The document discusses using an interactive robot to intelligently monitor patients' health at home. It describes developing a system for the robot to understand contexts in order to determine the right time each day to ask a patient to complete a symptoms questionnaire. The system considers data from sensors and parameters like a patient's presence and facial recognition. Implementing this system on a NAO robot and enhancing its features improved the system's average precision from 76.8% to 18.2% higher. Future research will focus on understanding context within the semantic web to further increase effectiveness.

1. Context Understanding for Medico-Social
Assistance with an Interactive Robot
W. BEN GHEZALA, G. BREDA, L. DEVIGNE, P. BAZIN, N. BEAUMATIN
ALTRAN Technologies, firstname.lastname@altran.com
2 Rue Paul DAUTIER, 78160 Vélizy-Villacoublay
Introduction
The aging population and the increase in chronic diseases stretch to
increase care and health spending needs. According to INSEE [1], the
medical population will decrease by 10% until 2019 before returning to its
current level in 2030. New solutions must therefore be found to optimize
resources and health expenditures, and improve prevention, care and
monitoring of patients and vulnerable people throughout their course of
social and professional life. Most current IT medical monitoring systems are
based on technological platforms consisting of information systems (e.g.
patient record management information systems) and connected objects
(e.g. communicating medical devices). They aim to provide better
professionals’ coordination and prevent the degradation of patients’ health
condition.
However they lack of intelligence and interactivity to personalize the
relationship with the patient.
The scope of this article hence generally addresses the technical
feasibility of using an interactive “human-friendly” robot to perform
intelligent medical monitoring of people. More precisely, we got interested
here on the feasibility of making the robot able to find the appropriate
situation for patient’s completion of a symptoms’ questionnaire. Indeed this
is a common function for a number of cases where there is a need to
monitor patients’ symptoms at a distance (for example in the case of home
chemotherapy).
In this paper we will first present our analysis of the existing robots that
are tested or used with the elderly and/or sick people. Afterwards, we will
describe our system of context understanding that aims to find the right time
for completing a symptoms’ questionnaire. Finally, the implementation of
this system on the NAO robot and the main results will be exposed.

2. Robotic systems for medical monitoring
Service robotics is one of the most growing businesses. It aims to help
people in their daily lives and to provide basic services such as transport,
cleanliness, safety, care and support. Indeed, the development of assistive
robotics is an emerging field that could help to bring this interactive
dimension, smart and easy to access. These robots are designed to provide
support services to daily life at home (reminder for taking medication,
making meals, porterage, security ...) and / or have a role of companions to
preserve the autonomy and quality of life at home. They may have
humanoid or animal varied forms and are designed to interact with humans
via various interfaces: touch, kinesthetic, sensory, emotional, cognitive and
socio-behavioral. Some are even called emotional robots because they are
likely to trigger positive emotions. These emotional or social robots,
animaloid-form like the Paro robot [2], are opening interesting perspectives
of multimodal management of people with cognitive dementias such as
Alzheimer's disease. Indeed many studies indicate that the use of these
robots as a means of communication can promote social interaction and
soothe some behavioral disorders in these patients [3]. Other home care
robots1
are more focused on social telepresence like the BEAM robot.
However the detection of emotions includes universal emotions (happiness,
surprise, anger, sadness and doubt) but not anxiety / stress, and generally do
not combine physiological approaches. Moreover, robots are usually bulky
and their visual appearance is sometimes a bit scary, and so no "human-
friendly", even if left to the subjectivity of each observer. Moreover, the
home deployment of these robots with human size such as Kompaï, poses
major technical problems far from being solved [8]. Indeed, despite the
promises offered by home care robots, over 40% of developments are
stopped for lack of understanding of socio-technical factors. These factors
are determining in the adoption of such robots. In [4] the authors find that
intends of use depend on social influence, expected performance,
confidence issues, privacy, and ethical concerns. Among the determinants,
social influence is the strongest predictor. In addition, monitoring of vital
signs, easy communication with family, and recall of medication are the
most requested applications by respondent people (recruited in health
service companies).
This state of the art shows that tele-monitoring robots on the market and
under development have varying features and different aspects. Few of
them are now connected to a sensors’ environment.
1
Home health care robots, i.e. making clinical information available at the right time and the right place to
reduce the risks of error, increase the safety and quality of care.

3. Home monitoring features that help observance does not cover the
understanding of several context dimensions to ask symptoms’
questionnaires, which is one of the main preoccupation of professionals
when patients are at home, and which is likely to allow better anticipating
emergency situations.
Context understanding
We then designed, developed and tested a system to understand the context
in order to allow the robot to ask a questionnaire of symptoms to the patient
every day. This task is most of the time annoying for the patient. Indeed, if
the robot asks questions to the patient at any time of the day, it can quickly
be perceived as intrusive and disturbing instead of motivating and
entertaining the patient. To make it less painful and more fun, the robot
should determine when the patient is willing to answer the questionnaire
and send it to his doctor. In addition, each patient is different; the robot
must be able to adapt its decision based on his experience with a particular
patient, and to several parameters.
Here we have considered a context taking into account the data of an
activity/temperature sensor (MOVISENS [5]) worn by a patient undergoing
chemotherapy at home [6], but also parameters such as the presence of the
patient near the robot, facial recognition, and time (calendar concept).
Figure 1 : Examples of assistive robots and virtual agents (left to right):
NAO, Care-O-Bot, SAM, PR2, Twendy one, Giraff, Kompai, Asimo, Paro

