Lessons from the MOnarCH Project

ao Silva Sequeira

and Isabel Aldinhas Ferreira

IST/ISR, Universidade de Lisboa, Lisbon, Portugal

Centro de Filosoﬁa da Universidade de Lisboa, Universidade de Lisboa, Lisbon, Portugal

Keywords:

Human-Robot Interaction, Social Robotics.

Abstract:

The paper describes the main conclusions issued from three years of development, of which approximately

eighteen months of observations, of an autonomous social robot interacting with children and adults in the

Pediatrics ward of an Oncological hospital. After this long period of trials and continuous interaction, the

integration of the robot in that particular social environment can be considered highly successful. The results

taken from all the long run experiments yield valuable lessons in what relates social acceptance and user

experience to be considered in the case of robot deployment at institutions or even at households. These are

detailed in the paper.

1 INTRODUCTION

The main goal of the MOnarCH project

is to test the

integration of robots in a social environment, assess-

ing the relationships established between humans and

robots and identifying guidelines for developments to

be incorporated in future social robots.

The project is evolving in the Pediatrics ward on

an Oncological hospital

where the robot engages

in edutainment activities with the inpatient children.

Though this environment poses no special difﬁculties

on the locomotion of a wheeled robot, it can be chal-

lenging in what concerns obstacle avoidance as the in-

habitant children can use the space of the ward to play

with bulky toys and tend to be naturally entropic.

The robot is shown in Figure 1. It is about the

height of an 8-10 year old child, capable of omnidi-

rectional motion with linear velocities up to 2.5 m/s,

thus capable to walk along a person at fast pace. The

volume of the robot is compatible to that of a child

companion.

There are a number of factors difﬁcult to quan-

tize when dealing with inpatient children. Nowa-

days there is a wide variety of toy robots commer-

cially available, some with sophisticated interfacing

skills. Nevertheless, the great majority of children

had never been close to a robot this size. Even though

many of them already know what a robot is and have

clear prototypical images of a robot (see (Sequeira

www.monarch-fp7.eu

IPOLFG, in Lisbon.

Figure 1: The MOnarCH robot (source: SIC TV Network).

and Ferreira, 2014)) this is a perception obtained from

movies, books, toys, etc. At the hospital it has been

observed that children often show surprise when they

see the robot for the ﬁrst time moving autonomously.

This initial perception may have a lasting effect and

inﬂuence future behaviors (see for instance (Boden,

2014)).

This paper does not report results supported on

rigid observation methodologies as often used in the

literature, with Likert questionnaires and statistical

processing of the results (preliminary analysis can be

found in project MOnarCH Deliverables, (MOnarCH

Project Consortium, 2015; MOnarCH Project Con-

Sequeira, J. and Ferreira, I.

Lessons from the MOnarCH Project.

DOI: 10.5220/0005998102410248

In Proceedings of the 13th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2016) - Volume 1, pages 241-248

ISBN: 978-989-758-198-4

241

sortium, 2016)). The main reason is that such ap-

proach can disturb the normal operation of the envi-

ronment (the Pediatrics ward in this case) and bias

any results (the argument in (Boden, 2014) on the im-

portance of initial perceptions can also be used here).

Instead, the paper is supported on direct observations

of all the actors in the environment. The paper’s con-

clusions are grounded on almost 18 months of intense

observation of the reactions of children and adults at

the hospital with the robot evolving through different

stages of development. In a sense, it can be assumed

that multiple inpatient children with essentially simi-

lar social behaviors were interacting, directly and in-

directly, with the robot. This number is estimated in

the order of hundreds. Therefore the number of chil-

dren that knew about the robot is large enough for the

observations to have some signiﬁcance.

We claim that social experiments involving robots

must unfold with the dynamics of the social environ-

ment itself. Otherwise there is a risk of biasing the re-

sults. Rigid planning of experiments, as often done in

lab environments often cannot be used. In fact, while

preparing an experiment a team is changing the en-

vironment and potentially biasing the reaction of the

inhabitants.

The paper is organized as a collection of the main

lessons learned, one per section, without any rank-

ing concerns. All the lessons are equally important.

Even though some of these lessons are well known of

Robotics developers, it is worth to recall them. More-

over, these lessons are not intended to be a method-

ological approach to project development.

