scieee Science in your language
[en] (orig)

Human-smart rollator interaction for gait analysis and fall prevention using learning methods and the i-Walker

Cortés Martínez, Atia · 2018

Abstract

The ability to walk is typically related to several bio-mechanical components that are involved in the gait cycle (or stride), including free mobility of joints, particularly in the legs; coordination of muscle activity in terms of timing and intensity; and normal sensory input, such as vision and vestibular system. A walk is composed of the stance and swing phases. The faster we walk, the shorter the stance phase will be. Thus, gait requires input from the brain, spinal cord, peripheral nerves, muscular power and joint and cardiovascular health. Because all of these systems are required to coordinate gait, gait speed is an indicator of the health of many physiological systems. At the same time, a relation between gait and cognition has been widely analysed from the medical point of view, and we can find several reviews in the literature. As people age, they tend to slow their gait speed, and their balance is also affected. Also, the retirement from the working life and the consequent reduction of physical and social activity contribute to the increased incidence of falls in older adults. Moreover, older adults suffer different kinds of cognitive decline, such as dementia or attention problems, which also accentuate gait disorders and its consequences. Assistive technologies (AT) play a key role in today's' society, especially when it comes to the older adults. ATs have enabled improvements in their Quality of Life, extending their autonomy and community living. This is important, as they can stay active safely and independently. During the last decade, research has focused on developing ATs with a sensor system integrated with the device or located in the human body. Efforts are focused especially on mobility assistance for different targets of people (visual impairment, frailty, chronic diseases or rehabilitation) and activity recognition, which could be used, for instance, to monitor elderly population living in autonomy and community-dwelling. This thesis proposes a methodology to analyse how do older adults at high risk of falling interact with a smart rollator, the i-Walker, to navigate in indoor, flat environments. The i-Walker is equipped with a set of sensors and actuators and can collect data for long periods of time (several hours). It has already been tested in post-stroke rehabilitation and fall prevention clinical trials with successful results. In this work, we present results on our approach from a narrative perspective. Results are promising since we can relate the data obtained from human-rollator interaction to clinical parameters. The machine learning approach uses the data obtained with the force sensors of the i-Walker based on the interaction of individuals of different ages and health conditions. The analysis complements our extracted gait parameters with biological and clinical data to learn new characteristics of human gait at a stride-to-stride level. We believe that users, caregivers and clinicians would benefit from the new knowledge that the i-Walker can generate from this work.

Read accessible full text

Human-smart rollator interaction for gait analysis and fall prevention using learning methods and the i-Walker

