Emotion Based Motion Generator Using

Page 1 of 5

Summer Simulation Multiconference - SCSC S 2005

www.liophant.org/scsc/scsc2005 ‫ ﺳﺎﯾﺖ‬- ‫ ﮐﺎﻧﺎدا‬SCSC ‫ﻣﻘﺎﻟﻪ ﻣﻨﺘﺸﺮه درﮐﻨﻔﺮاﻧﺲ‬

Emotion Based Motion Generator Using Computational Theory of Perception Mohsen Davoudi MSC, Department of Control Eng. AIU Tehran south University

MB Menhaj Prof, Department of Electrical Eng. Amir Kabir University of Technology

Abstract Human has some perception based behavior seen in his body movements. Movements of hands, legs, head, etc show the inner emotion of the person in specific times. We define a standard GCL for humanoid robots to categorize data needed for trajectory tracking for each joint which lead to Motion Generation. An Artificial Neural Network (ANN) generates velocity and acceleration for each joint of humanoid using GCL values. A concept that plays a key role in Motion Generation with CTP is GCL.

1. INTRODUCTION Humanoid Robots are going to do many tasks like humans. But as we know human has some perception based behavior. Humans have capability to perform a wide variety of physical motions without any measurements and any computations. In performing such movements, humans employ perceptions of time, situation, possibility, inner emotions and other attributes of physical and mental objects. Reflecting the bounded ability of the human brain to resolve details of motions, perceptions are intrinsically imprecise. The computational theory of perceptions (CTP) contains a special capability to compute and reason with perception-based information. This paper introduces a fuzzy logic-based analyzer that interprets the linguistic emotions that are common among people into what is called the Generalized Constraint Language (GCL) [1]. Fuzzy Logic-Based probability theory has the fundamental ability to operate on perception-based information, which bivalent logic-based probability theory does not posses. Hence, fuzzy logic offers a conceptual structure, which accommodates partiality and granularity. We define a standard GCL for humanoid robots to categorize data needed for

Mehdi Davoudi BSC, Department of Electrical Eng. Imam Hossein University

trajectory tracking for each joint which lead to Motion Generation. Uncertainty is an attribute of the input information. This information, whatever its form is, after Fuzzy analysis, may be represented as GCL [2]. Fuzzy analyzer gets two types of Linguistic proposition that are detected from a sentence: 1) The type of emotion, 2) The intensity of emotion. For example I say “move your hands very angry”, two words are detected: angry and very. An idea which underlies the approach described in this paper is that an emotion may be viewed as a proposition that fuzzy analyzer approximates by means of the intensity and type of emotion. Furthermore, a proposition plays the role of a carrier of information. In the next step, an Artificial Neural Network (ANN) generates velocity and acceleration for each joint of humanoid using GCL values as inputs. A concept that plays a key role in Motion Generation with CTP is GCL. Because it is an interface between two sections of this analyze: a) emotion interpreter (Fuzzy analyzer), b) a data generator for physical motion of the humanoid robot. The computational theory of perceptions enhances the ability of AI to deal with real-world problems such as robots in which decision relevant information is a mixture of measurements and perceptions. Humanoid Motion Generation is compromise between perceptions of emotions in common sense and measured data for trajectory tracking in humanoid’s joints. Figure 1 shows the concept that is outlined in this paper. In order to deal with perception-based analysis of the motion, an array of tools centering on Computing with Words and Perceptions (CWP), Computational Theory of Perceptions (CTP), Precisiated Natural Language (PNL), and Perception-Based Probability Theory (PPT), are required [3].

Summer Simulation Multiconference - SCSC S 2005

Figure1. The whole Motion Generator system

2. FUZZY ANALYZER CW (Computing with Words) serves two major purposes in Motion Generation project: a) provides a machinery for dealing with emotions in which precise information is not available b) provides a machinery for dealing with emotions in which precise information is available, but there is a tolerance for imprecision which can be exploited to achieve tractability, robustness, simplicity and low solution cost PNL is a sublanguage of precisiable propositions in NL which is equipped with two dictionaries: (A) NL to GCL; (B) GCL to PFL (Protoform Language); and (C) a collection of rules of deduction (rules of generalized constrained propagation) expressed in PFL. Perceptions in Motion Generation are f-granular in the sense that (a) the boundaries of perceived classes are fuzzy; and (b) the values of perceived attributes are granular, with a granule being a clump of values drawn together by indistinguishability, similarity, proximity or functionality of emotions. p=proposition in a natural language a proposition is a perception if it contains fuzzy quantifiers: many, most, few, …; fuzzy qualifiers: usually, probably, possibly, typically, generally, …; fuzzy modifiers: very, more or less; extremely, …; and/or fuzzy nouns, adjective or adverbs such as emotion nouns that is common among people in a culture. We considered a number of inner emotions in this paper: Angriness, kindness, stress, happiness. The mathematics of perception-based information has a higher level of generality than the mathematics of measurement-based information.