2 DESIGNING THE

EXPERIMENTS

The hospital environment is socially constrained often

to a high degree. Besides the ethical and legal regula-

tions, there are a large number of unwritten rules (the

social rules) the robot should comply with to maxi-

mize the acceptance by the inhabitants of the social

environment, i.e., children, parents, relatives and vis-

itors, and staff. Moreover, some of these rules may

change in time depending, for example, on the occu-

pancy of the ward.

The routines of the hospital cannot be disturbed at

any time and even though the robot is constantly un-

der development, experiments must occur only when

the adequate conditions in the social environment

emerge. This means that seldom the experiments can

be planned. Instead, the robot must match its capabil-

ities to the current state of the environment and gener-

ate observations from which relevant conclusions can

be withdrawn. In a sense, experiments must occur as

the (uncontrolled) events arise. This randomness al-

lowed nonetheless a rough scheduling for classes of

experiments.

The robot design followed principles also recom-

mended by other authors, (see for example (Kim,

2007)) namely that of using test frameworks and en-

sure that the project team have a thorough understand-

ing of the HRI interfaces, of the environment, and of

the inﬂuence of design aspects. In MOnarCH, besides

the testing in the hospital, numerous tests of isolated

components were conducted i lab testbeds at some of

the partners of the project.

The robot was ﬁrst deployed at the environment

mainly in a static exhibition, most of the time being

kept inside a room but being available for the chil-

dren to see upon request to the staff. This ﬁrst period

lasted for around 5 months, during which the robot

was taken out in order for the team to assess different

functionalities, e.g., the navigation system.

The following period lasted for approximately 7

months during which the core functionalities of the

system were developed, namely the games the robot

can play with the children and the communicative

actions and expressions used for human-robot inter-

action. The robot was shown in a normal develop-

ment cycle (sometimes failing) and the children could

watch it during the test runs.

During the last period the robot was used in semi

and full autonomy, wandering around the ward, play-

ing with the children in some occasions. The tun-

ing of the communicative actions and expressions is

currently taking the majority of the effort. This is a

lengthy process, requiring the careful observation of

the environment and trial runs to assess the efﬁcacy of

the interactions.

The Wizard-of-Oz (WoZ) technique (Green et al.,

2008) has proved highly valuable in early stages,

namely as it allowed the testing of a number of inter-

actions (communicative acts in the project parlance,

see (Alonso-Martin et al., 2013)) for later use. In

these experiments, a Wizard controlled some of the

verbal capabilities of the robot and could decide nav-

igation goals. The Wizard role was played by a

project team member located remotely but maintain-

ing a good perception of the environment. This form

of WoZ usage, with real technology in a real environ-

ment, has been reported to be seldom used in research,

(Riek, 2012).

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

242

3 SOFTWARE

The underlying structure of MOnarCH is a network

of software components. It can be assumed that each

of them was developed independently and that the

communication among them is based on some pre-

deﬁned protocol. The dynamics of each of the soft-

ware components is of major importance to the over-

all stability and performance. A component that does

not clearly broadcast its state can prevent other com-

ponents to run properly. Also, the handling of ex-

ception conditions must be carefully deﬁned, e.g.,

transitions between states must occur cleanly. In a

sense this amount to “controllability”and “observabil-

ity” concerns on the overall system (the relevance

of which has been recognized by some authors, e.g.,

(Naganathan and Eugene, 2009)).

The visible part of the system is composed by the

outer shell of the robot (Figure 1), the corresponding

Human-Robot Interaction (HRI) interfaces, and sen-

sors. The HRI interfaces must be used much in a hu-

man way, meaning that visual effects must be well

coordinated with sound effects so that the observers

get the correct perception and the interaction is efﬁ-

cient. The MOnarCH robot has two 1-dof arms and

a 1-dof neck, variable luminosity eyes and cheeks, a

LED matrix for a mouth, loudspeakers, microphone,

bumpers, touch sensors placed in strategic places of

the body shell, and a RFID tag reader. The adequate

coordination of HRI interfaces can be framed as a sta-

bility problem.

In MOnarCH, a key architectural feature that ac-

counts for this concern is called Situational Aware-

ness Module (SAM). In a sense this is a routing in-

formation matrix that deﬁnes what are the informa-

tion exchanges, and, if necessary, applies basic trans-

formations to that information (see (Messias et al.,

2014)).