Author: Cortés Martínez, Atia
Publisher: Universitat Politècnica de Catalunya
Year: 2018
DOI: 10.5821/dissertation-2117-121026
Source: https://upcommons.upc.edu/bitstream/2117/121026/1/TACM1de1.pdf
Human-sma olla o in e ac ion o
gai analysis and all p e en ion
using lea ning me hods and
he i-Walke
A ia Co és Ma ínez
ADVERTIMENT La consul a d’aques a esi queda condicionada a l’accep ació de les següen s
condicions d'ús: La di usió d’aques a esi pe mi jà del eposi o i ins i ucional
UPCommons (h p://upcommons.upc.edu/ esis) i el eposi o i coope a iu TDX
(h p://www. dx.ca /) ha
es a au o i zada pels i ula s dels d e s de p opie a in el·lec ual
únicamen pe a usos p i a s
emma ca s en ac i i a s d’in es igació i docència. No s’au o i za
la se a ep oducció amb inali a s
de luc e ni la se a di usió i posada a disposició des d’un lloc
aliè al se ei UPCommons o TDX. No s’au o i za la
p esen ació del seu con ingu en una ines a
o ma c aliè a UPCommons ( aming). Aques a ese a de
d e s a ec a an al esum de p esen ació
de la esi com als seus con ingu s. En la u ili zació o ci a
de pa s de la esi és obliga indica el nom
de la pe sona au o a.
ADVERTENCIA La consul a de es a esis queda condicionada a la acep ación de las siguien es
condiciones de uso: La di usión de es a esis po medio del eposi o io ins i ucional UPCommons
(h p://upcommons.upc.edu/ esis) y el eposi o io coope a i o TDR (h p://www. dx.ca /?locale-
a ibu e=es) ha
sido au o izada po los i ula es de los de echos de p opiedad in elec ual
únicamen e pa a usos
p i ados enma cados en ac i idades de in es igación y docencia. No
se au o iza su ep oducción
con inalidades de luc o ni su di usión y pues a a disposición desde
un si io ajeno al se icio UPCommons No se au o iza la p esen ación de su con enido en una
en ana o ma co ajeno a UPCommons ( aming). Es a ese a de de echos a ec a an o al
esumen de p esen ación de la esis como a sus
con enidos. En la u ilización o ci a de pa es
de la esis es obligado indica el nomb e de la
pe sona au o a.
WARNING On ha ing consul ed his hesis you’ e accep ing he ollowing use condi ions:
Sp eading his hesis by he ins i u ional eposi o y UPCommons
(h p://upcommons.upc.edu/ esis) and he coope a i e eposi o y TDX (h p://www. dx.ca /?locale-
a ibu e=en) has been au ho ized by he
i ula o he in ellec ual p ope y igh s only o p i a e
uses placed in in es iga ion and eaching
ac i i ies. Rep oduc ion wi h luc a i e aims is no
au ho ized nei he i s sp eading no a ailabili y
om a si e o eign o he UPCommons se ice.
In oducing i s con en in a window o ame o eign o he
UPCommons se ice is no au ho ized
( aming). These igh s a ec o he p esen a ion summa y o he
hesis as well as o i s con en s.
In he using o ci a ion o pa s o he hesis i ’s obliged o indica e
he name o he au ho .
Human - Sma Rolla o In e ac ion
o Gai Analysis and Fall P e en ion
Using Lea ning Me hods and he
i-Walke
A ia Co ´
es-Ma ´
ınez
CS
Uni e si a Poli `
ecnica de Ca alunya - Ba celonaTECH
A hesis p oposal submi ed o he deg ee o
Ph.D. in A i icial In elligence
2018
2
Abs ac
The abili y o walk is ypically ela ed o se e al bio-mechanical componen s ha a e in-
ol ed in he gai cycle (o s ide), including ee mobili y o join s, pa icula ly in he legs;
coo dina ion o muscle ac i i y in e ms o iming and in ensi y; and no mal senso y inpu ,
such as ision and es ibula sys em. A walk is composed o he s ance and swing phases. The
as e we walk, he sho e he s ance phase will be. Thus, gai equi es inpu om he b ain,
spinal co d, pe iphe al ne es, muscula powe and join and ca dio ascula heal h. Because
all o hese sys ems a e equi ed o coo dina e gai , gai speed is an indica o o he heal h
o many physiological sys ems. A he same ime, a ela ion be ween gai and cogni ion has
been widely analysed om he medical poin o iew, and we can ind se e al e iews in he
li e a u e. As people age, hey end o slow hei gai speed, and hei balance is also a ec ed.
Also, he e i emen om he wo king li e and he consequen educ ion o physical and social
ac i i y con ibu e o he inc eased incidence o alls in olde adul s. Mo eo e , olde adul s
su e di e en kinds o cogni i e decline, such as demen ia o a en ion p oblems, which also
accen ua e gai diso de s and i s consequences.
Assis i e echnologies (AT) play a key ole in oday’s’ socie y, especially when i comes
o he olde adul s. ATs ha e enabled imp o emen s in hei Quali y o Li e, ex ending hei
au onomy and communi y li ing. This is impo an , as hey can s ay ac i e sa ely and indepen-
den ly. Du ing he las decade, esea ch has ocused on de eloping ATs wi h a senso sys em
in eg a ed wi h he de ice o loca ed in he human body. E o s a e ocused especially on mo-
bili y assis ance o di e en a ge s o people ( isual impai men , ail y, ch onic diseases o
ehabili a ion) and ac i i y ecogni ion, which could be used, o ins ance, o moni o elde ly
popula ion li ing in au onomy and communi y-dwelling.
This hesis p oposes a me hodology o analyse how do olde adul s a high isk o alling
in e ac wi h a sma olla o , he i-Walke , o na iga e in indoo , la en i onmen s. The i-
Walke is equipped wi h a se o senso s and ac ua o s and can collec da a o long pe iods
o ime (se e al hou s). I has al eady been es ed in pos -s oke ehabili a ion and all p e-
en ion clinical ials wi h success ul esul s. In his wo k, we p esen esul s on ou app oach
om a na a i e pe spec i e. Resul s a e p omising since we can ela e he da a ob ained om
human- olla o in e ac ion o clinical pa ame e s. The machine lea ning app oach uses he da a
ob ained wi h he o ce senso s o he i-Walke based on he in e ac ion o indi iduals o di e -
en ages and heal h condi ions. The analysis complemen s ou ex ac ed gai pa ame e s wi h
biological and clinical da a o lea n new cha ac e is ics o human gai a a s ide- o-s ide le el.
i
We belie e ha use s, ca egi e s and clinicians would bene i om he new knowledge ha he
i-Walke can gene a e om his wo k.
ii

