An approach of multi-agent control of bio-robots using intelligent recognition diagnosis of persons with moving disabilities/Daugiaagentis biorobotu valdymas taikant intelektualizuotus asmenis su judejimo negalia diagnozes atpazinimo metodus.
Bielskis, Antanas Andrius ; Dzemydiene, Dale ; Denisov, Vitalij 等
1. Introduction
The developing processes of intelligent systems with adaptive
e-services are complex and important issues for providing user-friendly
e-health and e-social care for people with moving disabilities. Such
systems include different intellectual components for the control and
monitoring of sensors, by supporting multi-agent activities and, in
accordance to the recognition of certain situations, integrate the
possibilities to affect and control the devices of disabled persons
(Bielskis et al. 2008; Gricius et al. 2008; Kaklauskas et al. 2006).
The robots under development for medical care applications are
based on a new conceptual understanding of several different scientific
areas. Substantial works exist in the robotics literature on the
mechanical design, modeling, gait generation and implementation of
adulatory robotic prototypes (Sfakiotakis, Tsakiris 2007; Zavadskas et
al. 2007, 2008a, b, c). New human-robotics interface, consisting in an
eye-tracker interfaced with a transcranial magnetic stimulator device,
is designed in mobile robots (Mariottini et al. 2007). The bio-robotic
devices need the integration of the means of different knowledge
interpretation techniques for making the intelligence medical diagnosis
(Bernardi et al. 2008; Dzemydien? et al. 2008a, b; Samanta, Nataraj
2008; Treigys et al. 2008).
We recognize the possibilities to integrate different types of
knowledge representation techniques and to construct the control
sub-systems of bio-robot system. The control subsystem is working as
on-line diagnostic systems of complex mechanisms with the cooperation of
multi-agent's activities in recognizing human affect sensing.
Being able both to provide an intelligent accident preventive
robot-based support for people with moving disabilities and to include
affect sensing in Human Computer Interaction (HCI, in providing e-health
care for people with moving disabilities), Human-Robot Interaction (HRI,
for assisting tele-healthcare patients remaining autonomous), and
Computer Mediated Communication (CMC, in providing adaptive user-robot
friendly collaboration), such system should depend upon the possibility
of extracting emotions without interrupting the user during HCI, HRI, or
CMC (Mandryk, Atkins 2007; Bielskis et al. 2007; Pentland 2004).
Emotion is a mind-body phenomenon that is accessible at different
levels of observation (social, psychological, cerebral and
physiological). The models of an intelligent multi-agent based e-health
care systems for people with moving disabilities are recently proposed
by (Villon, Lisett 2006; Bielskis et al. 2008). The continuous
physiological activity of a disabled person is being made accessible by
the use of intelligent agent-based bio-sensors coupled with computers.
The aims of this research are devoted to the investigations of the
integration of different knowledge representation techniques for the
development of the reinforcement framework within multiple cooperative
agents' activities for the recognition of the prediction criteria
of diagnosis of the emotional situation of disabled persons. The
research results present further development of a multi-layered model of
this framework with the integration of the evaluation of fuzzy neural
control of speed of two wheelchair type robots working in real time by
providing moving support for disabled individuals. The approach of
reasoning by using fuzzy logical Petri nets (Jiang, Zheng 2000;
Dzemydiene 2001) is described to define physiological state of disabled
individuals by recognizing their emotions during their relaxation
activities based on playing computer games.
2. An adaptive control of robot motion in the system
The proposed reinforcement framework of intelligent remote
interaction of bio-robots is based on the distributed information
systems with important personal data of the patients and of monitoring
sensor's data. The framework is presented at Fig. 1. It includes
two adaptive moving wheelchair-type robots which are remotely
communicating with two wearable human's affect sensing bio-robots.
To capture towards e-social and e-health care context relevant episodes
based on humans affect stages (Vilon, Lisett 2006), the context aware
sensors are incorporated into the design of the Human's Affect
Sensing Bio-Robot-x (HASBR-x) for every disabled individual, and
information based on these episodes is used by local Intelligent
Decision Making Agent-x (IDMA-x) to control every intelligent support
providing robot.
This framework allows a multi-sensor data fusion before
transmitting the data to the Remote Control Server (RCS) to minimize the
TCP/IP bandwidth usage. Multi-agent based adaptive motion control of
both robots is based on an adaptive Fuzzy Neural Network Control (FNNC)
in according to the approach shown in Fig. 2. The architecture of the
FNNC controller represents an approach of the Adaptive Neural Fuzzy
Inference System, the ANFIS that combines the field of fuzzy logic and
neural networks shown in Fig. 2a (Rubaai et al. 2005).
The possibility to learn about the nonlinear dynamics and external
disturbances of the motor speed controller with a stable output, small
steady error, and fast disturbance rejection is integrated in this
framework. At the k-th moment, the difference between motor speed
reference value v(k) and motor speed output value [v.sub.o](k) is split
to speed error e(k) and speed error change [DELTA](k).Those values are
used by the proposed in (Jiang, Zheng 2000) NN Learning Agent presented
on Fig. 2b for the learning of artificial neural network Artificial NN
on Fig. 2a as well as 2nd order input vector of the Artificial NN. The
output of the Artificial NN generates percentage value of pulse width change [DELTA]PW(k) to describe how much pulse width value PW(k) of the
real motor speed control value at the moment k should be changed. This
value is then generated in real time by the ATmega32 microcontroller to
perform online calculating:
PW(k) = PW(k - 1) + [DELTA]PW(k).