Moreover, performance and stability concerns

have also been addressed through the inclusion of

watchdog mechanisms (see (MOnarCH Project Con-

sortium, 2016)).

The ROS middleware (www.ros.org) supports the

integration of all components. Key features of the de-

velopment are (i) local networking capabilities with

distributed node representation, (ii) global variables

accessible from any node in the system (the topics

in the ROS middleware), (iii) global networking is

useful for monitoring, (iv) fully independent control

of each component, (v) callback mechanisms to deal

with asynchronous events (actionlib), and (vi) state

machine frameworks (smach) for fast behavior devel-

opment, among others. Structures like the SAM mod-

ule can be easily implemented in the ROS middleware

using the built-in facilities. In a sense, a SAM is a col-

lection of direct mappings between ROS topics.

4 MOTION

Motion is the most basic tool for a robot to interact

with humans. In the particular case of inpatient chil-

dren (in the Pediatrics age range velocity and accel-

eration must be carefully selected. Relatively small

accelerations can be perceived by children as menac-

ing. Similarly, fast decelerations can be perceived as

if the robot is untrusty.

Obstacle avoidance is tightly connected to mo-

tion. Real environments can be really challenging,

e.g., children can move in unpredictable ways (see

Figure 2), and collisions are to be expected. When-

ever a robot collides it is a good practice that it clearly

shows that it knows a collision just occurred. A prac-

tical way is to issue a non-verbal sound that can easily

be associated to a collision, e.g., ouch.

The space in the Pediatrics ward is reasonably

structured (see the pictures presented in the paper).

Using an a priori map built with the gmapping ROS

package, (http://wiki.ros.org/gmapping, 2016), is ob-

tained by teleoperating the robot such that all the ward

is covered it is possible to localize the robot only with

the information from the laser scans. The naviga-

tion system in MOnarCH is based on a potential ﬁeld

approach combining a fast marching method to con-

struct an optimal path to the goal and the repulsive

ﬁelds from obstacles, (Ventura and Ahmad, 2014).

Localization is based on the AMCL package, avail-

able in ROS, (http://wiki.ros.org/amcl, 2016), and

uses information from two laser range ﬁnders that

scan the full space around the robot. The accuracy of

the pose estimate is in the order of a few cm and de-

grees and the system has been show to have a remark-

able robustness. Observations so far indicate that, on

average, the robot loses its localization less than once

a day. When this happens the unusual behavior of the

robot is enough for an external observer to recognize

it and either alert the project team or push the robot to

a speciﬁc location where it can auto-localize easily.

In addition to the motion of the body of the robot,

the 1-dof neck can rotate right/left and it conveys

quite effectively the idea that the robot has a clear fo-

cus of attention. If the head turns because a person

is detected on that side a child tends to conclude that

the robot knows where the person is. If the head turns

to the opposite side where a person is then a plausi-

ble perception is that the robot does not care about the

person and has some other focus of attention. In both

cases the observations suggest that the motion of the

Lessons from the MOnarCH Project

243

(a)

(b)

Figure 2: Children running a toy bike chasing the robot; the

robot must account as much as possible for sudden devia-

tions of the straight line path by the child.

neck is a relevant feature for human-robot interaction.

Also, the two 1-dof arms are used to improve live-

ness. A simple balancing of the arms while the robot

is moving can easily convey the perception that the

robot has a focus while moving or a goal location

where to go.

5 HUMAN-ROBOT

INTERACTION INTERFACES

A basic speech interface is used to convey either ver-

bal or non-verbal sounds. Tone and volume are highly

relevant for a sound message to be perceived properly.

Short verbal and non-verbal sounds can be used

to show liveness. The trigger of such sounds can be

done following an exponentially distributed process,

with the corresponding parameter eventually tuned to

match the environment.

Experiments performed with off-the-shelf Text-

To-Speech systems using adult voices have been

shown to trigger negative reactions by some children,

namely of very young age (even if the sentences are

not aggressive). As an alternative, pre-recorded child-

ish tone speech has been very well received both by

children and adults.

