review article bioinspired intelligent algorithm and...

Review ArticleBioinspired Intelligent Algorithm and Its Applications forMobile Robot Control A Survey

Jianjun Ni12 Liuying Wu1 Xinnan Fan12 and Simon X Yang3

1College of IOT Engineering Hohai University Changzhou 213022 China2Changzhou Key Laboratory of Special Robot and Intelligent Technology Hohai University Changzhou 213022 China3Advanced Robotics and Intelligent Systems (ARIS) Laboratory School of Engineering University of Guelph GuelphON Canada N1G 2W1

Correspondence should be addressed to Jianjun Ni njjhhucgmailcom

Received 1 September 2015 Accepted 8 November 2015

Academic Editor Jose Alfredo Hernandez

Copyright copy 2016 Jianjun Ni et al This is an open access article distributed under the Creative Commons Attribution Licensewhich permits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

Bioinspired intelligent algorithm (BIA) is a kind of intelligent computing method which is with a more lifelike biological workingmechanism than other types BIAs havemade significant progress in both understanding of the neuroscience and biological systemsand applying to various fields Mobile robot control is one of the main application fields of BIAs which has attractedmore andmoreattention because mobile robots can be used widely and general artificial intelligent algorithms meet a development bottleneckin this field such as complex computing and the dependence on high-precision sensors This paper presents a survey of recentresearch in BIAs which focuses on the research in the realization of various BIAs based on different working mechanisms andthe applications for mobile robot control to help in understanding BIAs comprehensively and clearly The survey has four primaryparts a classification of BIAs from the biomimetic mechanism a summary of several typical BIAs from different levels an overviewof current applications of BIAs in mobile robot control and a description of some possible future directions for research

1 Introduction

Recently a new type of intelligent computing methods hasbeen developed to overcome the limitations of traditionalartificial intelligent methods One of the most importantfeatures of these intelligent computing methods is that theirworking mechanisms are more lifelike to an individual ora group of organisms which can be understood very wellAnd these methods usually have higher efficiency thanthe traditional artificial intelligent methods The intelligentcomputing methods of this type are defined as bioinspiredintelligent algorithms (BIAs) to distinguish them from thetraditional artificial intelligent methods BIAs have made sig-nificant progress in both understanding of the neuroscienceand biological systems and applying to various fields such asmobile robot control

The purpose of this paper is to provide a survey ofthe recent research in the bioinspired intelligent algorithmand its applications for mobile robot control We focus onthe research in the realization of various BIAs based on

different mechanisms To help in understanding BIAs wefocus exclusively on the applications for mobile robot controlthat is a main application field of BIAs Several other surveysof the literature in the field of BIAs and mobile robot controlare available that complement this paper (see eg [1 2])

The main contributions of this paper are summarizedas follows (1) A detailed analysis on the features of BIAsis given out And a new classification is proposed from thebiomimetic mechanism of BIAs (2) A survey of recent BIAsis provided and some BIAs are chosen elaborately to focuson the biomimetic mechanism and realization processing ofBIAs (3) An overview of the applications in mobile robotcontrol by BIAs is given out to further illustrate BIAsFurthermore some possible future directions of the researchon BIAs are discussed

This paper is organized as follows In Section 2 weprovide a general introduction of BIAs and give out a clas-sification of these BIAs Section 3 analyzes and summarizessome typical BIAs The main applications of BIAs in mobilerobot control are introduced in Section 4 Section 5 discusses

Hindawi Publishing CorporationComputational Intelligence and NeuroscienceVolume 2016 Article ID 3810903 16 pageshttpdxdoiorg10115520163810903

2 Computational Intelligence and Neuroscience

the future research directions for the theory of BIAs and theirapplications in mobile robot control Finally conclusions aregiven out in Section 6

2 Classification of BIAs

There are two main aspects for human beings to learn fromnature The first one is to make inventions by imitating thestructures of organisms For example human beings createdthemanned glider flight based on the flying principle of birdsand invented the echolocation sonar system inspired from theecholocation function of bats Another one is to design somealgorithms (technologies) inspired from various principles ofnature For example based on the process of ant foraging AntColony Algorithm was developed and based on the processof natural evolution such as inheritance mutation selectionand crossover Genetic Algorithm was proposed With thedevelopment of science and technology human beings willface more and more challenges Some of the traditionalartificial intelligent algorithms are too simple in structures orfunctions to satisfy the requirements of developments Fromthe late 20th century more and more novel artificial intel-ligent algorithms have been proposed such as bioinspiredneural network Artificial Immune Algorithm and Mem-brane ComputingAlgorithmThese novel artificial intelligentmethods as well as some other intelligent algorithms (suchas Ant Colony Algorithm and Genetic Algorithmmentionedabove) have a common feature Namely they have morelifelike biological working mechanisms than other types Theartificial intelligent methods of this type are the bioinspiredintelligent algorithms (BIAs) discussed in this paper

Now BIAs are still in the stage of development so thereis no strict definition and uniform classification Binithaand Sathya [1] described the origin and advantages of thebioinspired computing algorithms and pointed out that BIAswere heuristic methods that imitated the strategy of naturewhich was a simple and nonrepresentational definition ofBIAs Bongard [3] introduced the development process ofthe bioinspired computing algorithm and analyzed the rela-tionship between the BIAs and the traditional intelligentcomputing methods Based on our expertise the bioinspiredintelligent computing algorithm can be defined as followsit is a type of intelligent computing methods with a quitelifelike biologicalworkingmechanism to imitate the functionand structure of the organism the individual and swarmbehaviors and the evolution process of life and societyHowever it is not an easy job to exclude those methodsfrom BIAs which are not strictly bioinspired In this paperwe do not intend to distinguish different types of intelligentcomputing algorithms but to analyze the main features ofBIAs classify these algorithms from the simulated biologicalworkingmechanisms and survey different categories focusedon the realization processing as well as the applications formobile robot control

Based on our research work and the overview of lots ofliterature the remarkable features of BIAs discussed in thispaper are summarized as follows [1 4ndash7]

(1) Bioinspired feature the working mechanism is veryclose to the biological or ecological mechanism of

natural organismsThebioinspired algorithms imitatethe biological nature as much as possible to deal withthe real-world problems

(2) Simplicity and emergence the strategy and computa-tion are often very simple but their resultant effectsare very amazing which reflects the principle ofemergence

(3) Robustness these algorithms have strong robustnessagainst the change of environments parameters andtasks namely these algorithms have good applicabil-ity and flexibility

(4) Self-organization these algorithms can improve theadaptability by self-learning or self-organization andrealize the evolution successfully

(5) Other features these algorithms have some othergood features such as parallelism in essence and non-determinism

As introduced above BIAs have many excellent charac-teristics that can meet the needs of researchers To introduceBIAs clearly and understand them easily BIAs should beclassified From different view angle various classificationscan be obtained for example from the whole effects we cansee all BIAs as a type of evolutional optimization algorithmsIn this paper we will classify BIAs from their sources of thebiomimetic mechanism then BIAs can be divided into threecategories (1) inspired from organism behaviors (2) inspiredfrom organism structure (3) inspired from evolution Theclassification in this paper is single that is to say onebioinspired algorithm will not be classified into two ormore different categories The classification map is shown inFigure 1 and each category will be introduced in the nextsection clearly by some typical algorithms

3 Theories Overview

There are some similarities in the property and application ofeach category with the same biomimetic mechanism In thispaper we will focus on the realization processing of severalnovel algorithms proposed in the last two decades as well assome distinctive or representative methods as far as we know

31 Inspired from Organism Behaviors Every living creaturecan survive and protect itself in its own way such as foragingof food defending against natural enemy and courtshipdisplaying Based on the characteristics of these importantbehaviors of organisms some BIAs have been developedIn these BIAs most of them are inspired from the foragingbehaviors of organisms such as Ant Colony AlgorithmBee Colony Algorithm Bacterial Foraging Algorithm FishSwarm Algorithm and Shuffled Frog Leaping AlgorithmThere are also some algorithmswhich are inspired from otherbehaviors of organisms For example Monkey ClimbingAlgorithm simulates the climbing behavior of monkeys andBacterial Chemotaxis Algorithm simulates the chemotaxisbehavior of bacteria There are several other surveys of theliterature on the behavior-based BIAs (see [8ndash10]) so we just

Computational Intelligence and Neuroscience 3

Bioinspired intelligentalgorithms

Inspired fromorganism structure

Inspired fromorganism behaviors

Inspired fromevolution

Based on the foraging behavior

Based on other behaviors

Ant ColonyAlgorithm

Bee ColonyAlgorithm

Bacterial ForagingAlgorithm

Shuffled FrogLeaping Algorithm

Monkey ClimbingAlgorithm

DNA

Membrane

Artificial Immune System

Bioinspiredneural network

Invasive WeedAlgorithm

Chemical Genetic Program

Selfish GeneAlgorithm

Coevolutionalgorithm

Culture Firefly

algorithm

Bacterial ChemotaxisAlgorithm

Based on biological evolution

Based on social evolution

Computing

Computing

Algorithm

Figure 1 The classification of BIAs from the biomimetic mechanism

select two algorithms to introduce the realization process indetail of the foraging behavior- (Bacterial Foraging Algo-rithm) and other behaviors- (Monkey Climbing Algorithm)based BIAs respectively

311 Bacterial Foraging Algorithm In 2002 Passino pro-posed a novel BIA inspired from the behavior of E colibacterium of foraging nutrition and food in peoplersquos intesti-nal canal which is called Bacterial Foraging Algorithm[11] There are some special behavior patterns in bacterialforaging such as swimming tumbling and chemotacticbehavior (see Figure 2) The basic steps of Bacterial ForagingAlgorithm are abstracted from these behaviors which areintroduced as follows [11ndash13]

(1) ChemotaxisThis is the behavior of bacteria gathering intothe nutrition area which includes tumbling and swimmingSwimming means moving to an arbitrary direction in a unitstep and tumbling decides the directions of the bacteriaBacterium will refresh its fitness value at each step oftumbling and will not stop tumbling till the fitness value is nolonger better or the number of steps is out of limitation Let120579119894(119895 119896 119897) represent the state of the 119894th bacterium (the location

of the 119894th bacterium) then the tumbling behavior can bedenoted by

120579119894(119895 + 1 119896 119897) = 120579

119894(119895 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (1)

where 119894 is the sequence number of the bacterium 119896 is thenumber of reproduction 119895 is the number of chemotaxis 119897 isthe number of dispersal119862(119894) denotes a basic chemotactic stepsize 120579119894(119895+1 119896 119897) represents the next state of the 119894th bacteriumafter tumbling Δ(119894) is a random vector whose elements arerandom numbers in [minus1 1] The swimming behavior can bedenoted by

