Research Article Cognitive Emotional Regulation Model in ...downloads.hindawi.com/journals/ddns/2015/829387.pdf · emotions. From the eld theory, the activated intensity at any point

Research ArticleCognitive Emotional Regulation Modelin Human-Robot Interaction

Xin Liu1 Lun Xie1 Anqi Liu2 and Dan Li1

1School of Computer and Communication Engineering University of Science and Technology Beijing Beijing 100083 China2Beijing Shougang International Engineering Technology Limited Company Beijing 100043 China

Correspondence should be addressed to Lun Xie xielunustbeducn

Received 23 September 2014 Revised 20 December 2014 Accepted 21 December 2014

Academic Editor Qingang Xiong

Copyright copy 2015 Xin Liu et al This is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited

This paper integrated Gross cognitive process into the HMM (hidden Markov model) emotional regulation method andimplemented human-robot emotional interactionwith facial expressions and behaviors Here energy was the psychological drivingforce of emotional transition in the cognitive emotional model The input facial expression was translated into external energy byexpression-emotion mapping Robotrsquos next emotional state was determined by the cognitive energy (the stimulus after cognition)and its own current emotional energyrsquos size and sourcersquos positionThe two random quantities in emotional transition processmdashtheemotional family and the specific emotional state in the AVS (arousal-valence-stance) 3D spacemdashwere used to simulate humanemotion selection The model had been verified by an emotional robot with 10 degrees of freedom and more than 100 kinds offacial expressions Experimental results show that the emotional regulationmodel does not simply provide the typical classificationand jump in terms of a set of emotional labels but that it operates in a 3D emotional space enabling a wide range of intermediaryemotional states to be obtained So the robot with cognitive emotional regulation model is more intelligent and real moreover itcan give full play to its emotional diversification in the interaction

1 Introduction

Nowadays robot not only needs intelligent behavior butalso needs mental life such as cognition emotion andpersonality Robot evolves its own emotional intelligence foranthropomorphic and diversified states and even producesempathy Human-robot interaction requires emotional anal-ysis and regulation so emotional modeling is particularlyimportant In this section several valued and far-reachingapproaches about emotion modeling have been proposedThey can be divided into two categories the discrete modeland the emotional space model Moreover the two categoriesare independent and complement

11 The Discrete Model Izzard divides emotions into twocategories prototypical emotions and complex emotionsThe prototypical emotions include more or less discreteemotional states usually from 2 to 11 [1ndash3] Lazarus believesthat the growing importance of cognitive-mediational orvalue-expectancy approaches to mind and behavior in social

sciences has promoted the prosperity of emotions as discretecategories [4] Ekman proposes six prototypical emotionsbased on the facial expressions [5] Besides this analo-gous approach is followed by several authors The artificialemotions are divided into anger boredom fear happinessinterest and sadness in Canamerorsquos works with social robots[6] In Gadanhorsquos approach emotions (happiness fear sad-ness and anger) are related to certain events [7] Velasquezalso proposes an emotion-based control for autonomousrobots In his research six prototypical emotions (anger fearsorrow happiness disgust and surprise) are implementedwith innate personality and the capacity of acquired learning[8] Murphy et al put forward the artificial emotional states(happy confident concerned and frustrated) for multiagentsystems and the emotions are released depending on thetask process [9] Complex emotions consist of the followingthree categories (1) 2-3 prototypical emotions mixed (2)prototypical emotions and inner impulse mixed (3) pro-totypical emotions and affective-cognitive structure mixed[10]

Hindawi Publishing CorporationDiscrete Dynamics in Nature and SocietyVolume 2015 Article ID 829387 8 pageshttpdxdoiorg1011552015829387

2 Discrete Dynamics in Nature and Society

12 The Emotional Space Model Unlike discrete modelsthe emotional space models consider a continuous multi-dimensional space where each point stands for an emo-tional state and each dimension stands for a fundamen-tal property common to all emotions In order to fullyshow the emotional properties the classical 3D emotionalspace pleasantunpleasant excitementdepression and ten-sionrelaxation has already been used by Wundt [11] Formany years a large number of emotional dimensional the-ories have been proposed [12ndash15] One of the most acceptedtheories is described by Russell andMehrabian [16] pleasure-arousal-dominance space (PAD) Scherer and Ekam andOrtony et al also use and develop PAD space to deter-mine artificial emotions for social robots [17 18] Zecca etal construct the 3D psychological vector space arousal-pleasant-certain withmachine learning dynamic regulationand personality [19] In addition Breazealrsquos arousal-valence-stance (AVS) space with the social robot Kismet is the mostnoteworthy emotional space model [20] In that case theemotional state of the robot varies according to its interactionwith people In fact the external stimulus is labeled by threeemotional properties in the AVS space Ortony et al modelthe OCC reasoning process that can produce the cognitiveemotions and touch off complex emotional experiences viathe trend of events (including event object and agent) [18]Then this model has been developed and refined by Yang etal and Kim et al [21 22]

This paper discussed a cognitive emotional regulationmodel in the active field state space First based on Grosscognitive strategy the cognitive reappraisal model satisfiedFischna-Weber law was built and a physiological endurancecoefficient was defined to meet the psychological diver-sity Second emotional state space in the active field wasestablished on the solid basis of the physiological energydistribution and the transition probability among the emo-tional families was figured out Third an observational statesequence was yielded from emotional family by HMMFinally the emotional regulationmodel is devoted to human-robot experiment for the validity analysis Figure 1 gives outgeneral ideas of this research