Off-the-shelf automatic speech recognition (ASR)

has been also tested (using for example the

SpeechRecognition-3.4.6 Python based package)

Reasonable recognition rates (from the perspective of

basic HRI) can be achieved when sentences are spo-

ken with clear diction which is often not the case with

children. This reduces the range of interactions that

an be supported on ASR.

A touchscreen in the chest is useful to (i) quickly

convey information on the “emotional state” of the

robot through emoticon-like graphics. Children tend

to be curious about the meaning of the graphics shown

and quickly come up with their own interpretation,

which in a sense amounts to the establishing relation

with the robot.

Touch sensors are placed in various locations be-

low the outer shell. These are non-visible interfaces.

This type of interface tends to be effective with young

children as they are more prone to engaging in touch-

ing behaviors.

Facial expressions, generated by the LED mouth,

eyes, and cheeks, tend to be interpreted as indicators

that the robot is active (often in conjunction with the

graphics displayed in the chest screen).

6 CHILDREN’S REACTIONS

Though the Pediatrics ward covers the full range of

Pediatric ages, the observation showed that ages in

the range 3-8 are the most interested. Moreover, chil-

dren with mild degrees of Asperger syndrome are also

highly receptive to robot interactions. Also, children

love to be recognized by name by the robot. Wearing

a ﬂexible RFID tag attached to the clothes allows the

robot to recognize the children and greet them occa-

sionally.

Children above 10-12 years old may look at the

robot but seldom initiate an interaction. They often

direct gaze to the robot but rarely touch it.

During the initial stages of the project the robot

was deployed at the hospital and when no team mem-

ber was around it was kept in a separate room, with

reserved access. The inpatient children know the re-

served location of robot and often ask the staff to go

there to see it, just to see the robot sleeping.

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

244

Currently the robot is kept in a public space, that

is, a docking station that is also used to charge bat-

teries. During the periods at the docking station (the

sleeping periods) the robot maintains a small degree

of liveness, reacting to touch stimulae. Even in this

almost static display children often look for the robot

either to just see it or to touch it and observe the reac-

tion induced by the liveness module.

7 TEAM MEMBERS

The team members working at the environment must

be integrate themselves socially to avoid becoming

biasing/disturbing/distracting factors. Successful in-

tegration smooths potential social barriers that some

people may rise against the robot. In fact this strategy

resulted from a hard constraint imposed by the hos-

pital from the very beginning of the project, i.e., the

hospital should continue to operate normally and not

be transformed in a robotics lab, with the team mem-

bers disturbing the ward.

The integration of the team members occurred

through multiple forms, e.g., explaining the robot to

parents/visitors/staff/children and bringing the robot

to the rooms upon requests by inpatient children with

restricted mobility (see ﬁgure 3).

8 PARENTS AND VISITORS

REACTIONS

Parents and visitors were informed about the project

through postcards visible in the walls of the Pediatric

ward. The observations clearly show that parents are

curious about the robot. Direct interactions do occur,

with parents and visitors often looking at the robot

while moving in the ward, take pictures, and even

touching the robot.

As the autonomy of the robot increased parents

and visitors started to become conﬁdent asking the

team to adjust the behaviors of the robot. A typi-

cal example is the adjustment of the snoring behavior

the robot exhibits when sleeping, or in the non-verbal

sounds the robot occasionally issues.

9 STAFF’S REACTIONS

Staff was given initially basic information on the

project and on the robot, and on actions to adopt in

case of the robot is disturbing the normal operation of

the ward. A basic instruction was that the robot has

(a)

(b)

(c)

Figure 3: Team members and hospital staff assisting chil-

dren during interactions.

Lessons from the MOnarCH Project

245

an emergency button that can be pushed at any time

to stop it and allow any staff member to push it away.

During the whole period no observations of any staff

member pushing the button were made.

In fact, the staff adapted to this additional mem-

ber that is often seen moving along the main corridor

of the Pediatrics ward. Figure 4 shows a staff mem-

ber pushing a cart and deviation from the robot. The

staff knows that the robot is capable of deviating from

static and moving obstacles but does not wait and per-

forms the evasive maneuver

(a)

(b)

Figure 4: Staff adapting to the behavior of the robot (acting

compliantly).

In some occasions it has been observed that some

staff members like to test the obstacle avoidance ca-

pabilities of the robot, usually for the fun of it, and

try to verify the conditions under which the robot gets

trapped.