4. Figure 2 : State machine for context understanding
The system was designed as a state machine as shown in Figure 2.
Each state of this machine is a robot feature that will be implemented in the
comprehension of the context to ask the questionnaire.
The "Solitary mode" is the robot baseline mode. The robot is not in active
listening but can detect movement activity to consume less battery. This is
the initial state of our system.
Our system calls different features depending on the transitions described in
Figure 2. Hence, the response time and the effectiveness of our system
depends on the response times and efficiency of the different features.
Main results
We have established an implementation choice on the robot NAO because it
presents a "human-friendly" interface. To do so, we have improved the
performance of various features used in the state machine designed to be
able to be integrated into the robotic system. The implementation of our
system was made within the robotics software of Aldebaran, called
Choregraphe [7]. In this software we have designed and integrated the
blocks corresponding to different features related to each other as described
in the application's design of context understanding. An improvement of the
effectiveness of each feature has been performed beforehand. The
improvement was realized independently on each feature [9]. As an
example of improvement, we have significantly increased the accuracy of
facial recognition:
• Adding check of the variable representing the confidence level;
• Empowering of the learning repeated facial;
• Renewing the identifier of face.
In order to compare the system developed with the algorithms realizing the
original features of the robotic system (first result) and our system

5. implementing enhanced features for efficacy in the treatment (2), we
calculated the average precision with two implementations (1) and (2). We
observed a global average of system’s precision (according to the equation
below) of 76.8%, giving an improvement of 18.2%.
Of course this system has allowed us to increase the efficiency of our
system but we must elaborate further research to increase this ratio as our
context requires a response time closer to the real-time with effectiveness
convergent to 100%.
Conclusion
The subject of this article discusses a rich scientific research topic because it
addresses several axes starting from IT to health through robotics. Given the
promising results we have obtained in this article, we focus our future
research on understanding the context while building on what happens in
the semantic web.
Acknowledgment
The team thanks the EILIS division of Altran and the promoter of the work
outlined in this article. We also thank all the PiCADo project team, a French
project financed by the French Inter ministerial Fund, who contributed by
establishing the real need observed on the ground, i.e. application of the
symptoms’ questionnaire.
References
[1] BRUTEL, Chantal. Projections de population à l'horizon 2050: un vieillissement
inéluctable. 2001.
[2] Sabanovic, Selma, et al. "PARO robot affects diverse interaction modalities in group
sensory therapy for older adults with dementia." Rehabilitation Robotics (ICORR), 2013
IEEE International Conference on. IEEE, 2013.
[3] STACEY, Dawn, LÉGARÉ, France, COL, Nananda F., et al. Decision aids for people
facing health treatment or screening decisions. Cochrane Database Syst Rev, 2014, vol. 1,
no 1.
[4] ALAIAD, Ahmad et ZHOU, Lina. The determinants of home healthcare robots adoption:
An empirical investigation. International journal of medical informatics, 2014, vol. 83, no
11, p. 825-840.

6. [5] MÜLLER, Lars, RIVERA-PELAYO, Verónica, KUNZMANN, Christine, et al.From stress
awareness to coping strategies of medical staff: Supporting reflection on physiological
data. In : Human Behavior Understanding. Springer Berlin Heidelberg, 2011. p. 93-103.
[6] Maurice M, Lévi F, Breda G, Beaumatin N, Duclos A, Chkeir A, Hewson D, Duchêne J.
Innovative Project For Domomedicine Deployment, eTELEMED 2015
[7] POT, Emmanuel, MONCEAUX, Jérôme, GELIN, Rodolphe, et al. Choregraphe: a
graphical tool for humanoid robot programming. In : Robot and Human Interactive
Communication, 2009. RO-MAN 2009. The 18th IEEE International Symposium on.
IEEE, 2009. p. 46-51.
[8] DUPOURQUE, V. Kompai: home centric robot by Robosoft. Kompai pptx ebook. www.
robosoft. Com
[9] BAZIN, Paul et DEVIGNE, Louise l’élaboration d’un système d’aide médico-sociale à
l’aide d’un robot humanoïde, unpublished
Authors’ Info
Walid Ben GHEZALA, Telecom engineer with PhD in computer
sciences applied on rehabilitation robotics systems. Walid is a project
leader on robotics for e-health. Initially specialized in network security
for e-health systems, he was selected to complete his engineering
expertise in ICT for Health in Montpellier (UM2). His PhD in
computer science is made in cooperation between CEA and Telecom
SudParis (TSP) for specialization in robotics for health.
Gabrièle BREDA, Normalienne with a PhD in Neurosciences, Gabriele
is the scientific advisor of Altran Research projects related to human
health. Her main area of expertise is Domomedicine, i.e new health
solutions at home or during socio-professional activities, based on
modern technologies, aimed at providing medical progress. Since her
arrival five years ago at Altran Research, she has developed the
portfolio of R&D projects in human health with renowned partners,
which resulted in several international scientific communications.
Nicolas BEAUMATIN, Graduated from the French “Grande Ecole”
Ecole Centrale de Lyon (a selective engineering school in computer
Science), Nicolas is a system engineer and architect specialized in
technologies for health, wellness and autonomy of patients and elderly
people. His main area of expertise is Domomedicine, i.e new health
solutions at home or during socio-professional activities, based on
modern technologies, aimed at providing medical progress.