The rest of this paper is organized as follows Section 2discusses the Gross cognitive strategy and presents a cogni-tive reappraisal model Section 3 defines the emotional spaceand related state transition process Section 4 introducesthe emotional robot platform Section 5 shows experimentalresults and discussions The conclusion and directions forfuture work are offered in Section 6

2 Emotional Cognition and Modeling

21 Gross Cognitive Strategy Gross proposed five emotionalregulation strategiesmdashsituation selection situation modi-fication attention deployment cognitive reappraisal andresponse suppression [23] Cognitive reappraisal the fore-most antecedent-focused strategy in the early emotionalregulation stage depends on the internal willpower andpersonality factorsWhen the individual is in troubles or psy-chological expectation is different to the reality the negative

External stimulus emotion from

expression

Negative emotion

Cognitive reappraisal

Source distribution in emotional sates space

Emotional states transition probability

Next emotion

Yes

No

Current emotion

Robot expression

Cognitive emotional regulation model

Figure 1 General ideas of this research

emotions such as sadness anxiety anger and pain areproduced In general the cognitive reappraisal can well cutdown the negative emotional intensity and improve theindividual emotional experience In the cognitive processcognitive reappraisal and response suppression strategiesboth can improve the individual emotional experience How-ever the psychological research has proven that the cognitivereappraisal strategy as compared with the four other regu-lation strategies is most effective in the negative emotionalregulation

22 Cognitive Reappraisal Model Gross believes that spon-taneity cognitive reappraisal as an essential part of psy-chological defense mechanism could help individuals toconsider the emotional stimulus from the peaceful perspec-tive [23] Thus facing various external emotional stimuluspsychological systemwill first unconsciously start the sponta-neous cognitive reappraisal to reduce the emotional intensityIn other words when the external stimulus appears inemotional state space the spontaneous cognitive reappraisalcomes into play at once for cutting down the stimulus-intensity Because the spontaneous cognitive reappraisalmainly depends on individual physiological endurance aphysiological endurance coefficient 120594 isin (0 1) is defined forrobot

Fischna-Weber law is a psychophysical formula describ-ing the relationship of internal feel-intensity and externalstimulus-intensity The law shows that human feel-intensityis proportional to the logarithm of stimulus-intensity (shownin formula (1)) Under the spontaneity guidance cognitivereappraisal the feel-intensity of the negative emotion can bedrawn from formula (2)

120583 = 119870119898 log 119868 + 119862 (1)

1205831015840= 120594 (119870119898 log 119868 + 119862) (2)

Here 120583 is the feel-intensity before spontaneity cognitivereappraisal 1205831015840 is the feel-intensity after spontaneity cognitivereappraisal 119868 is the stimulus-intensity and 119870119898 and 119862 areconstant

When 120594 rarr 0 robot has better endurance and enhancesthe defense capability of external negative emotional stimuli

Discrete Dynamics in Nature and Society 3

InterestedAngrySober Sleepy

ControlledFriendly Contemptuous

Clam ExcitedDominant Compliant

Painful GladInterested Relaxed

Humble ArrogantExcited Wrathful

Stiff SurprisedInfluential Influenced

Predominant

0 1 2 3 41234

Valence

Arousal

Stance

Figure 2 A simplified version of the affect scale

Conversely when 120594 rarr 1 robotrsquos endurance declines and itsown emotion state is vulnerable to external impacts

Based on AVS emotional model a simplified version ofthe affect scale is designed In this scale each dimensionality(arousal valence and stance) is measured by four pairs ofemotional states as is shown in Figure 2 Each pair includestwo opposite emotional states divided into 9 levels in adimension and the two states are fundamentally the samein the other two dimensions There are 7 typical experimentscenes corresponding to 7 external stimulus emotions and500 volunteers of random selection in different ages includingfive groups of 11sim20 21sim30 31sim40 41sim50 and 51sim60 yearsGroups of 100 participants get external stimulus emotionand choose emotional level in the scale at once and afterthinking a while respectively By processing and analysis ofeach scenersquos data with the hypothesis test based on Gaussianiteration we can find that after the cognitive reappraisal theinfluence of positive scenes has no obvious change and thechange of negative emotion obeys Gauss distribution shownin Figure 3 In this hypothesis test we first calculate the Gaussmathematical expectation 120583 for each scene by maximumlikelihood estimation Calculate the distance between eachpiece of data and the center of Gauss distribution and deletethe date which has the max distance Then calculate the newmathematical expectation 1205831015840 by maximum likelihood andobtain the distance |1205831015840 minus120583| If the distance |1205831015840 minus120583| is less thanthe threshold the iteration will finish If the algorithm con-verges quickly the AVS data for this scene is subject to Gaus-sian distribution So the parameter 120594 a set of Gauss randomdata could imitate human spontaneity cognition and makerobot have a probabilistic and diversified cognitive style