[FIGURE 1 OMITTED]
[FIGURE 2 OMITTED]
A simplified architecture of the neural-fuzzy controller (Rubaai et
al. 2005) is presented on Fig. 3. The layer 1 in Fig. 3 represents
inputs X = e(k) and Y = [DELTA]e(k) to the fuzzy neural controller, the
speed error e(k) and the change in speed error [DELTA]e(k) =
e(k)-e(k-1), respectively. The layer 2 consists of 7 input membership
nodes with four membership functions, [A.sub.1], [A.sub.2], [A.sub.3],
and [A.sub.4], for input X and three membership functions, [B.sub.1],
[B.sub.2], and [B.sub.3] for input Y, the membership value specifying
the degree to which an input value belongs to a fuzzy set is determined
in this layer. The triangular membership function is chosen owing to its
simplicity. For the change in motor speed error [delta]e(k), the initial
values of the premise parameters (the corner coordinates [a.sub.j],
[b.sub.j] and [c.sub.j] of the triangle) are chosen so that the
membership functions are equally spaced along the operating range of
each input variable. The weights between input and membership level are
assumed to be unity. The output of neuron j = 1, 2, 3, and 4 for input i
= 1 and j = 1, 2, and 3 for input i = 2 in the second layer can be
obtained as follows:
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.]
where [a.sub.j], [b.sub.j] and [c.sub.j] are the corners of the
j-th triangle type membership function in layer 2 and [X.sub.i] is the
i-th input variable to the node of layer 2, which could be either the
value of the error or the change in error. The layer 1 in Fig. 3
represents inputs X = e(k) and Y = [DELTA]e(k) to the fuzzy neural
controller, the speed error e(k) and the change in speed error
[DELTA]e(k)=e(k)-e(k-1), respectively.
Layer 2 consists of 7 input membership nodes with four membership
functions, [A.sub.1], [A.sub.2], [A.sub.3], and [A.sub.4] for input X
and three membership functions, [B.sub.1], [B.sub.2], and [B.sub.3] for
input Y, as shown in Fig. 3. The weights between input and membership
level are assumed to be unity. Each node in Rule layer 3 of Fig. 3
multiplies the incoming signal and outputs the result of the product
representing one fuzzy control rule. It takes two inputs, one from nodes
[A.sub.1]-[A.sub.4] and the other from nodes [B.sub.1]-[B.sub.3] of
layer 2. Nodes [A.sub.1]-[A.sub.4] define the membership values for the
motor speed error and nodes [B.sub.1]-[B.sub.3] define the membership
values for the change in speed error. Accordingly, there are 12 nodes in
layer 3 to form a fuzzy rule base for two input variables, with four
linguistic variables for the input motor speed error e(k) and three
linguistic variables for the input change in motor speed change error
[DELTA]e(k).
[FIGURE 3 OMITTED]
The input/output links of layer 3 define the preconditions and the
outcome of the rule nodes, respectively. The outcome is the strength
applied to the evaluation of the effect defined for each particular
rule. The output of neuron k in layer 3 is obtained as [O.sup.3.sub.k] =
[[PI].sub.t] [w.sup.3.sub.jk] [y.sup.3.sub.j], where [y.sup.3.sub.f]
represents the j-th input to the node of layer 3 and [w.sup.3.sub.fk] is
assumed to be unity. Neurons in the output membership layer 4 represent
fuzzy sets used in the consequent fuzzy rules. An output membership
neuron receives inputs from corresponding fuzzy rule neurons and
combines them by using the fuzzy operation union. This was implemented
by the maximum function. The layer 4 acts upon the output of layer 3
multiplied by the connecting weights. These link weights represent the
output action of the rule nodes evaluated by layer 3, and the output is
given [O.sup.4.sub.m] = max([O.sup.3.sub.k] [w.sub.km]), where the count
of k depends on the links from layer 3 to the particular m-th output in
layer 4 and the link weight [w.sub.km] is the output action of the m-th
output associated with the k-th rule. This level is essential in
ensuring the system's stability and allowing a smooth control
action. Layer 5 is the output layer and acts as a defuzzifier. The
single node in this layer takes the output fuzzy sets clipped by the
respective integrated firing strengths and combines them into a single
fuzzy set. The output of the neuro-fuzzy system is crisp, and thus a
combined output fuzzy set must be defuzzified. The sum-product
composition method was used. It calculates the crisp output as the
weighted average of the cancroids of all output membership functions as
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], where a[c.sub.m]
and b[c.sub.m] for m = 1, 2, .., and 5 are the centres and widths of the
output fuzzy sets, respectively. The values for the b[c.sub.m]'s
were chosen to be unity. This scaled output corresponds to the control
signal (percent duty cycle) to be applied to maintain the motor speed at
a constant value. The only weights that are trained are those between
layers 3 and layer 4 of Fig. 3. The back-propagation network is used to
train the weights of this layer. The weights of the neural network were
trained offline by using an open source type R-programming environment
before they were used in the online real time experimental by applying
the modified learning algorithm from (Rubaai et al. 2005):
Step (1): Calculate the error for the change in the control signal
(duty cycle) for ATmega32-based microcontroller as [E.sub.o] = [T.sub.o]
- [O.sup.5.sub.o], where [E.sub.o], [T.sub.o], and [O.sup.5.sub.0] are
the output error, the target control signal, and the actual control
signal;
Step (2): Calculate the error gradient [[delta].sub.m] = [E.sub.o]
[MATHEMATICAL EXPRESSION NOT REPRODUCIBLE IN ASCII.], where a[c.sub.i]
for i = 1 ... 5 are the centres of the output fuzzy sets and
[O.sup.4.sub.j] is the firing strength from node j in layer 4;
Step (3): Calculate the weight correction [DELTA][w.sub.km] =
[eta][[delta].sub.m][O.sup.3.sub.j] to increasing the learning rate.