The constrained variable, X, may assume a variety of forms. In particular, • X is an n-array variable, X=(X1, …, Xn) • X is a proposition, e.g., X= very angry • X is a function • X is a function of another variable, X=f(Y) • X is conditioned on another variable, X/Y • X has a structure, e.g., X=Emotion(Offensiveness(Stress)) • X is a group variable. In this case, there is a group, G[A]; with each member of the group, Namei, i=1, …, n, associated with an attribute– value, Ai. Ai may be vector-valued. Symbolically G[A]: Name1/A1+…+Namen/An. Basically, G[A] is a relation. • X is a generalized constraint, X= Y isr R [4]. In this paper two forms of X is used: 1) proposition form, 2) function form as measure-based information.

Figure 2. Fuzzy sets of two parameters: 1) kindness, 2) stress.

Considering figure2 : P(KT) = S1\K1 + S2\K2 + S3\K3 We supposed four emotion that is common among people: U={e1,e2,e3,e3} e1: Kindness = k e2: Stress = s e3: Angriness = a e4: Happiness = h Decision = f (e1 , e2 , e3 ,… ) Offensiveness = f (e i ) Intensity = g (e i ) i=1…,4

Page 2 of 5

Page 3 of 5

Summer Simulation Multiconference - SCSC S 2005

In tables 1 ,2 a phrase is converted to prototype form. Membership functions of offensiveness:

µ o (i ) = sup e ( µ o ( ∫ µ ei (u ) f (u )du )) U

proposition in NL

precisiation

p

p* (GC-form)

John is so much happy

ΣCount (intensity. happiness /happiness) is much

Membership functions of intensity:

µ i (i ) = sup e ( µ o ( ∫ µ ei (u ) g (u )du )) U

TABLE 1. POSITION INTERPRETATION

precisiation

protoform

p* (GC-form)

PF(p*)

ΣCount (intensity. Q A’s are B’s happiness /happiness) is much TABLE 2.PRECISIATION INTERPRETATION Figure3. Fuzzy analyzer

A proposition, p, in a natural language, NL, is precisiable if it translatable into a precisiation language In the case of PNL, the precisiation language is the Generalized Constraint Language, GCL precisiation of p, p*, is an element of GCL. Figure 4 shows this concept [5].

P after parsing is converted to semantic parse that may be one of these forms: 1) Logical form 2) Semantic network 3) Conceptual graph 4) Canonical form Semantic parse can be abstracted to more than one protoforms. In order to computation of GCL form for emotion sets we use fuzzy analyzer as inference motor. A simple fuzzification for happiness is shown in figure 5 (a). µ is membership function of intensity. Q is happiness and ant (Q) is sadness. The happiness definition may be changed by cultures. (Figure5 (b))

Figure4. GCL form of input information An Example: John is so much happy today. He wants to play guitar. Combined propositions are: (happy+ act of playing guitar). Deduction with perceptions involves the use of protoformal rules of generalized constraint propagation. Guitar playing is an action that is done using organs specially hands. Therefore considering his inner emotion, his movement is so quick and we can guess happiness from his motion.

Figure 5. Emotion definition

Summer Simulation Multiconference - SCSC S 2005

3. GENERALIZED CONSTRANTS FOR MOTION GENERATION

A. Possibilistic of Motion Generation (r=blank)

Page 4 of 5

ts ( g ) = µ likely ( ∫ g (u ) µ happy (u )). As other consideration Bimodal (r=bm) In the bimodal constraint, X isbm R,

X is R Which R playing the role of the possibility distribution of X. For example: X is [a, b]

R is a bimodal distribution of the form R:

Σi ki\Ai

, i=1, …, n.

This means that Means that [a, b] is the set of possible values of X. Another example: X is “happy”.

Prob(X is Ai) is ki.

In this case, the fuzzy set labeled happy is the possibility distribution of X. If µsmall is the membership function of happy, then the semantics of “X is happy” is defined by Poss{X=happy} = µsmall(happy) Where u is a generic value of X [6].

Figure6. database interaction

B. Probabilistic in Motion Generation (r=p) X isp R, Which R playing the role of the probability distribution of X. For example: X isp N(m, σ2) Means that X is a normally distributed random variable with mean m and variance σ2. If X is a random variable which takes values in a finite set of emotions {e1, …, en} with respective probabilities k1, …, kn, then X may be expressed symbolically as X isp (k1\e1+…+ kn\en), With the semantics Prob(X=ei)=ki,

Motion characteristics in humans consist of measurement-based information and perception-based emotions. A big percent of human knowledge is perception-based. True mimicry or right motion expression in each situation is the goal that we consider to design humanoid robots. So defining some rules and creation database for emotion definition is a base for designing. (Figure 6) Information about trajectory of each joints in a humanoid, the time that motions accrue and the supposed age for the humanoid are measurement-based information that can be retrieved from sensors and the motion generator algorithm in a humanoid. The offensiveness and intensity variables which are GCL form of perception-based information, are inputs of ANN.