Con en s
1 Mo i a ion 1
1.1 Scope o he hesis ................................ 3
1.2 Plan o he Wo k ................................. 6
2 A e iew o Gai , Cogni ion and Falls 9
2.1 Gai cha ac e is ics in elde ly popula ion ..................... 10
2.2 Ageing and Falls ................................. 13
3 A Re iew on Assis i e Technologies 17
3.1 Assis i e De ices ................................. 18
3.2 SHARE-i ..................................... 22
3.2.1 CARMEN: an ARW wi h collabo a i e con ol ............. 24
3.3 I-DONT-FALL ................................... 28
3.3.1 Fall Managemen Se ice ........................ 29
3.3.2 IDF componen s ............................. 30
3.3.3 The IDF p o ocol ............................. 31
4 The i-Walke 35
4.1 Main componen s ................................. 37
4.2 Reac i e con ol .................................. 40
4.2.1 Applying he collabo a i e con ol philosophy o he i-Walke ..... 43
4.3 The i-Walke ’s Assis i e En i onmen ...................... 43
4.3.1 The ole o he i-Walke in pos -s oke ehabili a ion .......... 44
4.3.2 I-DONT-FALL Resul s .......................... 46
4.3.3 De ec ing Walking Beha iou Pa e ns .................. 47
4.4 Summa y ..................................... 53
iii
CONTENTS
5 Clinical Tes s: design and implemen a ion o a pilo p o ocol 57
5.1 De ini ion o a p o ocol .............................. 58
5.2 P o ocol design .................................. 59
5.2.1 Baseline Pilo ............................... 59
5.2.2 Ta ge Popula ion ............................. 60
5.2.3 Clinical Scales .............................. 61
5.2.4 Ambula o y Exe cises .......................... 63
5.2.4.1 Ten Me e Walking Tes .................... 66
5.2.4.2 Timed Walking Tes s ..................... 68
5.3 Pilo s ....................................... 69
5.3.1 IDF Pilo ................................. 69
5.3.2 MAD Pilo ................................ 71
5.3.3 CVI Pilo ................................. 72
5.4 Summa y ..................................... 72
6 Me hodology 75
6.1 Gai Analysis based on Human-Rolla o In e ac ion ............... 77
6.1.1 Da a p epa a ion ............................. 78
6.1.2 Vocabula y o s ides ........................... 80
6.1.3 Clus e ing Time Se ies .......................... 83
6.1.4 Exe cises as bags-o -s ides ....................... 86
6.1.5 Clus e s abili y .............................. 87
6.2 Spa io- empo al Analysis ............................. 88
6.2.1 Desc ip i e Gai Pa ame e s ....................... 88
6.2.2 Gai Veloci y ............................... 89
6.3 Use D i ing Skills ................................ 90
6.3.1 La e ali y ................................. 91
6.3.2 Di ec i i y ................................ 93
6.4 Modelling Exe cises by Spa io-Tempo al Gai Cha ac e is ics .......... 94
6.5 Modelling Fall Risk Assessmen ......................... 96
6.6 Summa y ..................................... 97
i
CONTENTS
7 Resul s 99
7.1 Spa ioTempo al Analysis ............................. 99
7.1.1 Desc ip i e Gai Pa ame e s .......................100
7.1.2 Gai Veloci y ...............................101
7.2 Clus e ing Resul s o he Gai Analysis ......................108
7.2.1 IDF +MAD pilo s ............................108
7.2.2 CVI pilo .................................113
7.2.2.1 Fi s Scena io .........................114
