.Pulkkinen, P., Laurikkala, M., Ropponen, A., & Koivisto, H. (2008). Quality

management in GPRS networks with fuzzy case-based reasoning. Knowledge-Based Systems, 21, 421–428.

Rhee, B. J., & Won, S. (2006). A new fuzzy Lyapunov function approach for a Takagi–Sugeno fuzzy control system design. Fuzzy Sets and Systems, 157(9), 1211–1228.

Riccardi, L., Naso, D., Janocha, H., & Turchiano, B. (2012). A precise positioningactuator based on feedback-controlled magnetic shape memory alloys.Mechatronics, 22, 568–576.

Rumelhart, D.E., Hinton, G.E., & Williams, R.J. (1986). Learning internalrepresentations by error propagation. In Parallel distributed processing (pp.318–362).

Sakaguchi, S., Horio, K., & Yamakawa, T. (2011). Multimodal sensory motorcoordination by learning in dynamical self-organizing relationship network.International Journal of Innovative Computing Information and Control, 7,6097–6114.

Sala, A., & Arino, C. (2007). Asymptotically necessary and sufficient conditions forstability and performance in fuzzy control: Applications of Polya’s theorem.Fuzzy Sets and Systems, 158, 2671–2686.

Sanchez, J., & Principe, J. (2007). Brain machine interface engineering. Morgan andClaypool.

Sanz, R., Hernandez, C., Hernando, A., Gomez, J., Bermejo, J. (2009). Grounding robotautonomy in emotion and self-awareness. In J.H. Kim, S.S. Ge, P. Vadakkepat, N.Jesse, A. AlManum, S. Puthusserypady, et al. (Eds.), Advances in robotics (pp. 23–43).

Scholkopf, B., Smola, A. J., Williamson, R. C., & Bartlett, P. L. (2000). New supportvector algorithms. Neural Computation, 12, 1207–1245.

Schuitema, E., Busoniu, L., Babuska, R., & Jonker, P. (2010). Control delay inreinforcement learning for real-time dynamic systems: A memorylessapproach. In IEEE 2010 international conference on intelligent robots andsystems (pp. 3226–3231).

Schultz, W. (2004). Neural coding of basic reward terms of animal learning theory,game theory, microeconomics and behavioural ecology. Current Opinion inNeurobiology, 14, 139–147.


Seiffert, J., Sanyal, S., & Wunsch, D. (2008). Hamilton–Jacobi–Bellman equations andapproximate dynamic programming on time scales. IEEE Transactions onSystems, Man and Cybernetics, Part B, 38(4), 918–923.

Shakev, N. G., Topalov, A. V., Kaynak, O., & Shiev, K. B. (2012). Comparative resultson stabilization of the quad-rotor rotorcraft using bounded feedbackcontrollers. Journal of Intelligent & Robotic Systems, 65, 389–408.

Si, J., Barto, A. G., Powell, W. B., & Wunsch, D. (2004). Handbook of learning andapproximate dynamic programming. USA: IEEE Press.

Statheros, T., Howells, G., & McDonald-Maier, K. (2008). Autonomous ship collisionavoidance navigation concepts, technologies and techniques. Journal ofNavigation, 61, 129–142.

Sutton, R. S., & Barto, A. G. (1998). Reinforcement learning – An introduction.Cambridge, Massachusetts: MIT Press.

Takagi, T., & Sugeno, M. (1985). Fuzzy identification of systems and its applicationsto modeling and control. IEEE Transactions on Systems Man and Cybernetics,15(1), 116–132.

Tanaka, K., & Wang, H. O. (2001). Fuzzy control systems design and analysis. A linearmatrix inequality approach. New York: John Wiley and Sons.

Tavakolpour, A. R., Darus, I. Z. M., Tokhi, O., & Mailah, M. (2010). Genetic algorithm-based identification of transfer function parameters for a rectangular flexibleplate system. Engineering Applications of Artificial Intelligence, 23, 1388–1397.

