Transcript
Page 1: [IEEE 2005 International Conference on Neural Networks and Brain - Beijing, China (13-15 Oct. 2005)] 2005 International Conference on Neural Networks and Brain - Study on Artificial

Study on Artificial Emotion ModelNing Shurong

Academy ofInformation andScience Technology

Beijing Institute of TechnologyBeijing, 100081,China

E-mail: [email protected]

Pang JieSchool of Information

Beijing Forestry UniversityBeijing, 100083, China

E-mail: [email protected]

Xu Guangmei, Zeng Guangping, Tu XuyanAcademy ofInformation EngineeringUniversity of Science and Technology

Beijing, 100083, ChinaE-mail: [email protected]

Abstract-Artificial emotion is a companion volume ofartificial intelligence. Emotion is key to human action ofdecision-making. Emotion, intelligence and behavior haveintimate relations. Emotion model is a vital question to theresearch of artificial emotion. This paper introducesGeneralized model of artificial emotion and emotional controlmodel in softman control system. Erf (Emotion restrainfactor) and Eef (Emotion excursion factor) are brought

forward. Erf , Eef and V,, are the three elements ofemotional control in softman. Results of simulation prove thatinfluence of Eef to emotional control is stronger than that of

Erf . It matches the natural behaviors ofhuman emotion.

I. INTRODUCTION

For a very long time, it is said that emotion has littlerelationship with intelligence. However, the research resultof neuroscience shows that emotion plays an important rolein human's action of reasoning and decision-making. That isto say that the emotion plays important and indispensablerole in intelligence. The other research shows that humanwill not make decision effectively if human's subsystem ofemotion is injured or dysfunctional. With more and morepeople realize the intelligent function of emotion, itbecomes a hot topic that how to apply the intelligentfunction of emotion into computer and how to makecomputer have intelligence.

Artificial emotion is a new subject and will makemachine have emotion. Artificial emotion is the companionvolume of artificial intelligence. How to make computer ormachine have emotion to promote the intercommunionbetween them inevitably becomes a new requirementModel is very important to artificial emotion. This paper

researches emotion model. Generalized model of artificialemotion and emotional control model in softman controlsystems are studied. Erf (Emotion restrain factor), Eef(Emotion excursion factor) and V are the three factors ofemotional control in softman. Results of simulation provethat influence of Eef to emotional control is stronger thanthat of Erf. It matches the natural behaviors of human

emotion.

HI. THE MODEL OF ARTIFICIAL EMOTION

A. The Relation among Emotion, Intelligence andBehavior

The researches of intelligent control system based onartificial emotion are mainly divided as two parts: emotionrecognition, emotion control. The main task of emotioncontrol is to construct an emotion model. Emotion control ismade up of emotional control model and control emotionmodel. From the point of main body of control, the action ofcontrol emotion is active, but the action of emotional controlis passive [5-10,7-8]

The relation of emotion, intelligence and behavior hasseveral types as follows:

Fig. 1: The relation ofemotion, intellect and behavior

Intelligence - behavior: The behavior controlled byintelligence

Emotion - behavior: The behavior controlled byemotion

Emotion - Intelligence - behavior: The behaviorcontrolled by intelligence andemotion

"'-": Represents the action of control.Usually, human behavior is under the control the

intelligence. In an emergency, it is not intelligence butemotion that controls human behavior. There are manyexamples like that emotion controls human behavior.

Generally, human choice and behavior are affected byemotion. The controlling action of intelligence is inferior to

0-7803-9422-4/05/$20.00 C2005 IEEE1420

Page 2: [IEEE 2005 International Conference on Neural Networks and Brain - Beijing, China (13-15 Oct. 2005)] 2005 International Conference on Neural Networks and Brain - Study on Artificial

that of emotion. Many scientists have proved this result.The clamness expressed by the captain of TITANIC

before the tremendous tragedy is the pretty example thatemotion controls human behavior.