3 Emotional Regulation Modeling

31 Spatial Description of Emotional State The state spaceof robotrsquos emotions 119878 where emotions can transform freelyis a set of robotrsquos all emotions Robotrsquos own emotional state119878119905can be considered as a spatial location at time 119905 in the

active field state space And each state has the correspondingpsychology energy which is determined by the potentialenergy in the active field state space Different to traditionallimited emotional states emotional regulation process isdefined in the case of continuous time and continuous state

ArousalValence

Stan

ce 01234

minus1minus2minus3minus4

0123

minus1minus2

minus3minus4 minus4 minus3 minus2 minus1 0 1 2 3

Figure 3 Two kinds of negative emotional distribution after thecognitive reappraisal

space for making the robot more lively and anthropopathicSo 119905rsquos value range is a nonnegative real number and 119878

119905is

a real number point Considering the influence of inputmicroexpressions from participants the stimulus emotionalspace 119882 = 119908

1 1199082 119908

119898 is introduced in the robot

emotional regulation Here 119898 is the number of emotionalstates contained within the microexpressions On the basisof Ekmanrsquos emotion theory calming is included in theemotional states space to make the robot more similar tohuman So the stimulus emotional state space is

119882 = anger disgust fear happiness

sadness surprise calming (3)

According to the microexpression recognition method in[24] the facial microexpression of participants in the com-munication is mapped into 7 prototypical emotions in orderto simplify the experiment

32 Emotion Modeling in Active Field

321 Emotional States Space in Active Field Dynamical psy-chology shows that human psychology also requires energy(namely psychological energy) as other physical dynamicalsystems According to the reactions to external excitations[25ndash28] in this research the facial emotion type and theaction range correspond to the position of energy source and


the energy size respectively Based on Kismetrsquos emotionalspace the concept of the field is introduced into the emotionalstate space for describing emotional spatiotemporal propertyand measuring energy change among emotions To this endwe need consider the following three problems First whichemotional state is activated in the field Second what isthe source of the activated emotions Third what is thefield distribution around them In our emotional model theinteraction between external stimulus and robotrsquos emotionin the active field forms the emotional state space Here thesize of field source is determined by the activated intensity ofemotional state and position of field source is determined byemotional category Field intensity distribution in the emo-tional state space is determined by the emotional state systemwhich is composed of stimulus states and robotrsquos currentemotions From the field theory the activated intensity

119872

at any point119872(119909 119910 119911) in emotional state space is

119872=

119899

sum

119894=1

119876119894

1205831199032119894

119903119900

119894 (4)

where 119894 is the number of emotional sources 119903119894is the distance

from emotional source 119894 to point119872(119909 119910 119911) 1199030119894is 119903119894rsquos direction

vector 120583 is a coefficient and 119876119894is the intensity of emotional

source 119894The emotional vector field corresponds to the scalar

potential field The emotional activated intensity describesthe field in the space from the force perspective Accordinglyrobot emotional state transforms from current to next viathe work done by the active field On the other handthe emotional potential 120576 describes the field by energy sothe emotional potential energy is represented by 120576rsquos valuewhich is only determined by field sources 120576rsquos value is equalto in numeral emotional potential energy of the unitageThe computing method about emotional potential energy119872(119909 119910 119911) having 119899 activated emotional states is (Figure 4)

120576119872(119909 119910 119911) =

119899

sum

119894=1

119876119894

120583119903119894

=

119899

sum

119894=1

119876119894

120583radic(119909 minus 119909119894)2

+ (119910 minus 119910119894)2

+ (119911 minus 119911119894)2

(5)

322 State Transition Probability for Emotions Individualemotional state is driven and produced by psychologicalenergy In the active field state space the family of nextemotional states is chosen by the potential energy generatedby the stimulus and robotrsquos current emotion The greater theemotional potential energy the point possesses the more theprobability of this potential surface the next emotional statehas Emotional activation threshold could effectively solve theproblem of emotional over sensitivity and overflow Whenthe emotional potential energy is in a certain interval [119886 119887]this emotional state might be activated And in other casesemotions do not have activated probability

Open stance

Positive valenceHigh arousal

Low arousal

Closed stance

Negative valenceEquipotential surface of

emotional source

Happy

Calming

Sadness

Fear

Anger

SurpriseDisgust

Current emotion

Stimulus emotion after

cognitive reappraisal

Guidance emotion

Any point

Figure 4The current and stimulus emotions influence any point inactive field state space

In [119886 119887] the sum of emotional potential energy for eachpoint along the field direction is

120576sum = int119887

119886

120576 (119909 119910 119911) 119889119904 (6)

The transition probability from the current emotionalfamily 119894 to the next 119895 is

119875119894119895=120576119895

120576sum (7)

33 Emotional Regulation Based on HMM Human emo-tional regulation can be divided into two steps the firststep is the basis of cognitive reappraisal and the secondis correlated with personality factor So this paper regardsemotional regulation process as a double stochastic processand the first one could not be directly observed In otherwords the emotional regulation process can be imitated bya hidden Markov model (HMM) as shown in Figure 5

In the hidden process 120576119905is the psychological energy of

emotional family 1198781015840119905at time 119905 and 119905 is a nonnegative real

number After the external stimulus emotional regulationcan be considered continuous in time and space and itsatisfies the following criteria (1)When the external stimulusoccurs Π is initial probability and it is for final state in thelast regulation process to decide (2) For any 119904