120579119894(119895 + 1 119896 119897) = 120579

119894(119895 + 1 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (2)

(2) Reproduction This is the behavior of selective multipli-cation of the bacteria Namely the bacteria with good fitnessare selected to propagate whose offsprings will have the samepositions and steps But the bacteria with bad fitness will beweeded outThe elimination mechanism is based on the sumof the costs of the bacterium when it goes over its lifetimeThe health of the 119894th bacterium 119869119894health is calculated by

119869119894

health =

119873119888+1

sum

119895=1

119869 (119894 119895 119896 119897) (3)

where 119869(119894 119895 119896 119897) is the cost at the location of the 119894th bac-terium 119873

119888is the length of the lifetime of the 119894th bacterium

as measured by the number of chemotactic steps The highercost means the lower health so the bacteria with the highestvalues of 119869health will die at first


(a) (b) (c)

Figure 2 Swimming tumbling and chemotactic behavior of E coli [11] (a) the behavior of swimming (b) the behavior of tumbling (c) thebehavior of chemotaxis

(3) Dispersal This is the behavior of dispersing the bacteriato random positions at a certain probability This behavioraims at avoiding the local minimum of Bacterial ForagingAlgorithm and realizing global search

There are many advantages in Bacterial Foraging Algo-rithm such as parameter insensitivity strong robustness andeasy implementation And a lot of research results have beenobtained by this algorithm for example Sanyal et al [14]presented an adaptive Bacterial Foraging Algorithm for thesegmentation of gray images where the Bacterial ForagingAlgorithm is employed for maximization of fuzzy entropyto achieve the desired threshold based segmentation Liuet al [15] proposed a mobile robot path planning methodbased on Bacterial Foraging Algorithm where the BacterialForaging Algorithm is used to minimize the path length andthe number of turns without colliding with an obstacle

312 Monkey Climbing Algorithm In 2008 Zhao and Tang[16] presented Monkey Climbing Algorithm inspired by theactions ofmonkeyswhen they climb the trees to the best placefor some reasons to solve global optimization problemsThere are three main processes in this algorithm namelyclimb process watch-jump process and somersault process(1) Climb process this process represents the local search ofthe algorithm The monkeys climb from the initial positionsand find the respective local optimum positions (2) Watch-jump process this process represents the search process fromlocal optimum to global optimum The monkeys will lookaround when arriving at the local optimal positions If somebetter positions can be found themonkeys will jump to thesenew positions otherwise they will stay where they are (3)Somersault process this process represents the search processfor avoiding the local optimumThemonkeys will somersaultto new positions at a certain probability and begin a newsearch

The numerical computing processes ofMonkey ClimbingAlgorithm are summarized as follows [16 17]

Step 1 Initialize the size of monkeys 119872 and the initialpositions of all the monkeys For the 119894th monkey its positionis denoted by 119909

119894= (1199091198941 1199091198942 119909

119894119899) where 119899 is the dimension

of the solution

Step 2 Optimize the positions of the monkeys by climbingSet 119910 = (119910

1 1199102 119910

119899) and 119910

119894= 119909119894119895+ 119886 sdot sgn(1198911015840

119894119895(119909119894)) where

sgn(sdot) is a sign function 119886 gt 0 is the step length of the climbprocess and 1198911015840

119894119895(119909119894) is the pseudo gradient of the objective

function 119891(sdot) at the position 119909119894 which is calculated by

1198911015840

119894119895(119909119894) =

119891 (119909119894+ Δ119909119894) minus 119891 (119909

119894minus Δ119909119894)

2Δ119909119894119895

119895 = 1 2 119899

(4)

where Δ119909119894= (Δ119909

1198941 Δ1199091198942 Δ119909

119894119899) and Δ119909

119894119895= 119886 or minus119886

with the same probability If position 119910 is feasible update theposition of the monkey 119909

119894with 119910

Step 3 Search for better places within the visual range ofmonkeys and update their positionsThe detailed computingprocess is as follows generate a number 119910

119895sube (119909119894119895minus 119887 119909119894119895+ 119887)

where 119887 is a constant to denote the eyesight of the monkey If119891(119910) ge 119891(119909

119894) and position 119910 is feasible the 119894th monkey will

move from positions 119909119894to 119910 and go to Step 2

Step 4 Generate a random number 120572 isin [119888 119889] where [119888 119889] isthe somersault interval Get a new119910

119895= 119909119894119895+120572(119901119895minus119909119894119895) where

119901119895= (1119872)sum

119872

119894=1119909119894119895 119895 = 1 2 119899 If position 119910 is feasible

then force the monkey to go to the new position 119910 and begina search again

Step 5 If the results meet the end conditions the searchprocess is ended and output the position vector (the optimalsolution) Otherwise go to Step 2

Remark 1 Every step in this algorithm (including Steps 2 3and 4) will be repeated if the feasible and optimal positionscannot be found till the number of iteration of each stepmeets the maximum or some conditions are satisfied

One advantage of Monkey Climbing Algorithm is thepseudo-gradient calculation of the objective function in theclimb process which only requires two measurements of theobjective function regardless of the dimension of the real-world problem This advantage can reduce the cost of cal-culation and avoid the local minimum problem In additionMonkey Climbing Algorithm has a few parameters to adjustwhich makes it particularly easy to implement For exampleYi et al [18] used a collaborativeMonkeyClimbingAlgorithmto deal with the optimal sensor placement problem forhealth monitoring of high-rise structure The effectiveness of


the proposed algorithm is demonstrated by a numericalexample with a high-rise structure and the results showthat the proposed algorithm can provide a robust designfor sensor networks Suguna and Maheswari [19] proposeda monkey search optimization based feature selection andextraction approach for performing image classification andrecognition in the medical image segmentation process Theresults show that the proposed approach is an appropriatealgorithm for image segmentation

32 Inspired from Organism Structure Nature creates allthings where life is the most amazing one Each part oflife composition can give us some inspirations For exampleinspired from the double helix structure and complementarybase pairing rule of DNA Adleman proposed DNA Com-puting [20] inspired from the structure of biological cellsand the cooperation among different cells Paun presentedthe concept of Membrane Computing [21] inspired from theimmune mechanism of biological immune system Bersiniand Varela firstly proposed an Artificial Immune Systembasedmethod to deal with real-world problems [22] Further-more inspired from the linked structure and the informationtransmission mechanism in human beings or animal brainand neuron system some bioinspired neuron networks areproposed [23 24]

In this section three typical BIAs inspired from differentlevels of life are given out in detail respectively namelyDNA Computing (the level of chromosome) MembraneComputing (the level of cell) and Artificial Immune System(the level of system)

321 DNA Computing The study results of modern molec-ular biology indicate that the very complex structure ofliving body is the operation results on genetic informationrepresented by DNA sequence As ldquo00rdquo and ldquo10rdquo are used torepresent information in computer a single strand of DNAcan be looked at as a method for representing decodinginformation in themultitupleΣ = AGCT whereAGCand T denote adenine guanine cytosine and thyminerespectivelyThen the biological enzymes and other biochem-ical operations are the operators acted on the DNA sequenceBased on these ideas above Adleman published the papertitled ldquoMolecular Computation of Solutions to CombinatorialProblemsrdquo in the journal Science (1994) which means thebirth of a new research field DNA Computing [25]

The mapping between the DNA reaction in biology andthe process of DNA Computing is shown in Figure 3 Thebasic process of DNA Computing is (1) to code a problem byusing the special double helix structure and complementarybase pairing rules of DNA and to map objects into DNAmolecular chain (2) to produce various data pools by theactions of biological enzymes in the test tube with DNAsolution (3) to map the data operations of the original prob-lem into the controllable biochemical processes according tocertain rules (4) to obtain the calculating results by usingmolecular biological techniques such as polymerase chainreaction parallel overlap assembly affinity chromatographymolecular purification and magnetic bead separation

DNA Computing has two major advantages by using thefeatures of biological molecule and its biochemical reactionsthat are the massive storage and the parallel capacity It hasbeen applied in various fields and played a huge role such asclassification information security and robotic control [26ndash28] Of course there are still some disadvantages needed tobe further studied in DNA Computing For example theinformation of DNA cannot be copied and is difficult totransplant

322 Membrane Computing Membrane Computing wasfirstly presented by Paun in a research report in 1998 andthe related paper was published in 2000 [29] The goal ofMembrane Computing is not intended to model the functionof biological membranes but to get some new computingideas and design new computing models inspired from thecooperation among living cells and the cells of tissues organsor other structures In general a membrane system (alsocalled P-system see [29] for details) with 119898 degree can bedenoted by a multicomponent system

Π = (119881 119879 119862 120583 1199081 119908

119898 (1198771 1205881) (119877

119898 120588119898)) (5)

where 119881 is an alphabet whose elements are called objects119879 sube 119881 is an output alphabet 119862 sube 119881 minus 119879 is an activator 120583is a membrane structure containing 119898 membranes 119908

119894isin 119881lowast

is a multiset of the objects associated with the region 119894 of 120583where 119881lowast is a set of arbitrary strings composed of the stringsin119881119877

119894is a finite set of evolution rules and120588

119894is a partial order

relation over 119877119894 which is called a priority relation on the rule

of119877119894The evolution rule can be denoted by a pair (119906 V) where

119906 is a string in 119881 V = V1015840 or V = V1015840120575 (V1015840 is a string over the set119886here 119886out 119886in

119895

| 119886 isin 119881 1 le 119895 le 119898 where 119886here 119886out and119886in119895

are objects with different region destinations and 120575 is aspecial symbol not in 119881) If a rule containing 120575 is performedthis membrane is dissolved

In a nutshell themembrane system is constructed of threeparts hierarchy of membranes multiple sets for objects andevolutionary rules The membrane system uses evolutionaryrules in the membrane structure to complete the computingstarting from the initial state (represented by multiple sets ofobjects) [21 30]