Here Sejnowski--Rosenberg updating mechanism was used, which takes into
account the effect of past weight, changes on the current direction of
the moving in the weight space. This is given by [DELTA][w.sub.km](t) =
[eta](1 - [alpha])[[delta].sub.m][O.sup.3.sub.km] +
[alpha][DELTA][w.sub.km] (t - 1), where a is a smoothing coefficient in
the range of 0 ... 1,0, and [eta] is the learning rate;
Step (4): Update the weights [w.sub.km](t + 1) = [w.sub.km](t) +
[DELTA][w.sub.km](t), where t is the iteration number. The weights
linking the rule layer (layer 3) and the output membership layer (layer
4) are trained to capture the system dynamics and therefore minimize the
ripples around the operating point.
3. Human computer interaction in the system
There are many different methods of recognizing physical state or
behaviour by using data of wearer's emotion recognition sensors
(Mandryk, Atkins 2007; Pentland 2004). In this paper, a modified
Arousal--Valence model by (Mandryk, Atkins 2007) of Fig. 4 was used to
discover information in real time for providing some friendly advices
for a person with moving disabilities during his/her relaxation period
based on playing computer games with computer as well as with his/her
partner--another disabled person.
3.1. Fuzzy system for emotion recognition
The framework presented in Fig. 1 uses four emotion recognition
sensors for each of disabled individual: the ECG (Electrocardiogram);
the SCR (skin conductance response); the STH (skin temperature of head),
and the S[T.sub.F] (skin temperature of finger) to provide HR (heart
rate), HR[V.sub.H] (heart rate variability for the range of 0.15 to 0.4
Hz), HR[V.sub.L] (heart rate variability for the range of 0.015 to 0.15
Hz), SCR, S[T.sub.H], and S[T.sub.F] inputs for defining fuzzy values of
arousal and valence (Fig. 5).
The principal scheme of the integration of different components of
the modelling emotions is presented in Fig. 5 to provide HR(heart rate),
HR[V.sub.H](heart rate variability for the range of 0.15 to 0.4 Hz),
HR[V.sub.L] (heart rate variability for the range of 0.015 to 0.15 Hz),
SCR(skin conductance response), S[T.sub.H] (skin temperature of head),
and S[T.sub.F] (skin temperature of finger) inputs to define fuzzy
values of arousal and valence. The number of membership functions of
Fig. 5 and 6 applied to that input or output follows the
input/output labels. Within each input and output, there is a schematic representation of the location and form of the membership functions. In
Fig. 5, all membership functions are triangular, while in Fig. 6
membership functions are trapezoidal, exhibited by the flat ceilings,
rather than the peaked ceiling of a triangular membership function. The
system in Fig. 6 uses 67 rules proposed in (Mandryk, Atkins 2007) to
transform 2 inputs (the arousal and valence) into 5 outputs (fun,
challenge, boredom, frustration, and excitement).
[FIGURE 4 OMITTED]
[FIGURE 5 OMITTED]
[FIGURE 6 OMITTED]
The model of fuzzy system by (Mandryk, Atkins 2007) for the
recognition of five emotional states (fun, challenge, boredom,
frustration, and excitement) from arousal and valence is shown in Fig.
6.
The computing results of fuzzy reasoning were obtained by the
application of fuzzy logical Petri nets (Jiang, Zheng 2000). Classical
Petri nets are defined as a structure N = <S, T, F> where S means
a set of places, T is a set of transitions and F is F [??] (S x T)
[union] (T x S), where ([for all] t [member of] T)([there exists] p,q
[member of] S)(p,t), (t,q) [member of] F. Graphical representation is
set up by the following symbols:
Places--by rings, Transitions--by rectangles, and Relations--by
pointers between transitions and places or places and transitions. In
classical Petri nets, there is a token placed if the expression is true
(1) or not if it is false (0). Any IF-THEN rule is given of the form of
IF [X.sub.1] is [A.sub.1] AND [X.sub.1n is [A.sub.n] THEN Y is B ,
where [A.sub.1], .., [A.sub.n] and B are certain predicates
characterizing the variables [X.sub.1], ..., [X.sub.n] and Y. The set of
IF-THEN rules forms linguistic description:
[R.sub.1] : = IF [X.sub.1] is [A.sub.11] AND ... AND [X.sub.n] is
[A.sub.1n] THEN Y is [B.sub.1],
[R.sub.m] : = IF [X.sub.1] is [A.sub.m1] AND ... AND [X.sub.n] is
[A.sub.mn] THEN Y is [B.sub.m],
where each transition of the result fuzzy Petri net corresponds to
one rule of linguistic description. Models of Logical Petri Nets Applied
to Transforming 89 Fuzzy Inference Rules to Constructing Support
Information System for Bio-Robots are proposed in Table 1. In the Table
2, the set of 22 rules and corresponding transitions are proposed in
concerting galvanic skin response, the GSR, heart rate, the HR, heart
rate variability high, the HR[V.sub.H], heart rate variability low, the
HR[V.sub.L], skin temperature of head, the S[T.sub.H], and skin
temperature of finger, the S[T.sub.F] into arousal and valence. In Table
3, the corresponding transitions of logical Petri nets are prescribed to
the set of 67 rules used by (Mandryk, Atkins 2007) in concerting of
arousal-valence space transformation into five modelled emotional states
to converting arousal and valence into boredom, challenge, excitement,
frustration, and fun.