Similarly to children, a number of staff members

appreciates being recognized by the robot. RFID tags

worn in the service uniforms allow the robot to occa-

sionally greet the members.

10 METRICS AND ASSESSMENT

A number of metrics is being used to assess the

project. These are based on the activation rates of se-

lected micro-behaviors (units of activations per sec-

ond), annotated from the observations of the exper-

iments (this is a common strategy used by multiple

authors, e.g., (Mead et al., 2011; Scassellati, 2005;

Dautenhahn and Werry, 2002).

Table 1 shows results of a recent experiment at the

hospital. The experiment took place over a period of

two consecutive days.

Table 1: Activation rates (×10

−2

) for a typical experiment

(total duration of 5685 seconds over 2 consecutive days and

a total of 249 micro-behaviors annotated, units are activa-

tions per second).

Micro-behaviors

Activation

rates (robot

showing low

liveness)

Activation

rates (robot

showing high

liveness)

1. Looking to-

wards the robot,

without moving

1.0 0.86

2. Looking

towards the robot

and moving

(around, ahead

and/or at the back

of the robot)

0.64 0.78

3. Touching the

robot

0.16 0.16

4. Aggressive

movement

towards the robot

0.03 0

5. Ignoring the

robot

1.9 1.1

6. Following the

robot

0.32 0.23

7. Compliant

behavior towards

the robot

0.61 0.82

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

246

The values emphasize the importance of using

high levels of liveness. For example, the values sug-

gest that an intention to initiate an interaction may be

bigger if the robot shows that it is alive (lines 2 and 5).

Also, if the robot shows aliveness aggressive micro-

behaviors may be less likely (line 3). The values in

line 6 can be interpreted as a result of curiosity to un-

derstand the behavior of the robot when the liveness

shown is not enough to decide about interacting with

it.

Complex metrics can be deﬁned over the activa-

tion rates of collections of micro-behaviors. For ex-

ample, the acceptance of the robot can be measured

by comparing a combination of micro-behaviors 2,

3, 6, 7 against a combination of micro-behaviors 1,

4, 5. Designing these combinations will be in gen-

eral somewhat subjective but it may provide nonethe-

less useful guidance to adjust the behaviors of the

robot (see (MOnarCH Project Consortium, 2015;

MOnarCH Project Consortium, 2016) for additional

details).

Manual annotations are often subject to large er-

rors. To achieve statistical signiﬁcance the annota-

tions must (i) focus on experiments made in similar

conditions, namely in what concerns the robot and

the characteristics of the population interacting with

it, and (ii) annotated by multiple persons (or multi-

ple systems in case of automatic annotation). This is

often difﬁcult due to the constantly changing of the

inpatient population. For example, there are periods

where most of the children is in isolation rooms (and

hence cannot interact with the robot) and periods in

which many children are allowed to leave the rooms

(thus being able to play with the robot). This means

that either the experiments take place in close days, or

it may be necessary to wait for the right conditions to

be present.

Independently of the details on how the activa-

tion rates are combined, it seems clear that activa-

tion rates provide interesting quantitative information.

The use of automated system to annotate the micro-

behaviors could further provide, for example, the time

between activations of micro-behaviors and their du-

ration, which are necessary for a complete study of

their probabilistic nature. The use of such systems,

e.g., (www.noldus.com, 2016), is out of the possibili-

ties of the project.

11 CONCLUSIONS

It is commonly accepted that a real social environ-

ment is substantially different from common labora-

tory environments. Achieving successful integration

(a)

(b)

Figure 5: Snapshots acquired during the liveness experi-

ments.

of a robot in a real social environment, such as the

Pediatrics ward of the MOnarCH project has been

shown to depend on multiple factors, some outside the

pure engineering scope. Moreover, current off-the-

shelf and state-of-the-art robotics seem well adapted

to the objectives of the project, though technological

improvements may simplify similar systems. There

is no evidence that the lessons learned depend on the

speciﬁc type of robot used (wheeled, omnidirectional)

and hence may be useful in projects involving other

types of social robots.

Overall, the high acceptance of the robot by every-

one at the hospital is a clear indicator that the strategy

Lessons from the MOnarCH Project

247

that led to the lessons in the paper was successful.