7.2.2.2 Second Scena io ........................115
7.2.2.3 Thi d Scena io .........................119
7.3 CVI Clus e Explana ion om he Spa io-Tempo al Gai Cha ac e is ics ....127
7.3.1 Fi s Scena io ...............................128
7.3.2 Second Scena io .............................129
7.3.3 Thi d scena io ..............................131
7.4 Modeling Fall Risk ................................134
8 Conclusions 137
8.1 Discussion .....................................139
A Pilo P o ocol 143
A.1 Fondazione San a Lucia ..............................143
A.2 Residencia Los Nogales ..............................146
B In eg a ing he i-Walke as an in elligen se ice in a Social Ne wo k 149
B.1 A chi ec u e ....................................150
B.2 Social Ne wo k (SN) ...............................151
B.2.1 People ...................................152
B.2.2 De ices, Repo s and Messages .....................153
B.2.3 Mul i-Agen Sys em ...........................153
B.2.4 In eg a ion ................................155
B.3 Se ice Implemen a ion ..............................156
CONTENTS
C Resea ch Ac i i y 159
C.1 Eu opean P ojec s .................................159
C.2 Na ional P ojec s .................................160
C.3 Pa icipa ion in esea ch cou ses and/o semina ies ...............160
C.4 Pa icipa ion in con e ences ............................161
C.5 Publica ions ....................................162
C.6 Resea ch s ays and isi s .............................163
Re e ences 164
i
Chap e 1
Mo i a ion
Demog aphic ageing p oceeds apace in all wo ld egions, mo e apidly han i s an icipa ed
in Na ions (2003). The p opo ion o olde people as ly inc eases as mo ali y alls and li e
expec ancy inc eases. Popula ion g ow h slows as e ili y declines o eplacemen le els. La in
Ame ica, China and India a e expe iencing unp eceden edly apid demog aphic ageing. The
p opo ion o he popula ion aged 65 and o e is expec ed o iple in less de eloped coun ies
o e he nex 40 yea s, ising om 5.8 o 15% o he o al popula ion, while in he mo e de el-
oped coun ies his igu e is expec ed o g ow om 16 o 26% (an inc ease o mo e han 60%),
he ISSA epo says (Sca dino (2009)).
In nowadays ageing socie y, many people equi e app op ia ed and pe sonalised assis ance
and new echnologies o o e hem an ex ao dina y oppo uni y o pe o m hei ac i i ies o
daily li ing (ADL) and imp o e hei au onomy. A demog aphic s udy conduc ed by B aul
(2010) showed ha 56.7M people om he US (18.7% o he popula ion) had some le el o
disabili y and 38.3 million (12.6%) had a se e e impai men . Olde s udies by B aul (2005)
and B aul (2000) show ha hese numbe s a e s eadily inc easing yea a e yea , bu also ha
he majo i y o his popula ion is concen a ed on mo e olde adul s. O people aged 15 and
olde , 30.6 Million (12.6%) had di icul y wi h ambula o y ac i i ies o he lowe body and
15.2M people (6.3%) had ouble wi h cogni i e, men al o emo ional unc ioning.
In he case o he EU25, in 2011 he e we e mo e han 80 million people wi h a disabili y
in he popula ion age g oup o 16-54 yea s, and i is es ima ed ha his numbe inc eases
up o 84 o 107 million people in he Eu opean Union (de Pejil e al. (2011)). O all wo ld
egions, Eu ope has he highes p opo ion o he popula ion aged 65 o o e , a s a is ic ha
becomes mo e pessimis ic acco ding o he baseline p ojec ion o Eu os a , which shows ha
1