The membrane system is a distributed massively par-allel and nondeterministic computational model Since theemergence of Membrane Computing it has attracted a lotof attention and become a powerful theoretical frameworkin the field of natural computing For example Cheng etal [31] presented a membrane algorithm for numericaloptimization which is an appropriate combination of a dif-ferential evolution algorithm a local search and P-systemsThe effectiveness of the algorithm was tested on extensivenumerical optimization experiments and the results showthat this algorithm performs better than its counterpartdifferential evolution algorithm Zhang et al [32] presented amembrane algorithm based on PSO for solving broadcastingproblems which are regarded asNP-hard combinatorial opti-mization problems The experimental results from variousbroadcasting problems show that the proposed algorithmhas better search capability efficiency solution stability and


Code the special problem by DNA

sequence

Produce all the possible solutions by

various operations

Select the optimalsolution by constraints

Decode the DNAsequence to get the

solution

Generate the initialDNA fragments

Controllable biochemical reactions

by enzymes

Extract the optimal DNA chain by

biology detection

Obtain new DNA fragments

The process ofDNA Computing

The process ofDNA reaction

Figure 3 The mapping between the DNA reaction in biology and the process of DNA Computing

precision than other optimization methods reported in theliterature In addition Membrane Computing can be usedin many other fields such as computer graphics linguisticseconomics computer science and cryptography [30 33 34]

323 Artificial Immune System Biological immune systemis a complex adaptive system in nature The purpose of theimmune system is to identify all cells (or molecules) inthe body and to distinguish the self and nonself Then thenonself cells are further classified in order to construct anappropriate defense mechanism Artificial Immune System isa bioinspired intelligent computing method inspired by theprinciples and processes of the vertebrate immune systemWith the advances in biology more and more immunemechanisms are applied in computing algorithms includingB-cells T-cells antibody antigen immunological learningimmunological memory clone selection and immunologicalresponse

At present the theoretical study of Artificial ImmuneSystem focuses on four aspects (1)Artificial immunenetworkmodel currently two influential artificial immune networkmodels are Resource-Limited Artificial Immune System andaiNet [35] (2) Clone selection clone selection is an importantdoctrine of biological immune system theory and its basicidea is that only those cells who can recognize antigenwill amplify to be selected and retained [36] (3) Immuneevolutionary algorithm immune evolutionary algorithm isa general name of various immune optimization algorithmswhich is developed from an evolutionary algorithm frame-work by introducing many features of the immune system[5] (4) Negative selection algorithm negative selection isan important mechanism to make the immune system havethe function of immune tolerance The process is that theimmature T-cells will die if they reply to self during the

growth in the thymus The mature T-cells will be distributedin the circulatory system of human body combine withthe nonself and tolerate self Based on this self-nonselfprinciple of the immune system Forrest et al [37] proposedthe negative selection algorithm which has become oneof the main artificial immune methods for their uniquecharacteristics

Artificial Immune System has some outstanding quali-ties such as noise patience learning without teacher self-organization and no need of negative example so it hasbeen used widely For example Li et al [35] proposed anartificial immune network based anticollision algorithm fordense RFID readers where the major immune operatorsare designed to satisfy the practical convention of RFIDsystems and the experiments results show that the artificialimmune network based method is effective and efficient inmitigating the reader-to-reader collision problem in denseRFID networks Cai et al [38] introduced the clonal selectionalgorithm into the problem of the recognition of communitydetection in complex networks and the experiments onboth synthetic and real-world networks demonstrate theeffectiveness of the proposed method Artificial ImmuneSystem is suited for dealing with the problems of featureselection anomaly detection and so on [39ndash41]

33 Inspired from Evolution Natural selection and survivalof the fittest is the worldrsquos eternal unchanging law In biologythe evolution refers to the changes in the population genetictraits between generations Natural selection can favor thegenetic traits of survival and reproduction to become morecommon and the harmful traits to become rarer Humanbeings as advanced living creatures form a special socialsystem which evolves mainly on social culture Inspiredfrom these complex evolution processes of lives some


(1) Set 119866119887= 1198661= 1198662= 120601

where 119866119887means the best genome 119866

1and 119866

2are two temporary genomes

(2) Initialize 119901119894119895= 1119899

119894

where 119901119894119895means the frequency of the allele 119886

119894119895

(3) Set 119866119887= select best()

Select the best genome from the gene pool(4) Do

1198661= select rand() and 119866

2= select rand()

Select two genomes randomly from the gene poolIf fitness(119866

1) ge fitness(119866

2)

reward(1198661) and penalize(119866

2)

If fitness(1198661) ge fitness(119866

119887)

119866119887= 1198661

where fitness() is a function to calculate the fitness of the genomesreward() and penalize() are two functions used to realize the tournament mechanism

Else


1)


119887)

119866119887= 1198662

while the stopping condition is reached(5) Return 119866

119887

Algorithm 1

evolution algorithms are proposed such as Genetic Algo-rithm Evolutionary Strategy Chemical Genetic ProgramInvasive Weed Algorithm and Culture Algorithm

From three different evolution levels three BIAs areintroduced in detail in this section to show the workmechanism and realization process of the BIAs inspiredfrom evolution namely Selfish Gene Algorithm (the levelof gene evolution) Invasive Weed Algorithm (the level ofpopulation evolution) and Culture Algorithm (the level ofsociety evolution) The first two algorithms belong to BIAsbased onbiological evolution andCultureAlgorithm is basedon society evolution

331 SelfishGeneAlgorithm SelfishGeneAlgorithm is a newmember of BIAs which is based on the selfish gene theorypresented by Dawkins [42] In the selfish gene theory theevolution is suggested to be viewed as acting at gene levelThe selection in organisms or populations is based on genesIn addition the population is seen as a genes pool wherethe number of individuals and their specific identities arenot of interest So Selfish Gene Algorithm focuses on thefitness of genes rather than individuals The main conceptsin Selfish Gene Algorithm are summarized as follows [43 44]and the pseudocode of SelfishGeneAlgorithm is summarizedin Algorithm 1

(1) Virtual Population The virtual population aims at mod-eling the gene pool concept defined by Dawkins which isdenoted by a vector

119901 = (1199011 1199012 119901

119899) (6)

where 119899 is the number of genes and 119901119894indicates the survival

rate of the 119894th gene In Selfish Gene Algorithm an individualis represented by its genome and the position in the genomeis called locus 119871

119894

(2) Allele The value appearing at the locus 119871119894is defined as

allele 119886119894119895 (119895 = 1 119899

119894) where 119899

119894is the number of gene in the

genome The success of an allele is based on the frequencywith which it appears in the virtual population

(3) EvolutionMechanismTheevolution of the virtual popula-tion proceeds by an unspecified kind of sexual reproductionof its individuals So Selfish Gene Algorithm does not relyon a crossover operator and the reproduction is performedimplicitly In Selfish Gene Algorithm evolution means thatthe organism which succeeds will increase its allelersquos fre-quency at the expense of its children and on the other handorganism which fails will decrease its allelersquos frequency

Selfish Gene Algorithm has been successfully tested inseveral problems For example Antonio [44] proposed astrategy based on the hybridization of Memetic Algorithmand Selfish Gene Algorithm to overcome the difficultiesin attaining a global solution for the optimization designproblem of composite structures Wang et al [45] exploitedthe selfish gene theory in the approach to improve the perfor-mance of the bivariate estimation of distribution algorithms

332 Invasive Weed Algorithm Invasive Weed Algorithm isintroduced by Mehrabian and Lucas [46] which simulatesthe weed invasion process Weeds have powerful viability


and survival strategies which turn them to undesirable plantsin agriculture However these features of weeds are veryuseful for certain computing algorithms There are threemain mechanisms of the invasive weed used in InvasiveWeed Algorithm (1) Propagation mechanism this mecha-nism means that the weeds will have different reproductivechances according to their fitness (2) Diffusion mechanismthis mechanism means that the offspring weeds will diffusewithin a space in a normal distribution mode taking theparent weeds as the axis (3) Competitive mechanism thismechanism means that all the weeds including the parentand offspring will be selected based on their merits whenthe number of weeds within a space reaches the upper limitAccording to the three mechanisms above the workflow ofInvasive Weed Algorithm is summarized as follows

Step 1 (population initialization) A certain number of weedsdiffuse in a119863 dimension space randomly

Step 2 (growth and reproduction) When the weeds grow tobloom they will produce seeds according to their fitnessThefitness of the parent weed and the number of its offspringweeds are in a linear relationship as follows

119873119904=119891 minus 119891min119891max minus 119891min

(119904max minus 119904min) + 119904min (7)

where 119873119904is the number of offspring weeds 119891 is the fitness

119891max and 119891min are the maximum and the minimum fitnessrespectively 119904max and 119904min are the maximum and the mini-mum number of the offspring weeds respectively

Step 3 (spatial diffusion) The offspring weeds diffuse into the119863 dimension space in a normal distribution mode and thestandard deviations 120590

119894of the normal distribution in different

iterations are changed according to the following rule

120590119894=(119894max minus 119894)

119899

(119894max)119899(120590initial minus 120590final) + 120590final (8)

where 119894 is the iteration number 119894max denotes the maximumiteration number 120590initial is the initial standard deviation 120590finalis the final standard deviation and 119899 is a nonlinear harmonicindex

Step 4 (competitive exclusion) When the bearing capacity ofenvironmental resources no longer satisfies the need of all theweeds the vulnerable groups namely weeds with low fitnesswill be weeded out

Step 5 If the end condition is satisfied output the optimalsolution Otherwise go to Step 2

There are some disadvantages in Invasive Weed Algo-rithm For example the parameter sensitivity is high thereis no information exchange among the individuals andthe searching deep is less than other algorithms such asParticle Swarm Optimization Algorithm and ExpectationMaximization Algorithm Now some improvements havebeen proposed in the standard InvasiveWeedAlgorithm and

most of them utilize the advantages of other algorithms todevelop a hybrid Invasive Weed Algorithm [47 48]

333 Culture Algorithm The evolution progress of humansociety is complex which is different from the naturalselection of any other living creatures The evolution ofhuman society is based on the heritage of knowledge andtransmission of culture Culture can make the populationevolve and adapt to the environment in a certain speed higherthan that of the general biological evolution relying solelyon gene inheritance Culture Algorithm [49] is proposedto simulate the evolution of human society In CultureAlgorithm the a priori knowledge in one field and theknowledge obtained from the evolution can be used to directthe searching process Culture Algorithm adopts a doubleevolutionary mechanism to simulate the culture evolutionprocess of human society A reliability space is established toextract various information in the evolutionary process basedon the traditional swarm evolutionary algorithms And theinformation is stored in the form of knowledge which is usedto lead the evolutionary process