3.2. Reasoning algorithms used by human arousal recognition agents
To derive emotions of user 1 and user 2 during their relaxation
state of interactive playing of computer-based games, agents HARA-1,
HARA-2, IDMA-1, and IDMA-2 of Fig. 1 were programmed by using the
reasoning algorithm of fuzzy logical Petri nets by (Pavliska 2006) (Fig.
7). This algorithm gets fuzzy Petri net as an input and creates a set of
linguistic descriptions corresponding to each output place of fuzzy
Petri net. Human arousal recognition agents HARA-1 and HARA-2 of Fig. 1
were programmed to use these reasoning algorithms to create some
friendly advices to disabled individuals who are taking part in the
model of e-social care system for people with moving disabilities of
Fig. 1 during their relaxation period based on playing computer games.
[TABLE 1 OMITTED]
Input : fuzzy Petri net: fpn
output: set of linguistic descriptions: lfln
lfln = [empty set];
foreach output place op of fpn do // create linguistic description
// create set ol Input variables (places) on whose op depends
inputs = [empty set];
foreach input transition if of op do
// add all Inputs ol transition it to inputs set
inputs = inputs [union] it.inputs;
end
// construct linguistic description (set of rules)
rule = [empty set];
foreach input transition if of op do
// construct rule corresponding to transition it
rule = [empty set];
foreach element in from inputs do
If rule [not equal to] [empty set] then rule = rule + AND;
if in [member of] it.inputs then
rule = rule + in.name is edge(in,it).value;
else
rule = rule -f in.name is UNDEF;
end
end
rule = rule + THEN op.name is edge(it,op).value;
rb = rb [union] rule; // add rule to rule base
end
lfln - lfln [union] rb: // add rule base to set of linguistic
description
end
Fig. 7. Reasoning algorithm by (Pavliska 2006) used for the
implementation of fuzzy logical Petri nets of this framework
3.3. Scenario of using fuzzy logical Petri nets in the system
Let us consider the following scenario of the organization of some
relaxation activities based on playing computer games by disabled
individuals who are taking part in the model of e-social care system for
people with moving disabilities of Fig. 1.
Step (1): The IDMA-1 agent initiates the IDMA-2 agent to asking
Wheelchair-2 robot to take the Human's Affect Sensing Bio-Robots
HASBR-1 and 2 from the shelf, bring them to user 1 and 2 for taking
their on for both users, ask users to log on into the system to start
relaxation activities based on playing of computer games.
Step (2): Agents HARA-1 and HARA-2 initiate starting of the game,
record measured data into the Personal Information Database of each
user, transforming emotion measurements into arousal-valence space, and
infer level of emotions of each player.
Step (3): Human Computer Interaction (HCI) in the system adaptively
generates friendly advices to each player into his/her Personal
Information Database and periodically provides necessary e-game support
advices for user1 and user2 based on their emotional states discovered
during the game in real time.
Step (4): To propose self-adaptively controllable social care aware
moving actions by robot 1 and 2 for a given user with moving
disabilities, the Q-learning algorithm of Fig. 8 (Touzet, Watkins 1989)
was implemented in the modified agent based adaptive FNNC-type DC motor
speed controller of Fig. 2a for defining an optimal path of robots
Wheelchair 1 and 2 of Fig. 1 to go from randomly selected point A to
point B under unknown environmental conditions.
The following fuzzy logical Petri net of this scenario is given on
Fig. 9. In Fig. 9, BLACK BOX 1 implements transitions
[T.sub.1]-[T.sub.22.sup.OR] (Table 2), and BLACK BOX 2 represents
[T.sub.23.sup.AND]-[T.sub.89] (Table 3) transitions of Petri net.
Initialize Q(s,a) arbitrarily
Repeat (for each episode):
Initialize s
Repeat (for each step of episode):
Choose a from s using policy derived from Q (e.g., [epsilon]-greedy)
Take action a, observe r,s'
Q(s,a) [left arrow] Q(s,a)+[alpha][r+[gamma][max.sub.a]
Q(s',a')-Q(s,a)]
s[left arrow]s' until s is terminal
Fig. 8. Q--learning: An off-policy TD control algorithm of robots
Wheelchair 1 and 2 by (Touzet, Watkins 1989)
4. Conclusions
In the process of the elaboration of the reinforcement framework
with multiple cooperative agents for the recognition of an appropriate
prediction criteria of diagnosis of emotional situation of disabled
persons, an approach is presented based on further development of
multi-layered model of this framework with the integration of the
evaluation of fuzzy neural control of the speed of two wheelchair type
robots working in real time. This was done by the implemention of the
moving support for disabled individuals by using information based on
emotion state of each disabled person.