B. The Generalized Model ofArtificial EmotionThe definition of the model ofAECM (Artificial Emotion

Control Model) is shown as follows:AECM= {I,M,S,O,P} (1)

During above equation (1):

(D) I -- Initialized EmotionThis is a binary knowledge model describing the

initialized emotion in the artificial emotion controlsystemI = I [IES, IEM] (2)

IES -- Initialized Emotion StateIEM -- Initialized Emotion Measurement

) M -- Emotion Control ModelM=M[ECM, CEM] (3)ECM-- Emotional Control ModelCEM -- Control Emotion Model

( S -- Emotion Control System Structure() 0-- Objective Emotion

O=O[OES, OEM] (4)OES -- Objective Emotion StateOEM-- Objective Emotion Measurement

© P -- Emotion Control Process ModelThis is a multiple partial differential state

equation set describing the dynamic process ofemotional activity.

p_= (5)

x

te

-- Emotion State Vector-- Emotion Time Variable

-- Emotion Environment Variable

-- This partial differential equation denote the

rate of change of emotional state complieswith environment

at-- This partial differential equation denote the

rate of change of emotional state complieswith time

Without regard for the influencing factor of theenvironment t imposes on the emotion state x, the model ofartificial emotion control system can be simplified asnonlinear multiple ordinary differential state equation set.

P =P x,,ede

(6)

- -- Ordinary differential coefficient denotes thede

rate of change of emotional state complieswith environment.

Im MODEL OF ARTIFICIAL EMOTION IN SOFTMAN CONTROLSYSTEM

A. Visual EnvironmentThe creature in nature has ability to self-adaptation and

self-learning. It is a very important factor for this ability thatbody of creature can interact with its environment

The visual environment of sofanan living is introduced.There are four properties in this visual environment. Theyare understanding, friendship, listening and care. The targetof sofanan in this visual environment is to gain the emotionof calmness. Besides of the environmental properties, thereare several properties within self-softman, such as factor ofemotion restrain and factor of emotion excursion. Therelations between softman and properties are described asfollows:The value of property of 'understand' increases when

softman is understood. Otherwise, the value decreases.The value of property of 'friendship' increases when

softman gains friendship. Otherwise, the value decreasesThe value of property of 'listening' increases when

softman is listened. Otherwise, the value decreasesThe value of property of 'care' increases when sofminan is

cared. Otherwise, the value decreases

B. Artificial Emotion Control ModelThe Schematic Diagram of Softman Emotion Control is

shown as Fig. 2.

Fig. 2. Schematic Diagram of Softman Emotion Control

1421

Page 3: [IEEE 2005 International Conference on Neural Networks and Brain - Beijing, China (13-15 Oct. 2005)] 2005 International Conference on Neural Networks and Brain - Study on Artificial

The feature of each property is described as follows:The correlative symbol of artificial emotion control is

shown as table 1:TABLE I

LISTS ofCORRELATIVE SYMBOL

Symbol Range ofValue Description of Symbol

three decreases. If A is a set of these four properties, thenA={{U}, {F}, {L}, {C}}.4Ais one element of A.V(Ai)isthe value of A., V(A - A) is the values of the other threeelements of A.

Vout (V(A1)) = K3 (V(A) - V(A - Ai))

[0,4][0, 1 ][0, 1 ][0, 1 ]

[-0.25, -0.5, -0.75, -1][0.25,0.5,0.75,1]

[0,1][0, 1 ][ 0 , 1 ][ 0 , 1 ]

intensity of emotionintensity of the currentemotion at the time of toutput emotion of softmandominant emotionemotional restrain factoremotional excursion factorproperty ofunderstandingproperty of friendshipproperty of listeningproperty of care

E(i) t= KjVouE(i)t (I11)

Current emotion is influence by past emotion. The currentoutput emotion of softman is its emotion of next time.

E(i)' = Eout (12)Current emotion is influence by environment of softman,

Ef, Eef .

E(i)t+l = f {EI, E(t), Erf, Eef, Vout}

E(i)t = De = max{E(i)}

(13)

(14)

Softnan has four initialed emotions: rage, sadness,unhappiness and dread. It is convenient for comparison tomap the intensity of these four emotions to the same range[0,1]. Intensity of each emotion is represented by numericalvalue:

0

i i E [0,1]

i-i iE[1,2]

i-2 iE[2,3]

i-3 iE[3,4]

calmnessragesadnessunhappinessdread

(7)

E(i)t is modified as follows to make its range value ofbe [0,1].