There are three main parts in Culture Algorithm [550] (1) Population space the population space is used torealize any evolutionary algorithm based on populationwhere individuals are evaluated and the selection crossoverand mutation evolutionary operations are conducted Inaddition the excellent individuals are provided to the reli-ability space as samples (2) Reliability space the reliabilityspace receives samples from the population space by theacceptance function and selects the information in samplesby the update function which is abstracted described andstored in the form of knowledge Finally various knowledgeacts on the population space by the influence function toaccelerate the convergence of the evolution and improve theenvironmental adaptability of the algorithm (3) Commu-nication channels the communication channels include theacceptance function the update function and the influencefunction The population space and the reliability spaceare two independent evolution processes The acceptancefunction and the influence function provide channels for theupper knowledge model and the lower evolutionary processso the two functions are also called interface functions Thebasic structure of Culture Algorithm can be seen in [5]

Culture Algorithms offer power alternation for searchproblems which can also provide an understanding ofcultural phenomena and underlying technology utilized byhuman species Culture Algorithm can be used to deal withpattern recognition multirobot coordination fault classifica-tion and so on [50ndash52]

A brief summary of the bioinspired intelligent algorithmspresented in the subsections above is illustrated in Table 1where the main advantages and applications of these algo-rithms are given out And the related literatures are also listed

4 Applications Overview of BIAs forMobile Robot Control

As introduced in Section 3 BIAs have been used widely inlots of fields To introduce BIAs clearly we focus exclusively


Table 1 A brief summary of the BIAs introduced in this paper

Category Name Advantages Applications References

Inspired fromorganism behavior


Parameter insensitivity strongrobustness easy implementation

Image segmentation robot path planningoptimum scheduling optimal power flow [11ndash15 53]


A few parameters to adjust lowcalculation cost fast convergencerate

Optimal sensor placement feature selectionand extraction numerical optimization [16ndash19 54]


DNA Computing High parallelism massive storageability low energy consumption

Information security robotic control taskassignment problem clustering problem [20 25ndash28 55]

MembraneComputing

Inherent parallelism distributedfeature nondeterminism

Numerical optimization broadcastingproblem computer graphics travelingsalesman problem

[21 29ndash34]

Artificial ImmuneSystem

Noise patience learning withoutteacher self-organization andidentity

Community detection anomaly detectionfault diagnosis web page classification [5 22 35ndash41]



High convergent reliability andconvergent velocity

Optimization design problem travelingsalesman problem scheduling problem [42ndash45 57]


Easy to understand goodadaptability strong robustness

Image clustering problem parameterestimation problem numerical optimization [46ndash48 58 59]

Culture AlgorithmWith a double evolutionstructure high search efficiencywith a certain universality

Pattern recognition multirobot coordinationfault classification engineering designproblem

[5 49ndash52 60]

on the application field inmobile robot control in this sectionwhich is one of the most important application fields of BIAsThe main applications for mobile robot control based onBIAs summarized here are based on what the authors aremost aware of Although not comprehensive the applicationscited here can demonstrate some of the key features ofBIAs To reduce the repetition and introduce more BIAs theBIAs introduced in this section are different from those inSection 3

Mobile robot as an important robotic research branchdevelops very quickly which can be used to complete varioustasks that do not suit humans such as when the workingenvironment is dangerous and poisonous the task is fullof sameness and dull repetition and the exploration task isin the deep sea or outer space [61ndash63] The main problemsthat must be addressed in mobile robot control includepath planning simultaneous localization and mapping andcooperative control of multirobots which will be introducedin detail as follows

41 Path Planning Path planning is one of the basic problemsin the mobile robot control field which is very important torealize the intelligence and autonomy of mobile robots Thegoal of path planning is to find an optimal or suboptimalpath from the starting position to the target position [6465] Various methods have been used to deal with pathplanning problems such as the potential field methods andthe fuzzy control methods These methods have achievedcertain success However there are still some problemsneeded to be further studied including path planning inunknown and dynamic environments and the local mini-mum problem of most optimization algorithms Recentlysome BIAs have been proposed in solving the problems inpath planning of mobile robots The detailed information

of typical BIAs in robot path planning is described in thesequel

Siddique and Amavasai [66] proposed a behavior basedcontroller inspired by the concept of spinal fields foundin frogs and rats The core idea of this bioinspired be-havior-based controller is that the robot controller con-sists of a collection of spinal fields which are respon-sible for generating behaviors depending on their acti-vation levels The collection of spinal fields is denotedby 119861

1 1198612 1198613 119861

119870 The set of sensor fusion units is

denoted by 1198651 1198652 119865

119873 The mapping among sensors

fusion units and spinal fields is shown in Figure 4In their study four basic behaviors are chosen namelygo forward turn left turn right and go back Thefour behaviors are mapped into the corresponding robotwheel speeds 119881

119871 119881119877Then a function of spinal fields to real-

ize the behavior is expressed as 119881119871 119881119877 = 119891(119861

1 1198612 1198613 1198614)

Finally four experiments are carried out with Kheperarobot and the results show that their developed bioinspiredbehavior-based controller is simple but robust

Yang and Meng [67] used a bioinspired neural networkto realize the dynamic collision-free trajectory generation formobile robot in a nonstationary environment This bioin-spired neural network is based on a shunting model whichis obtained from a computational membrane model for apatch of membrane in a biological neural system (proposedby Hodgkin and Huxley in 1952 [23]) The core idea of thisbioinspired neural network based robot trajectory generationmethod is that the neural network is topologically organizedwhere the dynamics of each neuron is characterized by ashunting equation

119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8

Figure 4 Mapping among sensors fusion units and spinal system [66]

where 119909119894is the neural activity (membrane potential) of the 119894th

neuron 119860 119861 and 119863 are nonnegative constants representingthe passive decay rate and the upper and lower bounds of theneural activity respectively and 119878+

119894and 119878minus119894are the excitatory

and inhibitory inputs to the neuron The excitatory input 119878+119894

results from the target and the lateral connections from itsneighboring neurons while the inhibitory input 119878minus

119894results

from obstacles only Thus the differential equation for the 119894thneuron is given by

119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)

where 119902 is the number of neural connections of the 119894th neuronto its neighboring neurons within a receptive field 119908

119894119895is

the lateral connection weight from the 119894th neuron to the 119895thneuron function [120577]+ is defined as [120577]+ = max120577 0 andfunction [120577]minus is defined as [120577]minus = maxminus120577 0 119868

119894is the external

input to the 119894th neuron which is defined as

119868119894=

119864 if there is a target

minus119864 if there is an obstacle

0 otherwise

(11)

where 119864 is a positive constant and 119864 ≫ 119861The dynamic activ-ity landscape of the topologically organized neural network isused to determine the next robot location

119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)

where 119909119895 119895 = 1 2 119896 is the activity of all the neighboring

neurons of the present neuron (the present location of therobot) 119901

119899is the location of the neuron with the maximum

activity in these neurons (the next possible locations of therobot) One of the path planning results in a U-shaped

environment based on this bioinspired neural network isshown in Figure 5 where the generated path is shown inFigure 5(a) while the neural activity landscape in 3D is shownin Figure 5(b)

42 Simultaneous Localization and Mapping The solution tothe simultaneous localization andmapping (SLAM) problemis a key issue in the field of mobile robot control which isregarded as a ldquoHoly Grailrdquo of autonomous mobile robot [68]The core task of SLAM is that a robot explores in an unknownenvironment to learn the environment (map) by utilizingthe sensors onboard while simultaneously using that mapto locate within the environment A number of approacheshave been proposed to address the SLAM problem and themost typical and widely used SLAM algorithm is EKF-basedSLAM at present However the SLAM problems have notbeen resolved efficiently For example the accuracy of thesystem noise and the observation noise model will decide thefinal accuracy of the EKF-based SLAM algorithms Recentlymore and more attention has been focused on emulating thebiological system thought to be responsible for mapping andnavigation in animals which are introduced as follows

Milford and Wyeth [69] investigated the persistent nav-igation and mapping problem of an autonomous robotand a biologically inspired SLAM system based on modelsof mapping in the rodent hippocampus (RatSLAM) waspresented The proposed RatSLAM consists of three com-ponents a set of local view cells a network of pose cellsand an experience map (see [69] for details) The posecells are a three-dimensional continuous attractor network(CAN) For each pose cell local excitation and inhibition areachieved through a three-dimensional Gaussian distributionof weighted connections The distribution 120576 is given by

120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)

where 119896119901and 119896119889are variant constants for place and direction

respectively and 119886 119887 and 119888 represent the distances between


Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)

Figure 5 Real-time path planning of robot by the bioinspired neural network (a) the generated path of the robot (b) the neural activity ofthe bioinspired neural network

units in 1199091015840 1199101015840 and 1205791015840 coordinates respectivelyThe local viewcells are an array of rate-coded units used to represent whatthe robot is seeing The connections between local view cell119881119894and pose cell 119875

119909101584011991010158401205791015840 are stored in a matrix 120573 which are

calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)

where 120582 is the learning rate The experience map is a semi-metric topological map containing representations of places(called experiences) and links among experiences describingthe transitions among these places RatSLAM performsSLAM continuously while interacting with global and localnavigation systems and a task selection module that selectstasks among exploration delivery and rechargingmodesThereal robot experiments are conducted where the workplacesare floors in two different buildings at The University ofQueensland in Brisbane Australia The results demonstratethat RatSLAM can serve as a reliable mapping resource foran autonomous system in multiple environments

Barrera and Weitzenfeld [24] presented a robot architec-ture with spatial cognition and navigation capabilities thatcaptures some properties of rat brain structures involvedin learning and memory The detailed biological frameworkof their proposed model can be seen in their paper whichincludes four main modules path integration module land-marks processing module place representation module andlearning module The basic procedures of this model aresummarized as follows (1) the sensory information includingkinesthetic information visual information and affordancesinformation is input into the posterior parietal cortex (PPC)module which is suggested as part of a neural network

mediating path integration (2) the hippocampus receiveskinesthetic and visual information from the retrosplenialcortex and the entorhinal cortex existing in PPC modulerespectively (3) the internal state is combined with incen-tives which are under control of the lateral hypothalamus (4)all the process results (including the affordances perceptualschema the landmark perceptual schema and the learningresults) are integrated into the place representation modulecomprising a place cell layer and aworld graph layer to obtainthe decision-making basis for the action selection modulewhich will determine the robotic motor outputs From arobotics perspective this work can be placed in the gapbetween mapping and map exploitation currently existent inthe SLAM literature