[FIGURE 9 OMITTED]
The proposed framework uses four emotion recognition sensors for
each disabled individual: the ECG (Electrocardiogram), the SCR (Skin
Conductance Response), the S[T.sub.H] (Skin Temperature of Head), and
the S[T.sub.F] (Skin Temperature of Finger) to provide HR (Heart Rate),
HR[V.sub.H] (Heart Rate Variability for the range of 0.15 to 0.4 Hz),
HR[V.sub.L] (Heart Rate Variability for the range of 0.015 to 0.15 Hz),
SCR, S[T.sub.H], and S[T.sub.F] inputs for defining fuzzy values of
arousal and valence of disabled person.
The method of fuzzy reasoning by using fuzzy logical Petri nets
based on transforming of arousal-valence space into five modelled
emotional states to convert arousal and valence into boredom, challenge,
excitement, frustration, and fun is described. The method allows both
defining of physiological state of disabled individuals and giving them
online advices based on recognition of their emotions during their
relaxation activities based on playing computer games with each other
and against the computer.
doi: 10.3846/1392-8619.2009.15.377-394
Received 4 March 2009; accepted 20 August 2009
Reference to this paper should be made as follows: Bielskis, A. A.;
Dzemydien?, D.; Denisov, V.; Andziulis, A.; Drungilas, D. 2009. An
approach of multi-agent control of bio robots using intelligent
recognition diagnosis of persons with moving disabilities, Technological
and Economic Developemt of Economy 15(3): 377-394
References
Bernardi, P.; Sanchez, E.; Schillaci, M.; Squillero, G.; Sonza
Reorda, M. 2008. An effective technique for the automatic generation of
diagnosis-oriented programs for processor cores, IEEE Transactions on
Computer-Aided Design of Integrated Circuits and Systems 27(3): 570-574.
doi:10.1109/TCAD.2008.915541.
Bielskis, A. A.; Denisov, V.; Kusinskas, G.; Ramasauskas, O.;
Romas, N. 2007. Modeling of human physiological parameters in an
E-Laboratory by SOM Neural Networks, Elektronika ir elektrotechnika
3(75): 77-82.
Bielskis, A. A.; Denisov, V.; Drungilas, D.; Gricius, G.;
Ramasauskas, O. 2008. Modeling of intelligent multi-agent based E-health
care system for people with movement disabilities, Elektronika ir
elektrotechnika 6(86): 37-42.
Dzemydiene, D. 2001. Temporal information management and decision
support for predictive control of environment contamination processes,
in Proc. of Fifth East-European Conference Advances in Databases and
Information Systems' ADBIS2001. Eds. A. Capliskas, J. Eder.
Vilnius: Technika 1: 157-172.
Dzemydiene, D.; Maskeliunas, S.; Jacobsen, K. 2008a. Sustainable
management of water resources based on web services and distributed data
warehouses, Technological and Economic Development of Economy 14(1):
38-50. doi:10.3846/2029-0187.2008.14.38-50.
Dzemydiene, D.; Maskeliunas, S. ; Dzemyda, I. 2008b.
Interoperability of information system components for monitoring of
sewage and intelligent analysis of water resources, Technological and
Economic Development of Economy 14(3): 260-278.
Gricius, G.; Drungilas, D.; Sliamin, A.; Lotuzis, K.; Bielskis, A.
A. 2008. Multi-agent-based e-social care system for people with movement
disabilities, Technological Research Works in Western Lithuania 6:
67-77.
Jiang, Ch.; Zheng, Y. 2000. Fuzzy Reasoning Based on Petri Nets.
Available from Internet:
<http://www.ici.ro/SIC/sic2000_3/art04.html>.
Kaklauskas, A.; Zavadskas, E. K.; Ditkevicius, R. 2006. An
intelligent tutoring system for constructions and real estate. 3rd
International Conference on Cooperative Design, Visualization and
Engineering, September 17-20, 2006, Mallora, Spain, Books Series:
Lecture Notes in Computer Science 4101: 174-181.
doi:10.1007/11863649_22.
Mandryk, R. L.; Atkins, M. S. 2007. A fuzzy physiological approach
for continuously modelling emotion during interaction with play
technologies, Int. J. Human-Computer Studies 6 R.L. 5: 329-347.
Mariottini, G. L.; Oriolo, G.; Prattichizzo, D. 2007. Image-based
visual servoing for nonholonomic mobile robots using epipolar geometry,
IEEE Transactions on Robotics 23(1): 87-100.
doi:10.1109/TRO.2006.886842.
Pavliska, V. 2006. Petri Nets as Fuzzy Modelling Tool. Available
from Internet: <http://irafm.osu.cz/research_report/84_rep84.pdf>.
Pentland, A. 2004. Health wear: medical technology becomes
wearable, IEEE Computer 37(5): 42-49.
Rubaai, A.; Ofoli, A. R.; Burge, L.; Garuba, M. 2005. Hardware
implementation of an adaptive network-based fuzzy controller for DC-DC
converters, IEEE Transactions on Industry Applications 41(6): 1557-1565.
doi:10.1109/TIA.2005.857461.
Samanta, B.; Nataraj, C. 2008. Automated diagnosis of cardiac state
in healthcare systems using computational intelligence, International
Journal of Services Operations and Informatics 3(2): 162-177.
doi:10.1504/IJSOI.2008.019331.
Sfakiotakis, M.; Tsakiris, D. P. 2007. Biomimetic centering for
undulatory robots, The International Journal of Robotics Research
26(11-12): 1267-1282. doi:10.1177/0278364907083394.
Touzet, C.; Watkins, 1989. Q-learning for Robot. Available from
Internet: <http://www.up.univ-mrs.fr/Local/umr_6149/umr/page_perso/Touzet/ Q-learning_for_Robots.pdf>.