E(i)=- min[E(i)', 1] (15)C. Control Rules

Rule 1: At fixed time, one variable of Erf; Eef andVout is dominated in emotional control.Rule 2: The adjusted rules of emotional control:

I) Er(E(i)t ) -+ E(i)+ -< E(i)'2) Ee(E(i)t) -+ E(i)t'+ . E(i)'

3) Ev(E(i)t) -* E(i)t+1 < E(i)t

(16)

(17)

(18)Emotional restrain factor Erf: it is formalized description

of ability to emotion control of self-softman. If this abilityincreases, the Erf 's value of softman increases accordingly.Range value of Erf is [-0.25, -0.5, -0.75, -1]. It

corresponding ability of emotion restrain is {relatively weak;common; relatively strong; very strong}.

E(i) +1= K1E(i)t+Ef (8)

Emotional excursion factor Eef : it is formalizeddescription of degree of emotional attention of self-softnan.Emotional attention describes the degree that softman isabsorbed in one emotion. Emotion excursion is opposite toemotional excursion. With the value of Eef lager, itsinfluence to emotional attention is stronger. Range value ofEef is [0.25,0.5,0.75,1]. It corresponding degree ofemotion excursion is {relatively weak; common; relativelystrong; very strong}

E(i)'t+'= K2(1- Eef)E(i)t (9)U, F, L and C are properties of outer environment of

softman. The range value of each property is [0,1]. Value ofone's property increases at one fixed time, those of the other

Er(E(i)') : Softman begins to restrain its emotion fromtime of t.

Ee(E(i)'): Emotional attention of softman is influencedform time of t.

Ev(E(i)') :Emotion of softman is influenced byenvironment from time of t.

IV SIMULATION

Now we analysis the results of emotional control underthe condition of EI = 0.8 and without regard for influenceof V

out

When VO =1,Eef =O, Erf changes, the result ofsimulation is shown as Fig.3:

1422

EI

E(i)EoutDeErfEefUFLC

EI =

(10)

Page 4: [IEEE 2005 International Conference on Neural Networks and Brain - Beijing, China (13-15 Oct. 2005)] 2005 International Conference on Neural Networks and Brain - Study on Artificial

Fig. 3.Curve of Emotion Control' - ': Erf =0.75;'-*-': Erf =0.5;' - + - ': Erf =0.25When VP" = 1, Erf = 0, Eef changes, the result

simulation is shown as Fig.40.81

of

Fig.4.Curve of Emotion Control'-':Erf =0.25;'- *- ': Erf=0.5;'-+- ':Erf=0.75

From the result of simulation, we can see that emotioncan be controlled by Erf and Eef if the element of V isout

discarded. The influence of Eef to emotional control isstronger than that of Erf . This result matches the naturalbehaviors ofhuman emotion.

Ill. CONCLUSIONS

Research of emotional control model is very important forartificial emotion. This paper discusses Generalized modelof emotional control and the emotional control model insoftman control system.

During the model of emotional control, there are threeelements, Erf , Eef and V which can influence the

out

emotion of softman. If the influence of V is discarded, theout

influence of Eef to emotional control is stronger than thatof Erf . This result matched the natural behaviors ofhumanemotion. The results proved that Erf and Eef really

influence human emotion. Emotion can be controlled byErf and Eef.

Artificial emotion is a new and hot topic. There are manyresults and research conclusions about artificial emotion, butall these results and conclusions focus on emotionrecognition, emotion expression. The research aboutemotional control and intelligent control based on emotionalcontrol is relatively few. This paper is a primary researchabout emotional control model. Emotional control hasexpansive perspective of application[3][4]. Self-prompter andemotion adjuster are concerned with artificial emotion. Theresearch of artificial emotion will make intelligence strideforward into a new stage. But there are many new and hardquestions during the researches of artificial emotion.

ACKNOWLEDGMENT

I thank my tutor, Professor Tu xuyan and my classmatesin the same laboratory. I thank them for their help and theirexcellent suggestions for me

REFERENCES

[1] Tu Xu Yan, "artificial emotion", the advance of Chinese artificialintelligence. Beijing University of Posts and TelecommunicationPress, 2003,pp.27-31

[2] Custodial, Ventura, R., and Pinto-Ferreira, C., "ARTIFICIALEMOTIONS AND EMOTION- BASED CONTROL SYSTEM,"1999IEEE, pp. 1415-1420.