Teixeira, C. A., Ruano, M. G., Ruano, A. E., & Pereira, W. C. A. (2008). A soft-computingmethodology for noninvasive time-spatial temperature estimation. IEEETransactions on Biomedical Engineering, 55, 572–580.

Tong, S., & Li, Y. (2009). Observer-based fuzzy adaptive control for strict-feedbacknonlinear systems. Fuzzy Sets and Systems, 160(12), 1749–1764.

van de Ven, P. W. J., Johansen, T. A., Sorensen, A. J., Flanagan, C., & Toal, D. (2007).Neural network augmented identification of underwater vehicle models.Control Engineering Practice, 15, 715–725.

Verbruggen, H., Park, J., Halang, W.A., Irwin, G.W., & Zalewski, J. (2002). Milestonereport coordinating committee on computer control. In IFAC world congress (Vol.15, pp. 217–225). Bacelona, Spain.

Vousdoukas, M. I., Ferreira, P. M., Almeida, L. P., Dodet, G., Psaros, F., Andriolo, U.,et al. (2011). Performance of intertidal topography video monitoring of a meso-tidal reflective beach in South Portugal. Ocean Dynamics, 61, 1521–1540.

Vrabie, D., & Lewis, F. (2009). Neural network approach to continuous-time directadaptive optimal control for partially unknown nonlinear systems. NeuralNetworks, 22, 237–246.

Vrabie, D., Pastravanu, O., Abu-Khalaf, M., & Lewis, F. L. (2009). Adaptive optimalcontrol for continuous-time linear systems based on policy iteration.Automatica, 45, 477–484.

Wang, L. X. (1993). Stable adaptive fuzzy control of nonlinear systems. IEEETransactions on Fuzzy Systems, 1(2), 146–155.

Wang, FY, Zhang, H, & Liu, D (2009). Adaptive dynamic programming: Anintroduction. IEEE Computational Intelligence Magazine, 39–47.

Werbos, P.J. (1989). Neural networks for control and system identification. In IEEEconference on decision and control (pp. 260–265). Tampa, Florida.

Werbos, P. J. (2009). Intelligence in the brain: A theory of how it works and how tobuild it. Neural Networks, 22(3), 200–212.

Werbos, P. J. (1991). A menu of designs for reinforcement learning over time. In W.T. Miller, R. S. Sutton, & P. J. Werbos (Eds.), Neural networks for control.Cambridge: MIT Press.

Werbos, P. J. (1992). Approximate dynamic programming for real-time control andneural modeling. In D. A. White & D. A. Sofge (Eds.), Handbook of intelligentcontrol. New York: Van Nostrand Reinhold.

Wessberg, J., Stambaugh, C. R., Kralik, J. D., Beck, P. D., Laubach, M., Chapin, J. K., et al.(2000). Real-time prediction of hand trajectory by ensembles of corticalneurons in primates. Nature Neuroscience, 408(6810), 361–365.

Wojtowicz, H., & Wajs, W. (2012). Medical decision support system for assessmentof dermatoglyphic indices and diagnosis of Down’s syndrome. In J. Watada, T.Watanabe, G. Phillips Wren, R.J. Howlett, & L.C. Jain (Eds.), Intelligent decisiontechnologies (pp. 69–78).

Wolpaw, J., & Wolpaw, E. (2012). Brain-computer interfaces: Principles and Practice.Oxford University: Press.

Wong, K. W., Tikk, D., Gedeon, T. D., & Koczy, L. T. (2005). Fuzzy rule interpolationfor multidimensional input spaces with applications: A case study. IEEETransactions on Fuzzy Systems, 13, 809–819.

Yang, J., Chen, W. H., & Li, S. (2012). Robust autopilot design of uncertain bank-to-turn missiles using state-space disturbance observers. Proceedings of theInstitution of Mechanical Engineers Part G-Journal of Aerospace Engineering, 226,97–107.

Zadeh, L. A. (2008). Is there a need for fuzzy logic? Information Sciences, 178,2751–2779.

Zribi, N., Kacem, I., El Kamel, A., & Borne, P. (2007). Assignment and scheduling inflexible job-shops by hierarchical optimization. IEEE Transactions on SystemsMan and Cybernetics Part C-Applications and Reviews, 37, 652–661.