A quantitative assessment in medical terms of a

robot such as the one developed in the MOnarCH

project may take several years and hence it out of the

scope of the project. Meanwhile, all the indicators

observed in MOnarCH suggest that social robots may

have a highly positive role in socially difﬁcult envi-

ronments.

ACKNOWLEDGEMENTS

Work supported by projects FP7-ICT-9-2011-601033

(MOnarCH) and FCT [UID/EEA/50009/2013].

REFERENCES

Alonso-Martin, F., Malfaz, M., Sequeira, J., Gorostiza, J.,

and Salichs, M. (2013). A Multimodel Emotion De-

tection System during Human-Robot Interaction. Sen-

sors Journal, 13:15549–15581.

Boden, M. (2014). Aaron Sloman: A Bright Tile in AI’s

Mosaic. In Jeremy L. Wyatt and Dean D. Petters and

David C. Hogg, editor, From Animals to Robots and

Back: Reﬂections on Hard Problems in the Study of

Cognition. Springer. Cognitive Systems Monographs

22.

Dautenhahn, K. and Werry, I. (2002). A Quantitative Tech-

nique for Analysing Robot-Human Interactions. In

Procs. of the 2002 IEEE/RSJ Int. Conf. on Intelligent

Robots and Systems. Switzerland.

Green, S., Richardson, S., Stiles, R., Billinghurst, M.,

and Chase, J. (2008). Multimodal Metric Study for

Human-Robot Collaboration. In Procs. 1st Int. Conf.

on Advances in Computer-Human Interaction, pages

218–233. Sainte-Luce, Martinique, 10-15 February.

http://wiki.ros.org/amcl (2016). Adaptive Monte-Carlo Lo-

calization. Accessed April 2016.

http://wiki.ros.org/gmapping (2016). Gmapping. Accessed

April 2016.

Kim, M. (2007). The HRI Experiment Framework for De-

signer. In Procs. of The 16th IEEE International Sym-

posium on Robot and Human Interactive Communica-

tion (RO-MAN 2007). Jeju, Korea, 26-29 August.

Mead, R., Atrash, A., and Mataric, M. (2011). Prox-

emic Feature Recognition for Interactive Robots: Au-

tomating Metrics from Social Sciences. In In Social

Robotics, volume 7072 of Lecture Notes in Computer

Science, pages 52–61. Springer.

Messias, J., Ventura, R., and Lima, P. (2014). The

MOnarCH Situation Awareness Module. Project

MOnarCH deliverable document available at

http://www.monarch-fp7.eu.

MOnarCH Project Consortium (2015). Project Deliverable

D8.8.4 - Long-Run MOnarCH Experiments at IPOL.

http://users.isr.ist.utl.pt/∼jseq/MOnarCH/Deliverables/

D8.8.4 ﬁnal.pdf.

MOnarCH Project Consortium (2016). Project Deliver-

able D8.8.5 - Experiments on the MOnarCH Mixed

Human-Robot Environment. http://users.isr.ist.utl.pt/

∼jseq/MOnarCH/Deliverables/D8.8.5-v1c.pdf.

Naganathan, E. and Eugene, X. (2009). Software Stability

model (SSM) for Building Reliable Real time Com-

puting Systems. In Procs. of the Third IEEE Interna-

tional Conference on Secure Software Integration and

Reliability Improvement (SSIRI 2009). July 8-9.

Riek, L. (2012). Wizard of Oz Studies in HRI: A Systematic

Review and New Reporting Guidelines. Journal of

Human-Robot Interaction, 1(1):119–136.

Scassellati, B. (2005). Quantitative Metrics of Social Re-

sponse for Autism Diagnosis. In Procs of ROMAN’05.

Sequeira, J. and Ferreira, I. (2014). The concept of [Robot]

in Children and Teens: Some Guidelines to the Design

of Social Robots. International Journal of Signs and

Semiotic Systems, 3(2):35–47.

Ventura, R. and Ahmad, A. (2014). Towards Optimal Robot

Navigation in Domestic Spaces. July 25 , Joao Pessoa,

Brazil.

www.noldus.com (2016). Innovative solutions for behav-

ioral research. Accessed March 2016.

ICINCO 2016 - 13th International Conference on Informatics in Control, Automation and Robotics

248