1le 1199051le 1199042le

1199052sdot sdot sdot le 119904

119899le 119905119899 random variables 120576

1199051

minus1205761199041

1205761199052

minus1205761199042

120576119905119899

minus120576119904119899

are independent for each other (3) For any 119904 lt 119905 randomvariable 120576

119905minus 120576119904obeys the probability distribution of formula

(7) Here emotional state is 119895 at time 119904 and emotional state is119895 at time 119895 (4) Emotional regulation process is continuousnamely 119905 rarr 120576

119905is a continuous function of time 119905 Emotional

states containing with equal psychology energy have thesame transition probability and they are perceived as anemotional family In fact this Markov process expounds thetransformation among the emotional families


Table 1 26 possible directions of emotional sates

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

120579 0120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

4

120593 0 0 0 0 0 0 0 0120587

2

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4minus120587

2minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4

Raw data from camera

Abstraction clusteringand emotion decision

HMM linking observational state to underlying emotional families

Robot facial expressions continually change during

interactionHMM yields an

observational state froman emotional family

Figure 5 Emotional regulation based on HMM

Another stochastic process is used for outputting aspecific emotional state and this state can be expressed byexpression language behavior and so forth However thesesymbols all are emotional extrinsic manifestations and theemotional state is the motivation to them We have to bothconsider the emotional diversity and the system real-time sothere are 26 possibilities in each emotional family (namely26 emotional states) as shown in Table 1 The angle with thehigh arousal axis is 120579 and minus120587 le 120579 le 120587 The angle with thearousal-valence plane is 120593 and minus1205872 le 120593 le 1205872 And theangle between possible direction ] and the linkage betweenrobotrsquos current emotion and stimulus state is defined as 120590]Emotional state transition probability 1198751015840] is decided by theangle 120590] When the external stimulus is more than one angle120590] to different stimulus may be equal In this case the nextemotion will randomly select among them

1198751015840

] =120587 minus 120590]

119899120587 minus sum119899

]=1 120590] 119899 = 26 (8)

4 Emotional Robot

The robot mechanical structure and shape imitate cartoonAs shown in Figure 6 there are 10 degrees of freedomin the head and arms and more than 100 kinds of facialexpressions So the robot could express emotions with armshead and even facial expressions The robotrsquos upper-bodymotion mechanism shown in Figure 7 is made up of 10 DCmotors and some adapting piece In the interaction robot

Figure 6 Robot and its representative expressions

emotion will trigger a series of external performances likefacial expressions head gestures arm actions and so onThebehavior expression rule is as follows

IF emotional state THEN robot performances

Here the type of performances is closely related to theemotional family and there are some slight differences inthe specific emotional state The range is proportional to theemotional intensity (namely the emotion potential energy)So the robot vivid behavior is a hybrid between the emotionaltype and active degree and could realistically express therobot internal emotional changing

5 Experiment

51 Emotional Cognitive and Psychological Energy As dis-cussed in Section 1 the discrete model for emotional classi-fication and regulation is vital limitations So for expandingthe robot emotional output information to continuous spacewe use one of the most influential AVS 3D models inthe field of affective computing domain that proposed byBreazeal [20] Based on Gross reappraisal strategy explainedin Section 21 we describe the cognitive emotionalmodel andthis methodology in the AVS space The emotional coordi-nates were calculated on the basis of volunteersrsquo responsesto the emotional state along three emotional dimensions ofarousal valence and stance And all emotional dimensionsare bounded within a range of [minus10 10] and the ldquocalmingrdquostate is placed on the original point (0 0 0)

Figure 8 graphically shows an example of psychologicalenergy distribution in the active field state space where boththe current emotional state and the stimulus emotional stateafter cognitive reappraisal as psychological energy resourcesare trying to influence emotional regulation process At themoment the coordinate values of robotrsquos own emotional stateare (5 5 0) and external stimulus derived from microex-pression is sadness whose coordinate value is (minus6 minus4 0)Robot physiological endurance coefficient 120594 = 06 Ascan be seen from Figure 8 the emotional energy gradually


Figure 7 Robot mechanical structure

0

140120100

80604020

105

0minus5

minus10 minus10minus5

05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc

Figure 8 Psychological energy distribution caused by cognitivestimulus and robotrsquos current emotional state

dies down along with the increase of distance between theemotional state and the emotional source And psychologicalenergy rapidly declines round the emotional sources Sowe set a pair of threshold to limit emotional activationscope If the energy of emotional states is greater or lessthan the threshold these emotional states are impossible tohappenThis phenomenon is entirely consistent with classicalemotional theory in the field of psychology [29] Based on therelative positions of the next emotional state and emotionalsource that the emotional family contains higher energywill has greater transition probability than the emotionalfamily which contains lower energy From this the transitionprobability in the first stochastic process of HMM can befigured out

52 Emotional Regulation Process According to Section 33the distribution of emotional familyrsquos probability is calcu-lated under a pair of emotions robotrsquos cognitive stimulusemotional state and its own current emotional state andthen the output of robot emotional state has 26 kinds ofpossibility on the basis of the spatial relationship among thecurrent emotion stimulus emotion and next emotion Thismethod highlights the capability to find a large amount ofintermediate emotional states which are extremely vital since

2 3 4 5 6 7 8 9 101

2

3

4

5

6

7

Emotional state at present

CalmingHappinessSurprise

FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)