Treigys, P.; Saltenis, V.; Dzemyda, G.; Barzdziukas, V.; Paunksnis,
A. 2008. Automated optic nerve disc parameterization, Informatica 19(3):
403-420.
Villon, O.; Lisett, Ch. 2006. A user-modelling approach to build
user's psycho-physiological maps of emotions using bio-sensors, in
Proc. of IEEE Roman, 269-276.
Zavadskas, E. K.; Kaklauskas, A.; Vlasenko, A. 2008a. Web-based VSA analyser decision support system for e-examination, in 8th International
Conference on Fuzzy Logic and Intelligent Technologies in Nuclear
Science "Computational Intelligence in Decision and Control",
World Scientific Proceedings Series on Computer Engineering and
Information Science, September 21-24, 2008, Madrid, Spain, 1: 1153-1158.
Zavadskas, E. K.; Naimaviciene, J.; Kaklauskas, A.; Krutinis, M.;
Vainiunas, P. 2008b. Multi-criteria decision support system of
intelligent ambient assisted living environment, in 25th International
Symposium on Automation and Robotics in Construction. ISARC-2008.
Selected papers (June 26-29, Vilnius, Lithuania). Edited by E. K.
Zavadskas, A. Kaklauskas, M. J. Skibniewski, 717-724.
Zavadskas, E. K.; Krutinis, M.; Kaklauskas, A.; Vainiunas, P.;
Naimaviciene, J.; Povilaityte, L.; Tupenaite, L. 2008c. Model for
multiple criteria analysis of integrated intelligent and built-up
ambient assisted living environment, in 25th International Symposium on
Automation and Robotics in Construction, ISARC-2008. Selected papers
(June 26-29, Vilnius, Lithuania). Edited by E. K. Zavadskas, A.
Kaklauskas, M. J. Skibniewski, 725-730.
Zavadskas, E.; Kaklauskas, A.; Maciunas, E.; Vainiunas, P.;
Marsalka, A. 2007. Real Estate's Market Value and a Pollution and
health Effect Analysis, in 4th International Conference on Cooperative
Design, Visualization and Engineering, September 16-20, 2007 Shanhai
Peoples R China, Lecture Notes in Computer Science 4674: 191-200.
doi:10.1007/978-3-540-74780-2_23.
Antanas Andrius Bielskis (1), Dale Dzemydien (2), Vitalij Denisov
(3), Arunas Andziulis (4), Darius Drungilas (5)
(1,3,4,5) Klaipeda University, Manto g. 84, 92294 Klaipeda,
Lithuania E-mail: (1) andrius.bielskis@ik.ku.lt; (3)
vitalij.denisov@ik.ku.lt; (4) arunas@ik.ku.lt; (5) dorition@gmail.com
(2) Mykolas Romeris University, Ateities g. 20, 08303 Vilnus,
Lithuania E-mail: daledz@mruni.lt
Antanas Andrius BIELSKIS is professor, doctor habilitatis of the
Department of Informatics of the Faculty of Nature Science and
Mathematics of Klaip?da University (Lithuania). He holds a diploma of
radio-engineering in 1959 of Kaunas Institute of Technology, PhD in
electronic engineering in 1968 of Moscow Institute of Communications,
and Doctor of Science in power electronics and communications in 1983 of
the Institute of Electrodynamics of Academy of Science of Ukraine (Dr
Habilitatis of Technical sciences in 1994). In 1959-1964, he was a
designer, project manager at the Klaip?da Ship Design Institute, in
1964-1990--senior lecturer, associate professor, head of the departments
of electrical engineering and physics-mathematics, professor of the
Klaip?da faculty of the Kaunas University of Technology, and from
1991--professor of Klaip?da University. He is assigned as a member of
the Joint doctoral evaluation committee in the fields P160, P170, P175
of the Klaip?da University and the Institute of Mathematics and Computer
Science. He was: in 1993-1996--a team member of the TEMPUS project
JEP-6001 "Computer Based Environmental Studies"
COBES-Lithuania; in 1998-1999--a teem member of the COPERNICUS project
CP-0636 "SME Software Support System"; in 2000-2003,--the
coordinator of the Working Group "Theoretical Fundamentals of
Informatics" (WG-1) within the SOCRATES/ERASMUS CDA project
"MOCURIS"--"Modern Curriculum in Information Systems at
Master Level". Since 1999 to the present, he is a visiting
professor in the SOCRATES/ERASMUS exchange project between the Klaip?da
University and the Wilhelmshaven University of Applied Science. He has
published over 100 of research articles, two manual books and one
monograph book. His research interests include: artificial intelligence
methods, knowledge representation and decision support systems, ambient
intelligence, and intelligent eco-social support systems.
Dale DzEMYDIENE is professor, doctor, head of the Department of
Informatics and Software Programs of the Social Informatics Faculty of
the Mykolas Romeris University (Lithuania). She holds a diploma with
honour of applied mathematics in specialization of software engineering
in 1980, Dr in mathematics-informatics in 1995, habilitation doctor
procedure in the field of social sciences of management and
administration in 2004, and long time works at the Department of
Software Engineering at the Institute of Mathematics and Informatics.
She has published about 100 research articles, three manual books and
one monograph book. She is an organizer of international conferences in
the area of information systems and database development. She is the
head of the Legal informatics section of Lithuanian Computer Society
(LIKS), member of European Coordinating Committee for Artificial
Intelligence (ECCAI) and member of Lithuanian Operation Research
Association. Her research interests include: artificial intelligence
methods, knowledge representation and decision support systems,
evaluation of sustainable development processes.