Prob(Emotion(happy) is High) Is likely

∫ g (e4 ) µ high (e4 )de4 i=1, …, n.

Prob(X is kind) = ∫ µ small (u ) g (u ) du

Or, in annotated form, GC (g) = X / ∫ g (e) µ high (e) de is R/likely.

Hence The test-score of this constraint on f and g is given by

Page 5 of 5

Summer Simulation Multiconference - SCSC S 2005

4.CONCLUSION ts(f)= µ likely ( ∫ f (u ) µ high (u )du ) ts(g)= µ likely ( ∫ g (u ) µ high (u ) du )

The last step for mimicry is the generation of the movements from the GCL variables. Using an ANN with less than 30 neurons GCL form is translated to q& , q&& and delay time value for each joint of humanoid. (Figure 7)

Motion Generation is such an area, where presentations of the proper motion are based on a combination of various quantitative as well as qualitative (perceptive) measurements. In this perspective, a major shortcoming of existing approaches is that, based as they are on bivalent, they don’t provide tools for dealing with perception-based information. It is not possible to construct a computational theory of perceptions within the conceptual structure of bivalent logic and probability theory in this area.

REFERENCES [1]

Figure 7. Artificial Neural Networks Multi Layer Perceptron (MLP) , RBF and Recurrent Neural Networks are candidate networks for this purpose. In this paper a MLP is used. (Figure 8)

Figure 8. Multi Layer Perceptron (MLP)

A. Bardossy, L. Duckstein, Fuzzy Rule-based Modelling with Application to Geophysical, Biological and Engineering Systems, CRC Press, 1995. [2] D. Dubois, H. Prade, Gradual inference rules in approximate reasoning, Information Sciences: An International Journal 61 (1-2) (1992) 103-122. [3] D. Dubois, H. Fargier, H. Prade, The calculus of fuzzy restrictions as a basis for flexible constraint satisfaction, in: Proc. 2nd IEEE Int. Conf. on Fuzzy Systems, San Francisco, CA, 1993, pp. 1131-1136. [4] D. Dubois, H. Prade (Eds.), Fuzzy Information Engineering: a Guided Tour of Applications, John Wiley and Sons, 1996. [5] D. Filev, R.R. Yager, Essentials of Fuzzy Modeling and Control, Wiley-Interscience, 1994. [6] M. Higashi, G.J. Klir, Measures of Uncertainty and Information Based on Possibility Distributions, Fuzzy Sets for Intelligent Systems, Morgan Kaufmann Publishers, San Mateo, CA, 1993, pp. 217-232. [7] M. Jamshidi, A. Titli, L.A. Zadeh, S. Boverie (Eds.), Applications of Fuzzy Logic towards High Machine Intelligence Quotient Systems, Environmental and Intelligent Manufacturing Systems Series, vol. 9, Prentice Hall, Upper Saddle River, NJ, 1997. [8] A. Kaufmann, M.M. Gupta, Introduction to Fuzzy Arithmetic: Theory and Applications, Von Nostrand, New York, 1985. [9] E.H. Mamdani, S. Assilian, An experiment in linguistic synthesis with a fuzzy logic controller, International Journal of Man-Machine Studies 7 (1975) 1-13. [10] V. Novak, I. Perfilieva, J. Mockor, Mathematical Principles of Fuzzy Logic, Kluwer, Boston/ Dordrecht, 1999. [11] H.T. Nguyen, On Modeling of Linguistic Information Using Random Sets, Fuzzy Sets for Intelligent Systems, Morgan Kaufmann Publishers, San Mateo, CA, 1993, pp. 242-246. [12] T.J. Ross, Fuzzy Logic with Engineering Applications, second ed., Wiley, 2004. [13] J. Yen, R. Langari, Fuzzy Logic: Intelligence, Control and Information, first ed., Prentice Hall, Berlin, 1998. [14] L.A. Zadeh, The concept of a linguistic variable and its application to approximate reasoning, Part I: Information Sciences 8 (1975) 199-249; [15] L.A. Zadeh, A computational approach to fuzzy quantifiers in natural languages, Computers and Mathematics 9 (1983) 149184. [16] L.A. Zadeh, Toward a perception-based theory of probabilistic reasoning with imprecise probabilities, Journal of Statistical Planning and Inference 105 (2002)

Summer Simulation Multiconference - SCSC S 2005

Emotion Based Motion Generator Using Computational Theory of Perception

Journal: Manuscript ID: Track: Date Submitted by the Author: Complete List of Authors: Keywords:

Summer Computer Simulation Conference draft Life & Environmental Science Simulation n/a davoudi, mohsen Menhaj, Mohammad bagher; Amir kabir university, Electrical Eng. Emotion, fuzzy, neural networks, motion