Figure 9 Example of emotional familyrsquos probability

they enrich the output of the robot emotional regulationsystem In Figure 9 external stimulus emotional state comingfrom microexpression is ldquosadnessrdquo and we can observe thetransition probabilityrsquos microvariation of emotional familieswhere 6 prototypical emotions (except sadness because thepsychological energy of sadness exceeds the threshold valueand it will not happen) are with robotrsquos own emotional statechanges Here robotrsquos own emotion is located at any pointarousal = valence isin [2 10] and stance = 0

Figure 10 shows the robot emotional regulation pro-cess with calming initial emotional state In 0ndash15 s robotrsquosown emotion remained about the same under no externalstimulus At the 15 s an external stimulus ldquodisgustrdquo derivedfrom microexpression occurred so robot emotionrsquos negativedegree gradually increased during 15ndash35 s and achieved thebalance around the 35 s Then this emotional experience waswith the robot for about 15 s Because the ldquodisgustrdquo stimulushas disappeared a while robot emotional state graduallytrended to ldquocalmingrdquo during 50ndash65 s [30] At the 65 s an


0ndash15 s 20ndash35 s15ndash20 s 35ndash50 s 50ndash65 s 80ndash95 s 95ndash110 s65ndash80 s

Figure 10 Robot emotional regulation process with different external stimulus

external stimulus ldquohappinessrdquo derived from microexpres-sion occurred so robot emotionrsquos positive degree graduallyincreased during 65ndash80 s and achieved the balance aroundthe 80 s Then similar emotional experience was with therobot for about 15 s and gradually waned during 95 s

Filtered 34 volunteers participate in the human-robotinteractions and each person experiments 100-time interac-tion in accordance with specified criteria Each participantfills in the predictive scale before the experiment started Inthis scale participants forecast robotrsquos next output state bythe simplified affect scale (as Figure 2) including 7 typicalstimulus states and 7 typical own emotional sates (a totalof 7 times 7 kinds of possible typical inputs) The participantfills in the satisfaction survey for each interaction duringthe interaction This survey designs two options (agreementand disagreement) for each interaction From the evaluationresults if they reason out robotrsquos emotional output in advancethe average matching rate is 6975 after experience How-ever it rises to 9752 when only considering participantsrsquoagreement to robot output emotions Objectively speakingthe use of the emotion model based on the cognitive reap-praisal in active field allows robot to imitate the hominineemotional regulation and that is in fact the aim of our workBut the obtained results are difficult to compare with otheremotional regulation studies that can be found in literaturebecause most of such studies do not recognize stimulus emo-tions in microexpressions and transfer emotional states inarousal-valence-stance terms Moreover the few studies thatdo have not been tested under physical robot experimentalconditions (specific robot device and experimental platformrefer to [30 31]) and do not provide as output the coordinatesof the studied emotional state in the 3D space

6 Conclusion

In this paper the noteworthy feature of emotional regulationwork was out of the simply interactive mode providingthe classification and jump in terms of a set of emotionallabels and it operated in a 3D emotional space enabling awide range of intermediary emotional states obtained underthe external stimulus Moreover this system focused onthe research field of emotional regulation depending onnatural facial expression cognition and proposed a microex-pression cognition and emotional regulation model basedon Gross reappraisal strategy Gross cognitive reappraisalstrategy effectively decreased negative emotional experienceand behavioral expression so it could provide an intelligentcognition style to computerrobot acting as a positive role in

HCI HMM double stochastic process makes robot emotionshave more diversification in human-robot interaction Ingeneral the use of HMM emotional regulation model basedon cognitive reappraisal in active field allows robot to imitatethe hominine emotional regulation naturally

Our current research only proposed a computable emo-tion model applied to universal psychological significancein continuous space but not considered with specific emo-tion changes Following from this future works should beoriented to the study of nature inspired cognitive-affectivecomputing by means of emotion modeling in continuousactive space and especially need pay more attention to themultimodal external stimulus and the pervasive emotioncomputing

Conflict of Interests

The authors declare that there is no conflict of interestsregarding the publication of this paper

Acknowledgments

This work is supported by National Natural Science Foun-dation of China (no 61432004 61170115) National KeyTechnologies RampD Program of China (no 2014BAF08B04)and the Foundation of Beijing Engineering and TechnologyCenter for Convergence Networks and Ubiquitous Services

References

[1] S S Tompkins Affect Imagery Consciousness Volume 1 ThePositive Affects Springer London UK 1962

[2] C IzardTheFace of Emotion vol 23 Appleton-Century-CroftsNew York NY USA 1971

[3] G Valenza A Lanata and E P Scilingo ldquoThe role of nonlineardynamics in affective valence and arousal recognitionrdquo IEEETransactions on Affective Computing vol 3 no 2 pp 237ndash2492012

[4] R S Lazarus ldquoRelational meaning and discrete emotionsrdquo inAppraisal Processes in Emotion Theory Methods Research pp37ndash67 Oxford University Press New York NY USA 2001

[5] P Ekman Lie Catching and Microexpressions Oxford UniversitPress 2009

[6] L Canamero ldquoModeling motivations and emotions as a basisfor intelligent behaviorrdquo in Proceedings of the 1st internationalConference on Autonomous Agents (AGENTS rsquo97) pp 148ndash1551997