43 Cooperation Control Multirobot systems have been asubject of much research since the 1970s for various tasksThe coordination ofmultirobot systems has recently attractedmore and more attention because a mobile robot team canaccomplish a task rapidly and efficiently [6 70] A lot of workhas been done on cooperation in multirobot systems andthe main issues in this field include hunting scheduling andforaging In this paper some literature related to these issuessolved by BIAs is overviewed

Tan et al [71] proposed a multirobot cooperative controlalgorithm for environment exploration based on immunenetwork model of B-T-cell In their algorithm the environ-ment which will be detected is considered as an antigenThe robot is considered as a B-cell The robotic behaviorstrategies are considered as antibodies produced by B-cellsThe control parameters of the algorithm are equivalent tothe regulatory effects of T-cells The dynamic equation for


the antibody levels of incentives and concentration is asfollows

Θ119894 (119905) = Θ119894 (119905 minus 1) + ΔΘ119894 (119905 minus 1) sdot 120579119894 (119905 minus 1)

ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)

119873minussum119873

119896=1119903119896119894120579119896 (119905 minus 1)

119873)

minus 119888119894 (119905 minus 1) minus 119896119894 + 120573119892119894

(15)

where Θ119894(119905) is the incentive level of the 119894th antibody 120579

119894(119905)

is the concentration of the 119894th antibody 119873 is the numberof the antibodies 119898

119894119895represents the affinity coefficients

between the 119894th and 119895th antibody 119903119896119894represents the rejection

coefficient between the 119896th and 119894th antibody 120572 representsthe interaction rate between the 119894th antibody and otherantibodies 120573 represents the interaction rate between the 119894thantibody and the antigen 119888

119894(119905) represents the concentration of

T-cell which regulates the antibody concentration of B-cell 119896119894

represents the natural mortality rate of the 119894th antibody and119892119894represents the matching rate between the 119894th antibody and

the antigen Each robot detects the environment according tothe matching rate between the antibody and the antigen tillthe exploration in the environment is completed

Yang andZhuang [72] presented an improvedAntColonyOptimization Algorithm for solving mobile agent (robot)routing problem The ants cooperate using an indirect formof communication mediated by pheromone trails of scentand find the best solution to their tasks guided by bothinformation (exploitation) which has been acquired andsearch (exploration) of the new routeWhen the ant 119896 travelsthe probability that the ant 119896 chooses the next host computerto be visited is

119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)

where 119901119896(119903 119904) is the probability that the ant 119896 chooses the

119904th host computer as the next destination 119903 is the currenthost computer which the ant is at 119889(119894 119895) is the time that amobile agent needs from the 119894th to the 119895th host computer120591(119894 119895) is the pheromone trail on the route between the119894th and 119895th host computer 119901

119894is the probability that the

mobile agent completes the task in the 119894th host computerand then the time delay in this host computer is denotedas 119905119894 119869119894(119903) is the set of the host computers that remain to

be visited by the ant 119894 positioned in the 119903th host computer120573 (0 le 120573 le 1) is a parameter to control the tradeoffbetween visibility (constructive heuristic) and pheromonetrail concentration Then the more the pheromone trails onthe route that the ants choose the shorter the delay time andthe higher the probability that the ants will choose that routeThe pheromone trails on the route can evaporate as time goeson and also the pheromone trails in the route can be updated

after all ants complete their tours respectively and return tothe initial host computerThe algorithmhas been successfullyintegrated into a simulated humanoid robot system whichwon the fourth place of RoboCup 2008 World Competition

44 Other Applications Lots of studies have been done onBIAs inmobile robot control besides those introduced abovesuch as machine vision robotic exploration and olfactorytracking

Villacorta-Atienza et al [73] proposed an internal rep-resentation neural network (IRNN) which can create com-pact internal representations (CIRs) of dynamic situationsdescribing the behavior of a mobile agent in an environmentwith moving obstacles Emergence of a CIR in IRNN can beviewed as a result of virtual exploration of the environmentThe general architecture of this IRNN is shown in Figure 6which consists of two coupled subnetworks Trajectory Mod-eling RNN (TM-RNN) and Causal Neural Network (CNN)Theoutput of TM-RNN is time independent and hence it justmaps the immobile objects into CNNwhose dynamicsmodelthe process of virtual exploration

119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895

is the neuron state variable representing theconcentration of virtual agents at the cell (119894 119895) the timederivative is taken with respect to the mental (inner) time 120591Δ119903119894119895= (119903119894+1119895

+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1

minus 4119903119894119895) denotes the discrete

Laplace operator describing the local (nearest neighbor)interneuronal coupling whose strength is controlled by 119889 (aconstant number) and 119901

119894119895accounts for the target if (119894 119895) is

occupied by a target 119901119894119895= 1 otherwise 119901

119894119895= 0 In their work

the effectiveness of IRNN is proved by some tests in differentsimulated environments including the environment with asingle moving obstacle and the realistic environments

Sturzl and Moller [74] presented an insect-inspiredapproach to orientation estimation for panoramic imagesThe relative orientation to a reference image 119868119886 can beestimated by simply calculating the image rotation thatminimizes the difference to the current image 119868119887 that is

119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)

where 119868119886(120601) denotes image 119868119886 rotated by angle 120601 around anaxis and the weights 120596119886119887

119894(120601) are calculated using 120596119886119887

119894(120601) =

(var119886119894(120601) + var119887

119894(0))minus1 where varminus1

119894is supposed to give infor-

mation about the quality of the pixel for rotation estimationFor example it should be low if the pixel belongs to an imagepart showing close objects In their minimalistic approachthe varivalues are computed from pixel differences of imagesrecorded at nearby position with the same orientation

Martinez et al [56] presented a biomimetic robot fortracking specific odors in turbulent plumes where an olfac-tory robot was constructed which consisted of a Koala robot


(i j)

Trajectory ModelingRNN

Casual NN

Sens

ory

inpu

t

Internal representation neural network

Figure 6 The sketch map of the internal representation neural network (IRNN) [73]

with an onboard computer and two electronic noses (E-noses) placed on both sides of the robot And a simplebiologically inspired strategy was proposed where the robotwas navigated with two spatially separated sensor arraysLet 119862

119897(119905) and 119862

119903(119905) be the odor concentrations estimated at

time 119905 by the sensors at the left and right respectively Theconcentration difference119862

119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When

only the side hit by an odor patch is known the turning speedof the robot 120596 would be

120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)

where Ω(119905) is an adaptive parameter for the robotic turningspeed The proposed bioinspired method is well suited forbuilding dedicated circuits and onboard implementation onreal robots Similarly a bioinspired autonomous swimmingrobot is designed to study the goal-directed locomotionwhich is inspired by the lamprey The robotic lamprey iscontrolled by a simulated central pattern generator networkthat generates the species-specific locomotormovements (see[75] for details)

5 Future Directions

Bioinspired intelligent computing is a relatively novel inter-disciplinary field of research and there are many theoreticalproblems needed to be solved The most important andwidely studied theoretical problems are convergence prob-lem effectiveness problem and evaluation standard problemBecause most of the BIAs are based on probability searchingit is difficult to rigorously prove the convergence and effec-tiveness of BIAs in mathematics Currently many theoreticalresearch results are aiming at one specific algorithm to carryout the research work One of the future works is to integratesome related algorithms to a common theory frameworkfor the proof of convergence and effectiveness In additionthe evaluation of BIAs should be conducted in a specificaspect for a specific task such as the computing speed or thecorrectness of the solution because it is difficult to find a BIAwhich is suitable for all the problems When we design analgorithm the practical principle should be followed and thealgorithm with unnecessary complex should not be selectedOther research branches in the theory study of BIAs includethe selection problem of initial parameters and the problemof convergence speed We believe that lots of novel BIAswill be developed constantly with the deepening of biologicalresearch

In the future BIAs will play an important role in moreand more fields In the application field for mobile robotcontrol based on BIAs the main future directions includebioinspired sensors cognitive model and bioinspired robotsIn the robotic sensors the development of the traditionalsensors meets a big problem that is it becomes more andmore difficult to improve the precision of the sensors andthe production cost of the sensors is higher and higherHowever as we know lots of organisms do not have anyhigh-precision sensors while they can achieve perfect sensefunction So the bioinspired sensors may be a good solutionfor the robotic sensors In the cognitive model currentlymany research results have been obtained based on theworking mechanism of the brains of rats and human beingsAnother important issue which should be of concern is theperformance of BIAs in real applications of mobile robotsIn the near future the robot with high intelligence self-learning and self-perception will become the mainstream inthe mobile robot field

6 Conclusions

In this survey we have analyzed the main features of BIAsbased on our work and the overview of the literature andgiven out a classification of BIAs from the biomimeticmechanismAccording to the classification we have surveyedthe realization process of each category of BIAs by someconcrete algorithms selected carefully Furthermore we havesurveyed the applications of BIAs focused on the mobilerobot control field Finally we have provided some possibledirections for future study Due to the large and growingliterature in this area many interesting results have not beenincluded in an attempt to capture some of the key areas in thisfield

It is clear that BIAs are going to be one of the hottestresearch points in the computation intelligent field includingtheory and application research And BIAs will play animportant role in mobile robot control which will be a goodsolution to improve the intelligence and autonomy of mobilerobot and can solve the development bottleneck of traditionaltechnologies such as the balance between precision andcost Currently many basic problems of mobile robot controlbased on BIAs have been explored and the results are excitingto demonstrate the potential of BIAs However most of theresults available are only conducted by simulation additionalefforts are needed to develop some more efficient BIAs andtransit these results to real applications in mobile robotcontrol