António E. Ruano was born in Espinho, Portugal, in 1959. He holds a first degree inElectronic Eng. and Telecommunications, from the University of Aveiro, Portugal,in1982, a MSc degree in Electrotechnical Eng., from the University of Coimbra,Portugal, in 1989, a Ph.D. in Electronic Eng., from the University of Wales, UK, in1992 and an Habilitation in Electronic Eng. and Computing, from the University ofAlgarve, in 2004. He was a Lecturer in the University of Aveiro, from 1982 to 1992,and from 1992 an Auxiliar and Associate Professor at the Faculty of Sciences &Technology of the University of Algarve, where he is currently Associate Professorwith Habilitation and Pro-Rector of the University. His research activities are cen-








































































































































tered in Intelligent Control. He is Chairman of the IFAC Technical Committee onComputational Intelligence in Control since 2008. He has served in the EditorialBoard of Automatica and currently of International Journal of Systems Science andWorld Scientific, and chaired several IFAC and IEEE sponsored conferences. He hascoordinated more than 25 national and international projects, has supervised morethan 30 M.Sc., Ph.D. and Pos-Doc students, and has published more than 220 arti-cles, books and book chapters.

Shuzhi Sam Ge received the B.Sc. degree from Beijing University of Aeronautics andAstronautics, Beijing, China, in 1986, and the Ph.D. degree from the Imperial Collegeof Science, Technology and Medicine, University of London, London, U.K., in 1993.He is the Founding Director of the Robotics Institute and the Institute of IntelligentSystems and Information Technology, University of Electronic Science and Tech-nology of China, Chengdu, China. He is the Founding Director of the Social RoboticsLaboratory, Interactive Digital Media Institute, National University of Singapore. Heis a Professor in the Department of Electrical and Computer Engineering, NationalUniversity of Singapore, Singapore. He has authored or coauthored six books andmore than 300 international journal and conference papers. His current researchinterests include social robotics, multimedia fusion, medical robots, and intelligentsystems. Dr. Ge is the Editor-in-Chief of the International Journal of Social Robotics.He has served/been serving as an Associate Editor for a number of flagship journalsincluding the IEEE Transactions on Automatic Control, IEEE Transactions on ControlSystems Technology, IEEE Transactions on Neural Networks, and Automatica. Healso serves as an Editor of the Taylor & Francis Automation and Control EngineeringSeries. He also served as the Vice President of Technical Activities, 2009–2010, andthe Vice President for Membership Activities, 2011–2012, IEEE Control SystemsSociety.

Thierry Marie Guerra was born in Mulhouse, France in 1963. He is currently fullprofessor at the University of Valenciennes et du Hainaut-Cambrésis (UVHC),France. He received his Ph.D. degree in automatic control from the UVHC in 1991and the HDR in 1999. He is head of the Laboratory of Industrial and HumanAutomation, Mechanics and Computer Science (LAMIH CNRS UMR 8201; 100researchers and staff, 80 Ph.D. students and post-docs http://www.univ-valenci-ennes.fr/LAMIH/). He is vice-chair of the Technical Committee 3.2 ‘‘ComputationalIntelligence in Control’’ for IFAC (International Federation of Automatic Control),member of the IFAC TC 7.1 ‘‘Automotive Control’’, Area Editor of the internationaljournals Fuzzy Sets & Systems and IEEE Transactions on Vehicular Technology. Hismajor research fields and topics of interest are, wine, hard rock, chess, nonlinearcontrol, LPV, quasi-LPV (Takagi–Sugeno) models control and observation, LMIconstraints, Non quadratic Lyapunov functions. Applications to powertrain systems(IC engine, hybrid vehicles) and to disabled people.