[7] S Gadanho ldquoReinforcement learning in autonomous robots anempirical investigation of the role of emotionsrdquo in Emotions inHuman and Artifacts MIT Press 2002

[8] J Velasquez ldquoAn emotion-based approach to roboticsrdquo in Pro-ceedings of the IEEERSJ International Conference on IntelligentRobots and Systems (IROS rsquo99) vol 1 pp 235ndash240 KyongjuRepublic of Korea October 1999

[9] R R Murphy C L Lisetti R Tardif L Irish and A GageldquoEmotion-based control of cooperating heterogeneous mobilerobotsrdquo IEEE Transactions on Robotics and Automation vol 18no 5 pp 744ndash757 2002

[10] W Burgstaller R Lang P Porscht and R Velik ldquoTechnicalmodel for basic and complex emotionsrdquo in Proceedings of the5th IEEE International Conference on Industrial Informatics pp1007ndash1012 2007

[11] W Wundt Principles of Physiological Psychology MacmillanPress New York NY USA 1910

[12] H Schlosberg ldquoThree dimensions of emotionrdquo PsychologicalReview vol 61 no 2 pp 81ndash88 1954

[13] P J Lang M M Bradley and B N Cuthbert ldquoEmotion moti-vation and anxiety brain mechanisms and psychophysiologyrdquoBiological Psychiatry vol 44 no 12 pp 1248ndash1263 1998

[14] C OsgoodTheMeasurement of Meaning University of IllinoisPress 1975

[15] J Panksepp Affective Neuroscience The Foundations of Humanand Animal Emotions Oxford University Press New York NYUSA 2004

[16] J A Russell and A Mehrabian ldquoEvidence for a three-factortheory of emotionsrdquo Journal of Research in Personality vol 11no 3 pp 273ndash294 1977

[17] K Scherer and P Ekam Approaches to Emotions LawrenceErlbaum Associates 1984

[18] A Ortony G L Clore and A Collins The Cognitive Structureof Emotions Cambridge University Press London UK 1988

[19] M Zecca S Roccella M C Carrozza et al ldquoOn the devel-opment of the emotion expression humanoid robot WE-4RIIwith RCH-1rdquo in Proceedings of the 4th IEEE-RAS InternationalConference on Humanoid Robots pp 235ndash252 Tokyo JapanNovember 2004

[20] C Breazeal ldquoFunction meets style insights from emotiontheory applied to HRIrdquo IEEE Transactions on Systems Man andCybernetics Part C Applications and Reviews vol 34 no 2 pp187ndash194 2004

[21] H Yang Z Pan and G Liu ldquoComprehensive computationalmodel of emotionsrdquo Journal of Computer Research and Devel-opment vol 45 no 4 pp 579ndash587 2008

[22] W H Kim J W Park W H Lee and M J Chung ldquoStochasticapproach on a simplified OCC model for uncertainty andbelievabilityrdquo in Proceedings of the IEEE International Sympo-sium on Computational Intelligence in Robotics and Automation(CIRA rsquo09) pp 66ndash71 Daejeon Republic of Korea December2009

[23] J J Gross ldquoEmotion regulation affective cognitive and socialconsequencesrdquo Psychophysiology vol 39 no 3 pp 281ndash2912002

[24] S-J Wang H-L Chen W-J Yan Y-H Chen and X FuldquoFace recognition and micro-expression recognition based ondiscriminant tensor subspace analysis plus extreme learningmachinerdquo Neural Processing Letters vol 39 no 1 pp 25ndash432014

[25] J Zhang X Wang and H Xie ldquoPhonon energy inversion ingraphene during transient thermal transportrdquo Physics LettersA vol 377 no 9 pp 721ndash726 2013

[26] Q Xiong B Li J Xu X Wang L Wang and W Ge ldquoEfficient3D DNS of gas-solid flows on Fermi GPGPUrdquo Computers andFluids vol 70 pp 86ndash94 2012

[27] Q Xiong E Madadi-Kandjani and G Lorenzini ldquoA LBM-DEM solver for fast discrete particle simulation of particle-fluidflowsrdquo ContinuumMechanics andThermodynamics vol 26 no6 pp 907ndash917 2014

[28] J Zhang Y Wang and X Wang ldquoRough contact is not alwaysbad for interfacial energy couplingrdquoNanoscale vol 5 no 23 pp11598ndash11603 2013

[29] M A Salichs and M Malfaz ldquoA new approach to modelingemotions and their use on a decision-making system forartificial agentsrdquo IEEE Transactions on Affective Computing vol3 no 1 pp 56ndash68 2012

[30] L Xin X LunW Zhi-Liang and F Dong-Mei ldquoRobot emotionand performance regulation based on HMMrdquo InternationalJournal of Advanced Robotic Systems vol 10 article 160 2013

[31] P Xiaolan X Lun L Xin and W Zhiliang ldquoEmotional statetransition model based on stimulus and personality character-isticsrdquo China Communications vol 10 no 6 pp 146ndash155 2013