Conflict of Interests

The authors declared that they have no conflict of interestsregarding this work

Acknowledgments

The authors would like to thank the National NaturalScience Foundation of China (61203365 61573128) theJiangsu Province Natural Science Foundation (BK2012149)the Fundamental Research Funds for the Central Univer-sities (2015B20114 2015B14614) and the Jiangsu ProvinceInnovation Project for Postgraduate (KYLX15 0496) for theirsupport of this paperThe authors would also like to thank theauthors of all the references

References

[1] S Binitha and S S Sathya ldquoA survey of Bio inspired optimiza-tion algorithmsrdquo International Journal of Soft Computing andEngineering vol 2 no 2 pp 137ndash151 2012

[2] P Kopacek ldquoDevelopment trends in roboticsrdquo Elektrotechnik ampInformationstechnik vol 130 no 2 pp 42ndash47 2013

[3] J Bongard ldquoBiologically inspired computingrdquo Computer vol42 no 4 pp 95ndash98 2009

[4] C Chang ldquoUsing sensor habituation inmobile robots to reduceoscillatory movements in narrow corridorsrdquo IEEE Transactionson Neural Networks vol 16 no 6 pp 1582ndash1589 2005

[5] H Duan X Zhang and C Xu Bio-Inspired Computing SciencePress Beijing China 2011

[6] J Ni and S X Yang ldquoBioinspired neural network for real-time cooperative hunting by multirobots in unknown environ-mentsrdquo IEEE Transactions on Neural Networks vol 22 no 12pp 2062ndash2077 2011

[7] J Ni C Wang X Fan and S X Yang ldquoA bioinspired neuralmodel based extended Kalman filter for robot SLAMrdquo Mathe-matical Problems in Engineering vol 2014 Article ID 905826 11pages 2014

[8] J Yang and J Yang ldquoIntelligence optimization algorithms asurveyrdquo International Journal of Advancements in ComputingTechnology vol 3 no 4 pp 144ndash152 2011

[9] D Karaboga and B Akay ldquoA survey algorithms simulating beeswarm intelligencerdquo Artificial Intelligence Review vol 31 no 1ndash4 pp 61ndash85 2009

[10] B ChandraMohan and R Baskaran ldquoSurvey on recent researchand implementation of ant colony optimization in various engi-neering applicationsrdquo International Journal of ComputationalIntelligence Systems vol 4 no 4 pp 566ndash582 2011

[11] KM Passino ldquoBiomimicry of bacterial foraging for distributedoptimization and controlrdquo IEEE Control Systems vol 22 no 3pp 52ndash67 2002

[12] B Bhushan andM Singh ldquoAdaptive control of DCmotor usingbacterial foraging algorithmrdquo Applied Soft Computing Journalvol 11 no 8 pp 4913ndash4920 2011

[13] N Rajasekar N Krishna Kumar and R Venugopalan ldquoBacte-rial foraging algorithm based solar PV parameter estimationrdquoSolar Energy vol 97 pp 255ndash265 2013

[14] N Sanyal A Chatterjee and S Munshi ldquoAn adaptive bacterialforaging algorithm for fuzzy entropy based image segmen-tationrdquo Expert Systems with Applications vol 38 no 12 pp15489ndash15498 2011

[15] W Liu B Niu H Chen and Y Zhu ldquoRobot path planningusing bacterial foraging algorithmrdquo Journal of ComputationalandTheoretical Nanoscience vol 10 no 12 pp 2890ndash2896 2013

[16] R Zhao andW Tang ldquoMonkey algorithm for global numericaloptimizationrdquo Journal of Uncertain Systems vol 2 no 3 pp165ndash176 2008

[17] L Zheng ldquoAn improved monkey algorithm with dynamicadaptationrdquo Applied Mathematics and Computation vol 222pp 645ndash657 2013

[18] T-H Yi G-D Zhou H-N Li and X-D Zhang ldquoOptimalsensor placement for health monitoring of high-rise structurebased on collaborative-climb monkey algorithmrdquo StructuralEngineering and Mechanics vol 54 no 2 pp 305ndash317 2015

[19] S K Suguna and S U Maheswari ldquoPerformance analysis offeature extraction and selection of region of interest by seg-mentation in mammogram images between the existing meta-heuristic algorithms andMonkey SearchOptimization (MSO)rdquoWSEAS Transactions on Information Science and Applicationsvol 11 pp 72ndash82 2014

[20] N Pisanti ldquoDNA computing a surveyrdquo Bulletin of the EuropeanAssociation for Theoretical Computer Science no 64 pp 188ndash216 1998

[21] G Paun ldquoMembrane computing-a quick surveyrdquo Journal ofComputational and Theoretical Nanoscience vol 8 no 3 pp304ndash312 2011

[22] H Bersini and F J Varela ldquoHints for adaptive problem solvinggleaned from immune networksrdquo in Parallel Problem Solvingfrom Nature vol 496 of Lecture Notes in Computer Science pp343ndash354 Springer Berlin Germany 1991

[23] A L Hodgkin and A F Huxley ldquoA quantitative descriptionof membrane current and its application to conduction andexcitation in nerverdquoThe Journal of Physiology vol 117 no 4 pp500ndash544 1952

[24] A Barrera and A Weitzenfeld ldquoBiologically-inspired robotspatial cognition based on rat neurophysiological studiesrdquoAutonomous Robots vol 25 no 1-2 pp 147ndash169 2008

[25] L M Adleman ldquoMolecular computation of solutions to com-binatorial problemsrdquo Science vol 266 no 5187 pp 1021ndash10241994

[26] Z Ezziane ldquoDNA computing applications and challengesrdquoNanotechnology vol 17 no 2 pp R27ndashR39 2006

[27] H Jiao Y Zhong and L Zhang ldquoAn unsupervised spectralmatching classifier based on artificial DNA computing forhyperspectral remote sensing imageryrdquo IEEE Transactions onGeoscience and Remote Sensing vol 52 no 8 pp 4524ndash45382014

[28] B S E Zoraida M Arock B S M Ronald and R Ponala-gusamy ldquoDNA algorithm employing temperature gradient forFreeze-Tag Problem in swarm roboticsrdquo Transactions of theInstitute of Measurement and Control vol 34 no 2-3 pp 278ndash290 2012

[29] G Paun ldquoComputing with membranesrdquo Journal of Computerand System Sciences vol 61 no 1 pp 108ndash143 2000

[30] G-X Zhang and L-Q Pan ldquoA survey of membrane computingas a new branch of natural computingrdquo Chinese Journal ofComputers vol 33 no 2 pp 208ndash214 2010

[31] J Cheng G Zhang and X Zeng ldquoA novel membrane algorithmbased on differential evolution for numerical optimizationrdquoInternational Journal of Unconventional Computing vol 7 no3 pp 159ndash183 2011


[32] G Zhang F Zhou X Huang et al ldquoA novel membranealgorithm based on particle swarm optimization for solvingbroadcasting problemsrdquo Journal of Universal Computer Sciencevol 18 no 13 pp 1821ndash1841 2012

[33] M Li L Yi Z Gao Z Pei and H Peng ldquoSupport VectorMachine (SVM) based on membrane computing optimizationand the application for C-band radio abnormal signal identifi-cationrdquo Journal of Information and Computational Science vol11 no 11 pp 3683ndash3693 2014

[34] D Dıaz-Pernil A Berciano F Pena-Cantillana and M AGutierrez-Naranjo ldquoSegmenting images with gradient-basededge detection using membrane computingrdquo Pattern Recogni-tion Letters vol 34 no 8 pp 846ndash855 2013

[35] Z Li J Li and C He ldquoArtificial immune network-based anti-collision algorithm for dense RFID readersrdquo Expert Systemswith Applications vol 41 no 10 pp 4798ndash4810 2014

[36] G Chen Y Wang and Y Yang ldquoCommunity detection incomplex networks using immune clone selection algorithmrdquoInternational Journal of Digital Content Technology and itsApplications vol 5 no 6 pp 182ndash189 2011

[37] S Forrest L Allen A S Perelson and R Cherukuri ldquoSelf-nonself discrimination in a computerrdquo in Proceedings of theIEEE Computer Society Symposium on Research in Security andPrivacy pp 202ndash212 IEEE Oakland Calif USA May 1994

[38] Q Cai M Gong L Ma and L Jiao ldquoA novel clonal selectionalgorithm for community detection in complex networksrdquoComputational Intelligence vol 31 no 3 pp 442ndash464 2015

[39] U K Yuso M N A Khalid and A T Khader ldquoArtificialimmune system for flexible manufacturing system machineloading problemrdquo ICIC Express Letters vol 8 no 3 pp 709ndash716 2014

[40] M Marinaki and Y Marinakis ldquoA hybridization of clonal selec-tion algorithm with iterated local search and variable neigh-borhood search for the feature selection problemrdquo MemeticComputing vol 7 no 3 pp 181ndash201 2015

[41] D Li S Liu and H Zhang ldquoNegative selection algorithmwith constant detectors for anomaly detectionrdquo Applied SoftComputing vol 36 pp 618ndash632 2015

[42] R DawkinsThe Selfish Gene Oxford University Press OxfordUK 1976

[43] F CornoM SonzaReorda andG Squillero ldquoNew evolutionaryalgorithm inspired by the selfish gene theoryrdquo in Proceedings ofthe IEEE Conference on Evolutionary Computation (ICEC rsquo98)pp 575ndash580 Anchorage Alaska USA May 1998

[44] C C Antonio ldquoA memetic algorithm based on multiplelearning procedures for global optimal design of compositestructuresrdquoMemetic Computing vol 6 no 2 pp 113ndash131 2014

[45] F Wang Z Lin C Yang and Y Li ldquoUsing selfish gene theoryto construct mutual information and entropy based clusters forbivariate optimizationsrdquo Soft Computing vol 15 no 5 pp 907ndash915 2011

[46] A RMehrabian and C Lucas ldquoA novel numerical optimizationalgorithm inspired from weed colonizationrdquo Ecological Infor-matics vol 1 no 4 pp 355ndash366 2006

[47] Z Yin M Wen and C Ye ldquoImproved invasive weed optimiza-tion based on hybrid genetic algorithmrdquo Journal of Computa-tional Information Systems vol 8 no 8 pp 3437ndash3444 2012

[48] G G Roy S Das P Chakraborty and P N Suganthan ldquoDesignof non-uniform circular antenna arrays using a modifiedinvasive weed optimization algorithmrdquo IEEE Transactions onAntennas and Propagation vol 59 no 1 pp 110ndash118 2011