Vitalij DENISOV is associated professor, Dr head of the Department
of Computer Science at Klaip?da University. His research interests
include mathematical and simulation modeling of complex systems, systems
engineering, development of modeling and e-learning systems. He is the
author of more than 70 research papers, a number of textbooks and
educational software tools.
Arunas ANDzIULIS is professor, doctor, head of the Department of
Informatics engineering of the Faculty of Marine Engineering of the
Klaip?da University (Lithuania). He holds a diploma with honour of
physics in 1968 of Vilnius University, PhD in electronic engineering in
1968 of Kaunas Institute of Technology, habilitation doctor procedure in
the field of transport engineering sciences in 2007, and long time works
as a head, designer and project manager of the Department of microchip
technology of the MRI "Venta". The author has published about
200 research articles and one manual books. He is an organizer of
national conferences in the area of technical engineering. He is a
member of the Legal informatics section of Lithuanian Computer Society
(LIKS). His research interests include: artificial intelligence methods,
operation research, nanotechnology and modelling and optimization of
technical systems.
Darius DRUNGILAS is MSc. student at the Department of Computer
Science, Klaip?da University. He graduated BSc degree in computer
science, Klaip?da University (2007). He is the author of 2 scientific
publications. Research interests: methods of artificial intelligence,
agent-based modeling.
Table 2. The set of rules and corresponding transitions for recognition
of degree of emotional state
No Rules Transitions
1 If (GSR is high), then [[T.sub.1]]
(arousal is high)
2 If (GSR is high) or [[T.sub.2].sup.OR]
(HR is high), then
(arousal is high)
3 If (GSR is mid-low), then [T.sub.3]
(arousal is mid-low)
4 If (GSR is low) or [T.sub.4] OR
(HR is low), then
(arousal is low)
5 If (GSR is low) and [T.sub.5] AND
(HR is high), then
(arousal is mid-low)
6 If (GSR is high) and [T.sub.6] AND
(HR is low), then
(arousal is mid-high)
7 If (GSR is high) and [T.sub.7] AND
(HR is mid), then
(arousal is high)
8 If (GSR is mid-high) [T.sub.8] AND
and (HR is mid), then
(arousal is mid-high)
9 If (GSR is mid-low) and [T.sub.9] AND
(HR is mid), then
(arousal is mid-low)
10 IF ([HRV.sub.H] is high) and [T.sub.10] AND
(HRVL is low), then
(valence is very-high)
11 IF ([HRV.sub.H] is low) [T.sub.11] AND
and (HRVL is high), then
(valence is very-low)
12 IF ([HRV.sub.H] is medium) and [T.sub.11] AND
(HRVL is medium), then
(valence is neutral)
13 IF ([HRV.sub.H] is high) and [T.sub.13] AND
(HRVL is medium) then
(valence is high)
14 IF ([HRV.sub.H] is medium) and [T.sub.14] AND
(HRVL is high), then
(valence is low)
15 IF ([HRV.sub.H] is medium) and [T.sub.15] AND
(HRVL is low), then
(valence is high)
16 IF ([HRV.sub.H] is low) and [T.sub.16] AND
([HRVL.sub.L] is medium), then
(valence is low)
17 IF ([HRV.sub.H] is high) and [T.sub.17] AND
([HRV.sub.L] is high), then
(valence is neutral)
18 IF ([HRV.sub.H] is low) and [T.sub.18] AND
([HRV.sub.L] is low), then
(valence is neutral)
19 IF (HR is high) and [T.sub.19] AND
([HRV.sub.H] is high) and
([HRV.sub.L] is high), then
(valence is high)
20 IF ([ST.sub.H] is high) or [T.sub.20] OR
(STL is low), then
(valence is low)
21 IF ([ST.sub.H] is low) or [T.sub.21] OR
([ST.sub.L] is high), then
(valence is high
22 IF ([ST.sub.H] is medium) or [T.sub.22] OR
([ST.sub.L] is medium),
then (valence is medium)
Table 3. Set of rules and corresponding transitions of logical Petri
nets of transforming of the arousal-valence space
No Rules Transitions
23 If (arousal is not very low) and [T.sub.23] AND
(valence is mid-high), then (fun is low)
24 If (arousal is not low) and (valence is [T.sub.24] AND
mid-high), then (fun is low)
25 If (arousal is not very low) and [T.sub.25] AND
(valence is high), then (fun is medium)
26 If (valence is very high), [T.sub.26]
then (fun is high)
27 If (arousal is mid-high) and [T.sub.27] AND
(valence is mid-low), then
(challenge is low)
28 If (arousal is mid-high) and [T.sub.28] AND
(valence is mid-high), then
(challenge is low)
29 If (arousal is high) and [T.sub.29] AND
(valence is mid-low), then
(challenge is medium)
30 If (arousal is high) and [T.sub.30] AND
(valence is mid-high), then
(challenge is medium)
31 If (arousal is very high) and [T.sub.31] AND
(valence is mid-low), then
(challenge is high)
32 If (arousal is very high) and [T.sub.32] AND
(valence is mid-high), then
(challenge is high)
33 If (arousal is mid-low) and [T.sub.33] AND
(valence is mid-low), then
(boredom is low)
34 If (arousal is mid-low) and [T.sub.34] AND
(valence is low), then
(boredom is medium)
35 If (arousal is low) and [T.sub.35] AND
(valence is low), then