Submit your manuscripts athttpwwwhindawicom

Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

MathematicsJournal of


Mathematical Problems in Engineering

Hindawi Publishing Corporationhttpwwwhindawicom

Differential EquationsInternational Journal of

Volume 2014

Applied MathematicsJournal of


Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Journal of


Mathematical PhysicsAdvances in

Complex AnalysisJournal of


OptimizationJournal of


CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of


Operations ResearchAdvances in

Journal of


Function Spaces

Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014

International Journal of Mathematics and Mathematical Sciences


The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014


Algebra

Discrete Dynamics in Nature and Society



Decision SciencesAdvances in

Discrete MathematicsJournal of


Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014

Stochastic AnalysisInternational Journal of


12 The Emotional Space Model Unlike discrete modelsthe emotional space models consider a continuous multi-dimensional space where each point stands for an emo-tional state and each dimension stands for a fundamen-tal property common to all emotions In order to fullyshow the emotional properties the classical 3D emotionalspace pleasantunpleasant excitementdepression and ten-sionrelaxation has already been used by Wundt [11] Formany years a large number of emotional dimensional the-ories have been proposed [12ndash15] One of the most acceptedtheories is described by Russell andMehrabian [16] pleasure-arousal-dominance space (PAD) Scherer and Ekam andOrtony et al also use and develop PAD space to deter-mine artificial emotions for social robots [17 18] Zecca etal construct the 3D psychological vector space arousal-pleasant-certain withmachine learning dynamic regulationand personality [19] In addition Breazealrsquos arousal-valence-stance (AVS) space with the social robot Kismet is the mostnoteworthy emotional space model [20] In that case theemotional state of the robot varies according to its interactionwith people In fact the external stimulus is labeled by threeemotional properties in the AVS space Ortony et al modelthe OCC reasoning process that can produce the cognitiveemotions and touch off complex emotional experiences viathe trend of events (including event object and agent) [18]Then this model has been developed and refined by Yang etal and Kim et al [21 22]

This paper discussed a cognitive emotional regulationmodel in the active field state space First based on Grosscognitive strategy the cognitive reappraisal model satisfiedFischna-Weber law was built and a physiological endurancecoefficient was defined to meet the psychological diver-sity Second emotional state space in the active field wasestablished on the solid basis of the physiological energydistribution and the transition probability among the emo-tional families was figured out Third an observational statesequence was yielded from emotional family by HMMFinally the emotional regulationmodel is devoted to human-robot experiment for the validity analysis Figure 1 gives outgeneral ideas of this research

The rest of this paper is organized as follows Section 2discusses the Gross cognitive strategy and presents a cogni-tive reappraisal model Section 3 defines the emotional spaceand related state transition process Section 4 introducesthe emotional robot platform Section 5 shows experimentalresults and discussions The conclusion and directions forfuture work are offered in Section 6

2 Emotional Cognition and Modeling

21 Gross Cognitive Strategy Gross proposed five emotionalregulation strategiesmdashsituation selection situation modi-fication attention deployment cognitive reappraisal andresponse suppression [23] Cognitive reappraisal the fore-most antecedent-focused strategy in the early emotionalregulation stage depends on the internal willpower andpersonality factorsWhen the individual is in troubles or psy-chological expectation is different to the reality the negative

External stimulus emotion from

expression

Negative emotion

Cognitive reappraisal

Source distribution in emotional sates space

Emotional states transition probability

Next emotion

Yes

No

Current emotion

Robot expression

Cognitive emotional regulation model

Figure 1 General ideas of this research

emotions such as sadness anxiety anger and pain areproduced In general the cognitive reappraisal can well cutdown the negative emotional intensity and improve theindividual emotional experience In the cognitive processcognitive reappraisal and response suppression strategiesboth can improve the individual emotional experience How-ever the psychological research has proven that the cognitivereappraisal strategy as compared with the four other regu-lation strategies is most effective in the negative emotionalregulation

22 Cognitive Reappraisal Model Gross believes that spon-taneity cognitive reappraisal as an essential part of psy-chological defense mechanism could help individuals toconsider the emotional stimulus from the peaceful perspec-tive [23] Thus facing various external emotional stimuluspsychological systemwill first unconsciously start the sponta-neous cognitive reappraisal to reduce the emotional intensityIn other words when the external stimulus appears inemotional state space the spontaneous cognitive reappraisalcomes into play at once for cutting down the stimulus-intensity Because the spontaneous cognitive reappraisalmainly depends on individual physiological endurance aphysiological endurance coefficient 120594 isin (0 1) is defined forrobot

Fischna-Weber law is a psychophysical formula describ-ing the relationship of internal feel-intensity and externalstimulus-intensity The law shows that human feel-intensityis proportional to the logarithm of stimulus-intensity (shownin formula (1)) Under the spontaneity guidance cognitivereappraisal the feel-intensity of the negative emotion can bedrawn from formula (2)

120583 = 119870119898 log 119868 + 119862 (1)

1205831015840= 120594 (119870119898 log 119868 + 119862) (2)

Here 120583 is the feel-intensity before spontaneity cognitivereappraisal 1205831015840 is the feel-intensity after spontaneity cognitivereappraisal 119868 is the stimulus-intensity and 119870119898 and 119862 areconstant

When 120594 rarr 0 robot has better endurance and enhancesthe defense capability of external negative emotional stimuli








Predominant

0 1 2 3 41234

Valence

Arousal

Stance







ArousalValence

Stan

ce 01234


0123

minus1minus2




119905is


1 1199082 119908










119872


119872=

119899

sum

119894=1

119876119894

1205831199032119894

119903119900

119894 (4)