[49] R G Reynolds ldquoAn introduction to cultural algorithmsrdquo inProceedings of the 3rd Annual Conference on EvolutionaryProgramming pp 131ndash139 San Diego Calif USA 1994

[50] M Z Ali and R G Reynolds ldquoCultural algorithms a Tabusearch approach for the optimization of engineering designproblemsrdquo Soft Computing vol 18 no 8 pp 1631ndash1644 2014

[51] Y-N Guo M Yang and J Cheng ldquoPath planning methodfor robots in complex ground environment based on culturalalgorithmrdquo in Proceedings of the 1st ACMSIGEVO Summit onGenetic and Evolutionary Computation (GEC rsquo09) pp 185ndash191Shanghai China June 2009

[52] Q Wu J Zhang W Huang and Y Sun ldquoAn efficient imagematching algorithm based on culture evolutionrdquo Journal ofChemical and Pharmaceutical Research vol 6 no 5 pp 271ndash278 2014

[53] W J Tang M S Li Q H Wu and J R Saunders ldquoBacterialforaging algorithm for optimal power flow in dynamic envi-ronmentsrdquo IEEE Transactions on Circuits and Systems I RegularPapers vol 55 no 8 pp 2433ndash2442 2008

[54] M Soleimanpour and H Nezamabadi-Pour ldquoA modified mon-key algorithm for real-parameter optimizationrdquo Journal ofMultiple-Valued Logic and Soft Computing vol 21 no 5-6 pp453ndash477 2013

[55] Z Wang J Tan D Huang Y Ren and Z Ji ldquoA biologicalalgorithm to solve the assignment problem based on DNAmolecules computationrdquo Applied Mathematics and Computa-tion vol 244 pp 183ndash190 2014

[56] D Martinez O Rochel and E Hugues ldquoA biomimetic robotfor tracking specific odors in turbulent plumesrdquo AutonomousRobots vol 20 no 3 pp 185ndash195 2006

[57] A VillagraM De San PedroM Lasso andD Pandolfi ldquoMulti-recombined evolutionary algorithm inspired in the selfish genetheory to face the weighted tardiness scheduling problemrdquo inAdvances inArtificial IntelligencemdashIBERAMIA 2004 vol 3315 ofLectureNotes inComputer Science pp 809ndash819 Springer BerlinGermany 2004

[58] X Zhao and J Zhou ldquoImproved kernel possibilistic fuzzyclustering algorithm based on invasive weed optimizationrdquoJournal of Shanghai Jiaotong University vol 20 no 2 pp 164ndash170 2015

[59] A Ouyang L Liu Z Sheng and F Wu ldquoA class of parameterestimation methods for nonlinear muskingum model usinghybrid invasive weed optimization algorithmrdquo MathematicalProblems in Engineering vol 2015 Article ID 573894 15 pages2015

[60] P Rakshit A Konar P Bhowmik et al ldquoRealization of anadaptive memetic algorithm using differential evolution andq-learning a case study in multirobot path planningrdquo IEEETransactions on Systems Man and Cybernetics Part ASystemsand Humans vol 43 no 4 pp 814ndash831 2013

[61] D Lee G Kim D Kim H Myung and H-T Choi ldquoVision-based object detection and tracking for autonomous navigationof underwater robotsrdquo Ocean Engineering vol 48 pp 59ndash682012

[62] K Yoshida ldquoAchievements in space roboticsrdquo IEEERobotics andAutomation Magazine vol 16 no 4 pp 20ndash28 2009

[63] B Sun D Zhu and S X Yang ldquoA bioinspired filtered backstep-ping tracking control of 7000-m manned submarine vehiclerdquoIEEE Transactions on Industrial Electronics vol 61 no 7 pp3682ndash3693 2014

[64] J Ni and S X Yang ldquoA fuzzy-logic based chaos GA for coop-erative foraging of multi-robots in unknown environmentsrdquo


International Journal of Robotics and Automation vol 27 no 1pp 15ndash30 2012

[65] D Zhu H Huang and S X Yang ldquoDynamic task assignmentand path planning of multi-AUV system based on an improvedself-organizing map and velocity synthesis method in three-dimensional underwater workspacerdquo IEEE Transactions onCybernetics vol 43 no 2 pp 504ndash514 2013

[66] N H Siddique and B P Amavasai ldquoBio-inspired behaviour-based controlrdquo Artificial Intelligence Review vol 27 no 2-3 pp131ndash147 2007

[67] S X Yang andMQ-HMeng ldquoReal-time collision-freemotionplanning of a mobile robot using a neural dynamics-basedapproachrdquo IEEE Transactions on Neural Networks vol 14 no6 pp 1541ndash1552 2003

[68] M W M Gamini Dissanayake P Newman S Clark H FDurrant-Whyte andMCsorba ldquoA solution to the simultaneouslocalization and map building (SLAM) problemrdquo IEEE Trans-actions on Robotics and Automation vol 17 no 3 pp 229ndash2412001

[69] M Milford and G Wyeth ldquoPersistent navigation and mappingusing a biologically inspired SLAM systemrdquo International Jour-nal of Robotics Research vol 29 no 9 pp 1131ndash1153 2010

[70] A Zhu and S X Yang ldquoA neural network approach to dynamictask assignment of multirobotsrdquo IEEE Transactions on NeuralNetworks vol 17 no 5 pp 1278ndash1286 2006

[71] Y Tan Y Fang and J Yu ldquoApplication of improved immunealgorithm to multi-robot environment explorationrdquo Interna-tional Journal of Advancements in Computing Technology vol4 no 16 pp 158ndash164 2012

[72] J Yang and Y Zhuang ldquoAn improved ant colony optimizationalgorithm for solving a complex combinatorial optimizationproblemrdquoApplied SoftComputing Journal vol 10 no 2 pp 653ndash660 2010

[73] J A Villacorta-Atienza M G Velarde and V A MakarovldquoCompact internal representation of dynamic situations neu-ral network implementing the causality principlerdquo BiologicalCybernetics vol 103 no 4 pp 285ndash297 2010

[74] W Sturzl and R Moller ldquoAn insect-inspired active visionapproach for orientation estimation with panoramic imagesrdquo inBio-Inspired Modeling of Cognitive Tasks Second InternationalWork-Conference on the Interplay BetweenNatural andArtificialComputation IWINAC 2007 La Manga del Mar Menor SpainJune 18-21 2007 Proceedings Part I vol 4527 of Lecture Notes inComputer Science pp 61ndash70 Springer Berlin Germany 2007

[75] L Manfredi T Assaf S Mintchev et al ldquoA bioinspiredautonomous swimming robot as a tool for studying goal-directed locomotionrdquo Biological Cybernetics vol 107 no 5 pp513ndash527 2013

Submit your manuscripts athttpwwwhindawicom

Computer Games Technology

International Journal of

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Distributed Sensor Networks


Advances in

FuzzySystems

Hindawi Publishing Corporationhttpwwwhindawicom

Volume 2014


ReconfigurableComputing

Hindawi Publishing Corporation httpwwwhindawicom Volume 2014


Applied Computational Intelligence and Soft Computing

thinspAdvancesthinspinthinsp

Artificial Intelligence

HindawithinspPublishingthinspCorporationhttpwwwhindawicom Volumethinsp2014

Advances inSoftware EngineeringHindawi Publishing Corporationhttpwwwhindawicom Volume 2014


Electrical and Computer Engineering

Journal of

Journal of

Computer Networks and Communications


Hindawi Publishing Corporation

httpwwwhindawicom Volume 2014

Advances in

Multimedia


Biomedical Imaging


ArtificialNeural Systems

Advances in


RoboticsJournal of



Computational Intelligence and Neuroscience

Industrial EngineeringJournal of


Modelling amp Simulation in EngineeringHindawi Publishing Corporation httpwwwhindawicom Volume 2014

The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Human-ComputerInteraction

Advances in

Computer EngineeringAdvances in



the future research directions for the theory of BIAs and theirapplications in mobile robot control Finally conclusions aregiven out in Section 6

2 Classification of BIAs

There are two main aspects for human beings to learn fromnature The first one is to make inventions by imitating thestructures of organisms For example human beings createdthemanned glider flight based on the flying principle of birdsand invented the echolocation sonar system inspired from theecholocation function of bats Another one is to design somealgorithms (technologies) inspired from various principles ofnature For example based on the process of ant foraging AntColony Algorithm was developed and based on the processof natural evolution such as inheritance mutation selectionand crossover Genetic Algorithm was proposed With thedevelopment of science and technology human beings willface more and more challenges Some of the traditionalartificial intelligent algorithms are too simple in structures orfunctions to satisfy the requirements of developments Fromthe late 20th century more and more novel artificial intel-ligent algorithms have been proposed such as bioinspiredneural network Artificial Immune Algorithm and Mem-brane ComputingAlgorithmThese novel artificial intelligentmethods as well as some other intelligent algorithms (suchas Ant Colony Algorithm and Genetic Algorithmmentionedabove) have a common feature Namely they have morelifelike biological working mechanisms than other types Theartificial intelligent methods of this type are the bioinspiredintelligent algorithms (BIAs) discussed in this paper

Now BIAs are still in the stage of development so thereis no strict definition and uniform classification Binithaand Sathya [1] described the origin and advantages of thebioinspired computing algorithms and pointed out that BIAswere heuristic methods that imitated the strategy of naturewhich was a simple and nonrepresentational definition ofBIAs Bongard [3] introduced the development process ofthe bioinspired computing algorithm and analyzed the rela-tionship between the BIAs and the traditional intelligentcomputing methods Based on our expertise the bioinspiredintelligent computing algorithm can be defined as followsit is a type of intelligent computing methods with a quitelifelike biologicalworkingmechanism to imitate the functionand structure of the organism the individual and swarmbehaviors and the evolution process of life and societyHowever it is not an easy job to exclude those methodsfrom BIAs which are not strictly bioinspired In this paperwe do not intend to distinguish different types of intelligentcomputing algorithms but to analyze the main features ofBIAs classify these algorithms from the simulated biologicalworkingmechanisms and survey different categories focusedon the realization processing as well as the applications formobile robot control

Based on our research work and the overview of lots ofliterature the remarkable features of BIAs discussed in thispaper are summarized as follows [1 4ndash7]