Frank L. Lewis, Member, National Academy of Inventors. Fellow IEEE, Fellow IFAC,Fellow U.K. Institute of Measurement & Control, PE Texas, U.K. Chartered Engineer.UTA Distinguished Scholar Professor, UTA Distinguished Teaching Professor, andMoncrief-O’Donnell Chair at the University of Texas at Arlington Research Institute.Qian Ren Thousand Talents Professor, Northeastern University, Shenyang, China. Heobtained the Bachelor’s Degree in Physics/EE and the MSEE at Rice University, theMS in Aeronautical Engineering from Univ. W. Florida, and the Ph.D. at Ga. Tech. Heworks in feedback control, intelligent systems, cooperative control systems, and


nonlinear systems. He is author of 6 US patents, numerous journal special issues,journal papers, and 14 books, including Optimal Control, Aircraft Control, OptimalEstimation, and Robot Manipulator Control which are used as university textbooksworldwide. He received the Fulbright Research Award, NSF Research InitiationGrant, ASEE Terman Award, Int. Neural Network Soc. Gabor Award, UK Inst Mea-surement & Control Honeywell Field Engineering Medal, and IEEE ComputationalIntelligence Society Neural Networks Pioneer Award. Received Outstanding ServiceAward from Dallas IEEE Section, selected as Engineer of the year by Ft. Worth IEEESection. Was listed in Ft. Worth Business Press Top 200 Leaders in Manufacturing.Texas Regents Outstanding Teaching Award 2013. He is Distinguished VisitingProfessor at Nanjing University of Science & Technology and Project 111 Professorat Northeastern University in Shenyang, China. Founding Member of the Board ofGovernors of the Mediterranean Control Association.

Jose C. Principe is Distinguished Professor of Electrical and Biomedical Engineeringat the University of Florida since 2002. He is BellSouth Professor and FoundingDirector of the University of Florida Computational Neuro-Engineering Laboratory(CNEL). He joined the University of Florida in 1987, after an eight year appointmentas Professor at the University of Aveiro, in Portugal. Dr. Principe holds degrees inelectrical engineering from the University of Porto, Portugal, University of Florida,USA (Master and Ph.D.), and Honoris Causa degrees from the Universita Mediter-ranea in Reggio Calabria, Italy, Universidade do Maranhao, Brazil and Aalto Uni-versity, Finland. Dr. Principe is a Fellow of the IEEE (2000), AIMBE (2006), IAMBE(2012) and recipient of the INNS Gabor Award, the IEEE Engineering in Medicineand Biology Society Career Achievement Award, and the IEEE ComputationalIntelligence Society Neural Network Pioneer Award. He served as President of theInternational Neural Network Society in 2004, as Editor in Chief of the IEEETransactions of Biomedical Engineering from 2001 to 2007, and as a member of theAdvisory Science Board of the FDA from 2001 to 2004. He is currently the FoundingEditor in Chief of the IEEE Reviews in Biomedical Engineering. He has been heavilyinvolved in conference organization and several IEEE society administrative com-mittees. Dr. Principe chaired 78 Ph.D. and 61 Master student committees, and he isauthor of more than 600 refereed publications (5 books, 7 edited books, 19 bookchapters, 201 journal papers and 427 conference proceedings). He holds 22 patentsand has submitted seven.

Matjaz Colnaric received a degree in Electrical Engineering in 1980, a Mastersdegree in computer science in 1983 and a Doctorate in computer science in 1992.These were all from University of Maribor, where he has also been employed since1979 and was nominated full professor in 2002. In 1989 he founded the Real-TimeSystems Laboratory which he still chairs. He is a visiting professor at the Universityof Osijek, Croatia. He has authored a number of papers, book chapters, books andconference contributions. He serves on the editorial boards of the Journal of Com-puting and Information Technology, Zagreb and the International Journal of Elec-trical and Computer Engineering Systems, Osijek, Croatia. He has served as an EUexpert for evaluating of project proposals and projects within the area of embeddedsystems. He has been active in IFAC since 1996 and joined the Technical Board in2008 as a chairperson of the Coordination Committee on Computers, Cognition andCommunication, where he served until 2014. His research interests comprise themajor areas of embedded computing with an emphasis on hard real-time andsafety-related systems.


http://www.univ-valenciennes.fr/LAMIH/

http://www.univ-valenciennes.fr/LAMIH/



computational intelligence in control

Documents