120576119872(119909 119910 119911) =

119899

sum

119894=1

119876119894

120583119903119894

=

119899

sum

119894=1

119876119894

120583radic(119909 minus 119909119894)2

+ (119910 minus 119910119894)2

+ (119911 minus 119911119894)2

(5)


Open stance


Low arousal

Closed stance


emotional source

Happy

Calming

Sadness

Fear

Anger

SurpriseDisgust

Current emotion



Guidance emotion

Any point



120576sum = int119887

119886

120576 (119909 119910 119911) 119889119904 (6)


119875119894119895=120576119895

120576sum (7)





1le 1199051le 1199042le



1199051

minus1205761199041

1205761199052

minus1205761199042

120576119905119899

minus120576119904119899








1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

120579 0120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

4

120593 0 0 0 0 0 0 0 0120587

2

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4minus120587

2minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4









1198751015840

] =120587 minus 120590]

119899120587 minus sum119899

]=1 120590] 119899 = 26 (8)

4 Emotional Robot






5 Experiment





0

140120100

80604020

105

0minus5


05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc




2 3 4 5 6 7 8 9 101

2

3

4

5

6

7



FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)









6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Predominant

0 1 2 3 41234

Valence

Arousal

Stance







ArousalValence

Stan

ce 01234


0123

minus1minus2




119905is


1 1199082 119908










119872


119872=

119899

sum

119894=1

119876119894

1205831199032119894

119903119900

119894 (4)






120576119872(119909 119910 119911) =

119899

sum

119894=1

119876119894

120583119903119894

=

119899

sum

119894=1

119876119894

120583radic(119909 minus 119909119894)2

+ (119910 minus 119910119894)2

+ (119911 minus 119911119894)2

(5)


Open stance


Low arousal

Closed stance


emotional source

Happy

Calming

Sadness

Fear

Anger

SurpriseDisgust

Current emotion



Guidance emotion

Any point



120576sum = int119887

119886

120576 (119909 119910 119911) 119889119904 (6)


119875119894119895=120576119895

120576sum (7)





1le 1199051le 1199042le



1199051

minus1205761199041

1205761199052

minus1205761199042

120576119905119899

minus120576119904119899








1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

120579 0120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

4

120593 0 0 0 0 0 0 0 0120587

2

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4minus120587

2minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4









1198751015840

] =120587 minus 120590]

119899120587 minus sum119899

]=1 120590] 119899 = 26 (8)

4 Emotional Robot






5 Experiment





0

140120100

80604020

105

0minus5


05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc




2 3 4 5 6 7 8 9 101

2

3

4

5

6

7



FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)









6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











119872


119872=

119899

sum

119894=1

119876119894

1205831199032119894

119903119900

119894 (4)






120576119872(119909 119910 119911) =

119899

sum

119894=1

119876119894

120583119903119894

=

119899

sum

119894=1

119876119894

120583radic(119909 minus 119909119894)2

+ (119910 minus 119910119894)2

+ (119911 minus 119911119894)2

(5)


Open stance


Low arousal

Closed stance


emotional source

Happy

Calming

Sadness

Fear

Anger

SurpriseDisgust

Current emotion



Guidance emotion

Any point



120576sum = int119887

119886

120576 (119909 119910 119911) 119889119904 (6)


119875119894119895=120576119895

120576sum (7)





1le 1199051le 1199042le



1199051

minus1205761199041

1205761199052

minus1205761199042

120576119905119899

minus120576119904119899








1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

120579 0120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

4

120593 0 0 0 0 0 0 0 0120587

2

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4minus120587

2minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4









1198751015840

] =120587 minus 120590]

119899120587 minus sum119899

]=1 120590] 119899 = 26 (8)

4 Emotional Robot






5 Experiment





0

140120100

80604020

105

0minus5


05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc




2 3 4 5 6 7 8 9 101

2

3

4

5

6

7



FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)









6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26

120579 0120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

40 0

120587

4

120587

2

3120587

4120587 minus

120587

4minus120587

2minus3120587

4

120593 0 0 0 0 0 0 0 0120587

2

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4

120587

4minus120587

2minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4minus120587

4









1198751015840

] =120587 minus 120590]

119899120587 minus sum119899

]=1 120590] 119899 = 26 (8)

4 Emotional Robot






5 Experiment





0

140120100

80604020

105

0minus5


05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc




2 3 4 5 6 7 8 9 101

2

3

4

5

6

7



FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)









6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra











0

140120100

80604020

105

0minus5


05

10

Psyc

holo

gica

l ene

rgy

ValenceArousal50

5

al

50

minus5Valenc




2 3 4 5 6 7 8 9 101

2

3

4

5

6

7



FearAngerDisgust

Tran

sitio

n pr

obab

ility

(times10minus2)









6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra














6 Conclusion






Acknowledgments


References








































Volume 2014




Journal of











Journal of


Function Spaces






Algebra










































Volume 2014




Journal of











Journal of


Function Spaces






Algebra
















Volume 2014




Journal of











Journal of


Function Spaces






Algebra









Documents

Research Article Cognitive Emotional Regulation Model in ...downloads.hindawi.com/journals/ddns/2015/829387.pdf · emotions. From the eld theory, the activated intensity at any point