(1) Bioinspired feature the working mechanism is veryclose to the biological or ecological mechanism of

natural organismsThebioinspired algorithms imitatethe biological nature as much as possible to deal withthe real-world problems

(2) Simplicity and emergence the strategy and computa-tion are often very simple but their resultant effectsare very amazing which reflects the principle ofemergence

(3) Robustness these algorithms have strong robustnessagainst the change of environments parameters andtasks namely these algorithms have good applicabil-ity and flexibility

(4) Self-organization these algorithms can improve theadaptability by self-learning or self-organization andrealize the evolution successfully

(5) Other features these algorithms have some othergood features such as parallelism in essence and non-determinism

As introduced above BIAs have many excellent charac-teristics that can meet the needs of researchers To introduceBIAs clearly and understand them easily BIAs should beclassified From different view angle various classificationscan be obtained for example from the whole effects we cansee all BIAs as a type of evolutional optimization algorithmsIn this paper we will classify BIAs from their sources of thebiomimetic mechanism then BIAs can be divided into threecategories (1) inspired from organism behaviors (2) inspiredfrom organism structure (3) inspired from evolution Theclassification in this paper is single that is to say onebioinspired algorithm will not be classified into two ormore different categories The classification map is shown inFigure 1 and each category will be introduced in the nextsection clearly by some typical algorithms

3 Theories Overview

There are some similarities in the property and application ofeach category with the same biomimetic mechanism In thispaper we will focus on the realization processing of severalnovel algorithms proposed in the last two decades as well assome distinctive or representative methods as far as we know

31 Inspired from Organism Behaviors Every living creaturecan survive and protect itself in its own way such as foragingof food defending against natural enemy and courtshipdisplaying Based on the characteristics of these importantbehaviors of organisms some BIAs have been developedIn these BIAs most of them are inspired from the foragingbehaviors of organisms such as Ant Colony AlgorithmBee Colony Algorithm Bacterial Foraging Algorithm FishSwarm Algorithm and Shuffled Frog Leaping AlgorithmThere are also some algorithmswhich are inspired from otherbehaviors of organisms For example Monkey ClimbingAlgorithm simulates the climbing behavior of monkeys andBacterial Chemotaxis Algorithm simulates the chemotaxisbehavior of bacteria There are several other surveys of theliterature on the behavior-based BIAs (see [8ndash10]) so we just








Ant ColonyAlgorithm

Bee ColonyAlgorithm




DNA

Membrane







Culture Firefly

algorithm




Computing

Computing

Algorithm






120579119894(119895 + 1 119896 119897) = 120579

119894(119895 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (1)


120579119894(119895 + 1 119896 119897) = 120579

119894(119895 + 1 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (2)


119869119894

health =

119873119888+1

sum

119895=1

119869 (119894 119895 119896 119897) (3)





(a) (b) (c)







119894= (1199091198941 1199091198942 119909


of the solution


1 1199102 119910

119899) and 119910

119894= 119909119894119895+ 119886 sdot sgn(1198911015840

119894119895(119909119894)) where




1198911015840

119894119895(119909119894) =

119891 (119909119894+ Δ119909119894) minus 119891 (119909

119894minus Δ119909119894)

2Δ119909119894119895

119895 = 1 2 119899

(4)

where Δ119909119894= (Δ119909

1198941 Δ1199091198942 Δ119909

119894119899) and Δ119909

119894119895= 119886 or minus119886


119894with 119910


119895sube (119909119894119895minus 119887 119909119894119895+ 119887)





119895= 119909119894119895+120572(119901119895minus119909119894119895) where

119901119895= (1119872)sum

119872














Π = (119881 119879 119862 120583 1199081 119908

119898 (1198771 1205881) (119877

119898 120588119898)) (5)









119895







sequence


various operations



solution



by enzymes


biology detection












(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in










Ant ColonyAlgorithm

Bee ColonyAlgorithm




DNA

Membrane







Culture Firefly

algorithm




Computing

Computing

Algorithm






120579119894(119895 + 1 119896 119897) = 120579

119894(119895 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (1)


120579119894(119895 + 1 119896 119897) = 120579

119894(119895 + 1 119896 119897) + 119862 (119894)

Δ (119894)

radicΔT (119894) Δ (119894) (2)


119869119894

health =

119873119888+1

sum

119895=1

119869 (119894 119895 119896 119897) (3)





(a) (b) (c)







119894= (1199091198941 1199091198942 119909


of the solution


1 1199102 119910

119899) and 119910

119894= 119909119894119895+ 119886 sdot sgn(1198911015840

119894119895(119909119894)) where




1198911015840

119894119895(119909119894) =

119891 (119909119894+ Δ119909119894) minus 119891 (119909

119894minus Δ119909119894)

2Δ119909119894119895

119895 = 1 2 119899

(4)

where Δ119909119894= (Δ119909

1198941 Δ1199091198942 Δ119909

119894119899) and Δ119909

119894119895= 119886 or minus119886


119894with 119910


119895sube (119909119894119895minus 119887 119909119894119895+ 119887)





119895= 119909119894119895+120572(119901119895minus119909119894119895) where

119901119895= (1119872)sum

119872














Π = (119881 119879 119862 120583 1199081 119908

119898 (1198771 1205881) (119877

119898 120588119898)) (5)









119895







sequence


various operations



solution



by enzymes


biology detection












(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(a) (b) (c)







119894= (1199091198941 1199091198942 119909


of the solution


1 1199102 119910

119899) and 119910

119894= 119909119894119895+ 119886 sdot sgn(1198911015840

119894119895(119909119894)) where




1198911015840

119894119895(119909119894) =

119891 (119909119894+ Δ119909119894) minus 119891 (119909

119894minus Δ119909119894)

2Δ119909119894119895

119895 = 1 2 119899

(4)

where Δ119909119894= (Δ119909

1198941 Δ1199091198942 Δ119909

119894119899) and Δ119909

119894119895= 119886 or minus119886


119894with 119910


119895sube (119909119894119895minus 119887 119909119894119895+ 119887)





119895= 119909119894119895+120572(119901119895minus119909119894119895) where

119901119895= (1119872)sum

119872














Π = (119881 119879 119862 120583 1199081 119908

119898 (1198771 1205881) (119877

119898 120588119898)) (5)









119895







sequence


various operations



solution



by enzymes


biology detection












(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in











Π = (119881 119879 119862 120583 1199081 119908

119898 (1198771 1205881) (119877

119898 120588119898)) (5)









119895







sequence


various operations



solution



by enzymes


biology detection












(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in





sequence


various operations



solution



by enzymes


biology detection












(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(1) Set 119866119887= 1198661= 1198662= 120601


1and 119866


(2) Initialize 119901119894119895= 1119899

119894


119894119895

(3) Set 119866119887= select best()


1198661= select rand() and 119866

2= select rand()



2)


2)


119887)

119866119887= 1198661


Else


1)


119887)

119866119887= 1198662


119887

Algorithm 1





119901 = (1199011 1199012 119901

119899) (6)



119894


allele 119886119894119895 (119895 = 1 119899

119894) where 119899











(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in








(119904max minus 119904min) + 119904min (7)






120590119894=(119894max minus 119894)

119899


























MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in
















MembraneComputing



[21 29ndash34]













[5 49ndash52 60]






1 1198612 1198613 119861


denoted by 1198651 1198652 119865





1 1198612 1198613 1198614)



119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894) 119878+

119894minus (119863 + 119909

119894) 119878minus

119894 (9)


Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Sensors

Information fusion

Spinal system

Robot control

Robot

B1

B2

B3

BK

F1

F2

F3

FN

S1

S2

S3

S4

S5

S6

S7

S8







119894results


119889119909119894

119889119905= minus119860119909

119894+ (119861 minus 119909

119894)([119868119894]++

119902

sum

119895=1

119908119894119895[119909119895]+

)

minus (119863 + 119909119894) [119868119894]minus

(10)


119894119895is




119868119894=



0 otherwise

(11)


119901119899lArr997904 119909119901119899

= max 119909119895 119895 = 1 2 119896 (12)








120576119886119887119888

= 119890minus(1198862

+1198872

)119896exc119901 119890minus1198882

119896exc119889 minus 119890minus(1198862

+1198872

)119896inh119901 119890minus1198882

119896inh119889 (13)




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




Obstacle

Target

Robot

Obstacle

0

5

10

15

20

25

30

Y

5 10 15 20 25 300X

(a)

10

2030

1020

30

X

Y

Obstacles

Target

minus1

minus05

0

05

1

Neu

ron

activ

ity(b)




calculated by

120573119905+1

119894119909101584011991010158401205791015840 = max (120573119905

119894119909101584011991010158401205791015840 1205821198811198941198751199091015840 11991010158401205791015840) (14)









ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






ΔΘ119894 (119905 minus 1)

= 120572(sum119873

119895=1119898119894119895120579119895 (119905 minus 1)


119896=1119903119896119894120579119896 (119905 minus 1)

119873)


(15)


119894(119905)










119901119896 (119903 119904)

=

[120591 (119903 119904)] [119901119904119889 (119903 119904) sdot 119905119904]120573

sum119906isin119869119894(119903)[120591 (119903 119906)] [119901119906119889 (119903 119906) sdot 119905119906]

120573 119904 isin 119869

119896 (119903)

0 otherwise

(16)









119903119894119895= 119889 sdot Δ119903

119894119895minus 119903119894119895sdot 119901119894119895 (17)

where 119903119894119895


+ 119903119894minus1119895

+ 119903119894119895+1

+ 119903119894119895minus1





119894119895= 0 In their work



119886119887

= argmin120601

(sum

119894

120596119886119887

119894(120601))

minus1

sdot sum

119894

120596119886119887

119894(120601) (119868

119886

119894(120601) minus 119868

119887

119894)2

(18)



119894(120601) =

(var119886119894(120601) + var119887






(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in




(i j)


Casual NN

Sens

ory

inpu

t




119897(119905) and 119862



119897(119905)minus119862

119903(119905) is noted asΔ119862(119905)When


120596 (119905) = Ω (119905) sdot sgn (Δ119862 (119905)) (19)


5 Future Directions



6 Conclusions






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in






Acknowledgments


References






















































































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in

























































Advances in

FuzzySystems


Volume 2014












Journal of

Journal of





Advances in

Multimedia


Biomedical Imaging



Advances in


RoboticsJournal of










Advances in



review article bioinspired intelligent algorithm and...

Documents