[3] Tu Xuyan and Yin Yixin.ed, "Artificial Life and Applications",proceedings of CAAI-Symposium on Artificial Life and Applications.Oct. 2002. Beijing university of Science and Technology Beijing.

[4] Zeng Guangping, Tu Xuyan, "SoftMan" The 10-th NationalConference of CAAI-Chinese Association of Artificial Intelligence,2003, Beijing, China, pp.677-682.

[5] Wang Zhiliang, "Artificial Psychology", Journal of University ofScience and Technology Beijing, where, 2002,vol.22, pp.478-481

[6] Zheng Xue, Yan Biaobin, Qiu Lin, Zhang Xinggui, EudemoniaPsychology, JiNan University Press, 2004.6.

[7] Jia Guangyu, My Meager Opinion of "emotional intelligence", ChinaEducation of Light Industry, 2005.l,p23

[8] Zhao Nanyuan, The Relationship between Emotion and Intelligence,The l' Chinese Conference on Affective Computing and IntelligentInteraction.2003, p60-63.

[9] Lu Haiming, Li Yanda, Lu Zengxiang, Xia Huiyu, INFOEMAITONRECOMMENDATION BASED ON CONTROL THEORY ANDAFFECTIVE COMPUTING, ACTA AUTOMATICA SINICA,2002, 28(4): 481-487.

[10] LIU Ming, XU Li, Study on Artificial Emotion and It's Application inthe Behavior Selection Strategy of Agents, Journal of SouthernYangtze University (Natural Science Edition), 2003,No.6.

[11] Hidenori Ishihara and Toshio Fukuda, Distributed Control ofMultipleAgents by Emotional Algorithm, Proceedings of the 2002IEEE,P666-671

[12] Tatsuya Nomura, Problems ofArtificial Emotion in Mental Therapy,Proceedings 2003 IEEE International Symposium on ComputationalIntelligence in Robotics and Automation,July 16-20,2003,Kobe,p567-570

[13] Hidenori Ishihara and Toshio Fukuda, individuality of agent withemotional algorithm, Proceedings of the 2001 IEEE/RSJ InternationalConference on Intelligent Robots and Systems. P1195-1200

[14] Kazuya MERA.Shinji KAWAMOTO,Miysuko YAMURA-TAKEI

1423

Page 5: [IEEE 2005 International Conference on Neural Networks and Brain - Beijing, China (13-15 Oct. 2005)] 2005 International Conference on Neural Networks and Brain - Study on Artificial

and Teruaki AIZAWA, Emotion-based planning evaluation method,Fourth International Conference on Knowledge-based EngineeringSystems, 2000 IEEE, P852-855

[15] Luis M.Botelho , Pedro N.Ramos, Emotionally ControlledInter-agent Communication ,Proceedings of the IEEE IntelligentVehicle Symposium 2000,p410-417

[16] Takanori Shibata, Makoto,Junji Yamato, Artificial Emotion Creaturefor Human-Machine Interaction,1997 IEEE, p2269-2274

[17] Takashi Gomi, John Vardalas, Koh-ichi Ide, Elements of ArtificialEmotion,IEEE International Workshop on Robot and HumanCommunication, 1995, p265-268

[18] Takanori Shibata,Toshihiro Tashima, Kazoo Tanie, Emergence ofEmotional Behavior through Physical Interaction between Humanand Robot,Proceedings of 1999 IEEE International Conference onRobotics & Automation, p2868-2873

[19] Luis M.Botelho, Pedro N.Ramos ,Machinery for Artificial Emotion,Cybernetics and Systems,2000

[20] Picard R W. Toward Computers that Recognize and Respond to userEmotion. IBM technical journal, 2000, 38(2): 705-719

[21] Cowie.R.Emotion Recognition in human-computer interaction. IEEESignal Processing Magazine.2001, 18(1): 32-80

[22] Keith Waters,A Muscle Model for Animating Three-DimensionalFacial Expression, Computer Graphic, Vol 21,Number 4, July 1987

[23] James D, Steve Maddock, Expressive Visual Speech using GeometricMuscle Functions

1424


Recommended