(boredom is medium)
36 If (arousal is low) and [T.sub.36] AND
(valence is mid-low), then
(boredom is medium)
37 If (arousal is mid-low) and [T.sub.37] AND
(valence is very low), then
(boredom is high)
38 If (arousal is low) and [T.sub.38] AND
(valence is very low), then
(boredom is high)
39 If (arousal is very low) and [T.sub.39] AND
(valence is very low), then
(boredom is high)
40 If (arousal is very low) and [T.sub.40] AND
(valence is low), then
(boredom is high)
41 If (arousal is very low) and [T.sub.41] AND
(valence is mid-low), then
(boredom is high)
42 If (arousal is mid-high) and [T.sub.42] AND
(valence is mid-low), then
(frustration is low)
43 If (arousal is mid-high) and [T.sub.43] AND
(valence is low), then
(frustration is medium)
44 If (arousal is high) and [T.sub.44] AND
(valence is low), then
(frustration is medium)
45 If (arousal is high) and [T.sub.45] AND
(valence is mid-low), then
(frustration is medium)
46 If (arousal is mid-high) and [T.sub.46] AND
(valence is very low), then
(frustration is high)
47 If (arousal is high) and [T.sub.47] AND
(valence is very low), then
(frustration is high)
48 If (arousal is very high) and [T.sub.48] AND
(valence is very low), then
(frustration is high)
49 If (arousal is very high) and [T.sub.49] AND
(valence is low), then
(frustration is high)
50 If (arousal is very high) and [T.sub.50] AND
(valence is mid-low), then
(frustration is high)
51 If (valence is very low), then [T.sub.51]
(fun is very low) and
(challenge is very low)
52 If (valence is low), then [T.sub.52]
(fun is very low) and
(challenge is very low)
53 If (valence is high), then [T.sub.53]
(challenge is very low) and
(boredom is very low) and
(frustration is very low)
54 If (valence is v very high), then [T.sub.54]
(challenge is very low) and
(boredom is very low) and
(frustration is very low)
55 If (valence is mid-high), then [T.sub.55]
(boredom is very low) and
(frustration is very low)
56 If (arousal is very low), then [T.sub.56]
(challenge is very low) and
(frustration is very low)
57 If (arousal is low), then [T.sub.57]
(challenge is very low) and
(frustration is very low)
58 If (arousal is mid-low), then [T.sub.58]
(challenge is very low) and
(frustration is very low)
59 If (arousal is mid-high), then [T.sub.59]
(boredom is very low)
60 If (arousal is high), then [T.sub.60]
(boredom is very low)
61 If (arousal is very high), then [T.sub.61]
(boredom is very low)
62 If (arousal is very low) and [T.sub.62] AND
(valence is mid-high), then
fun is very low)
63 If (arousal is low) and [T.sub.63] AND
(valence is mid-high), then
(fun is very low)
64 If (arousal is very low) and [T.sub.64] AND
(valence is high), then
(fun is low)
65 If (valence is mid-low), then [T.sub.65]
(fun is very low)
66 If (arousal is very low) and [T.sub.66] AND
(valence is high), then
(boredom is low)
67 If (arousal is low) and [T.sub.67] AND
(valence is mid-high), then
(boredom is low)
68 If (arousal is very low) and [T.sub.68] AND
(valence is mid-high), then
(boredom is medium)
69 If (arousal is very high) and [T.sub.69] AND
(valence is very low), then
(challenge is medium)
70 If (arousal is very high) and [T.sub.70] AND
(valence is mid-high), then
(challenge is medium)
71 If (arousal is high) and [T.sub.71] AND
(valence is low), then
(challenge is low)
72 If (arousal is high) and [T.sub.72] AND
(valence is high), then
(challenge is low)
73 If (arousal is very high) and [T.sub.73] AND
(valence is low), then
(challenge is high)
74 If (arousal is very high) and [T.sub.74] AND
(valence is high), then
(challenge is high)
75 If (arousal is mid-high) and [T.sub.75] AND
(valence is mid-high), then
(excitement is low)
76 If (arousal is high) and [T.sub.76] AND
(valence is mid-high), then
(excitement is medium)
77 If (arousal is high) and [T.sub.77] AND
(valence is high), then
(excitement is medium)
78 If (arousal is mid-high) and [T.sub.78] AND
(valence is high), then
(excitement is medium)
79 If (arousal is very high) and [T.sub.79] AND
(valence is mid-high), then
(excitement is high)
80 If (arousal is very high) and [T.sub.80] AND
(valence is high), then
(excitement is high)
81 If (arousal is very high) and [T.sub.81] AND
(valence is very high), then
(excitement is high)
82 If (arousal is high) and [T.sub.82] AND
(valence is very high), then
(excitement is high)
83 If (arousal is mid-high) and [T.sub.83] AND
(valence is very high), then
(excitement is high)
84 If (arousal is mid-low), then [T.sub.84]
(excitement is very low)
85 If (arousal is low), then [T.sub.85]
(excitement is very low)
86 If (arousal is very low), then [T.sub.86]
(excitement is very low)
87 If (valence is very low), then [T.sub.87]
(excitement is very low)
88 If (valence is low), then [T.sub.88]
(excitement is very low)
89 If (valence is mid-low), then [T.sub.89]
(excitement is very low)