1. MOTIVATION
his pe cen age will almos double o mo e han 25% in he yea 2050 (WHO (2012). Besides,
li e expec ancy has con inued o ise sys ema ically in all o he EU Membe S a es in ecen
decades (Ko ze a (2015)).
Besides, in his popula ion sec o , he equency o alls inc eases wi h age and ail y le el
and a e he leading cause o unin en ional inju y (WHO (2007)). A combina ion o biological
ac o s and disease- ela ed condi ions a e he p ima y cause o mos alls among senio s. This
combina ion has se e al implica ions o he Quali y o Li e (QoL) o he elde ly popula ion:
as hey educe hei ac i i y, hey inc ease hei ail y and ea o alling while losing hei
esidual skills. This will ep esen a challenge o he public heal h sys ems ha will ha e
o ace a subs an ial socio-economic impac o deal wi h his demog aphic si ua ion. This
is al eady no sus ainable in some coun ies and will be a wo ldwide issue sho ly. One o
he p ima y objec i es o he H2020 p og am is o ocus on he analysis o he causes and
consequences o pa hologies o ind pa e ns ha will suppo ea ly de ec ion o a disease o
associa ed isks. Consequen ly, he ca e communi y could ake decisions on in e en ion and
educa ional in o ma ion o delay he physical o cogni i e decline o he elde ly and y o keep
hem independen as long as possible li ing in he communi y.
The e olu ion o ICT ools ( ega ding cos , size o a ailabili y) in collabo a ion wi h med-
ical knowledge has empowe ed he design and de elopmen o inno a i e solu ions o p o ide
ailo ed, emo e and p e en i e ca e o people wi h special needs. In pa icula , he e is an in-
c easing in e es in ambien assis ed li ing echnology, whe e indi iduals (in his case, elde ly
non-au onomous pe sons) and hei en i onmen a e equipped wi h a sys em o senso s ( om
localiza ion o bio-me ic, among o he s) ha will collec di e en ypes o measu es allowing
expe s o moni o hei ac i i ies in eal- ime, o by epo s gene a ed by he sys em. Ambi-
en assis ed li ing en i onmen s a e expec ed o gain pa icula ele ance wi h he incoming
In e ne o Things pa adigm as a ailo ed, cos -e ec i e solu ion o imp o e heal h sys ems
(Ve mesan and F iess (2013)).
Assis i e echnologies (AT) play a c ucial ole in he ca e o challenged indi iduals, such
as olde adul s o people wi h physical and/o cogni i e dys unc ion. Thei p ima y pu pose
is o main ain o imp o e indi idual’s unc ioning and independence in o de o acili a e pa -
icipa ion in he socie y and o enhance o e all well-being (WHO (2014)). ATs aim o p o-
ide assis ance o di e en so s o a ge publics, including low ision de ices, hea ing aids,
augmen a i e and al e na i e communica ion, and especially echnologies in ol ing he use
o mobile pla o ms, such as canes, scoo e s, wheelchai s and olla o s o p os heses, such as
2
1.1 Scope o he hesis
a i icial legs. Cane use has been p e alen among he elde ly o yea s, ollowed by walk-
e s (B aul (2000)). La ely, esea ch has ocused on he obo isa ion o hese de ices o assis
pe sons wi h physical and/o cogni i e disabili ies in hei ac i i ies o daily li ing (ADLs).
Independen mobili y is one o he mos c i ical ac o s in main aining he quali y o li e
o elde s, and o he clinical popula ions who need assis i e de ices, by delaying hei ins i-
u ionalisa ion. Mobili y is c ucial o pe o ming ADLs, as well as o main aining i ness
and i ali y (Alwan e al. (2007)). While obo ised wheelchai s ha e been mainly s udied o
assis in he mobili y and au onomy o he use (see §3.1), sma walke s can go a s ep u -
he acco ding o Ma ins e al. (2012): hey a e use ul o a b oade ange o use s and help in
he eco e y o ambula o y skills. Recen s udies on eal indi iduals ha e es ed he po en ial
o obo ised olla o s in he ehabili a ion p ocess o people hospi alised due o some ype o
impai men , like a s oke eco e y o a ca acciden (Giuliani e al. (2012) and Mo one e al.
(2016)). Thus sma walke s no only p omo e mobili y and na iga ion assis ance bu also can
p o ide gai moni o ing and pa ial body weigh . New ends in esea ch and u u e model
business solu ions will combine ATs wi h di e en embedded o on-body (wea able) senso s
ha will p o ide bene i s o di e en g oups o in e es : assis ance o he end-use , moni o ing
o ela i es o ca egi e s, ac i i y o clinical epo s o specialis s.
1.1 Scope o he hesis
This hesis wo k p esen s a new me hodology o analysing he esul s o he in e ac ion be-
ween a sma olla o equipped wi h di e en senso s, he i-Walke , and a g oup o olde
adul s wi h physical dys un ions due o he na u al ageing p og ess o due o a ecen all. The
i-Walke (see §4) is used o collec da a du ing he execu ion o di e en walking es s and
exe cises pe o med by a g oup o olun ee pa icipan s om a ious cen es and na ional-
i ies. Da ase s will be comple ed wi h biological in o ma ion om each indi idual, such as
age, gende , he numbe o alls du ing he las yea , as well as some cogni i e and physical
assessmen measu es. The i-Walke will be es ed wi h wo g oups: a baseline o elde use s
pe o ming a sho walking es (10 Me e Walk Tes ) and a g oup o people om a wide age
ange pe o ming a longe walking es (3 minu es Walk Tes ).
I is di icul o es ablish a bounda y ha de ines whe he an elde is heal hy o no bu ,
since we wo k wi h in-pa ien s om hospi als and esidences, we mus disc imina e he a ge
popula ion wi h whom we wan o wo k. As he es in ol es walking o se e al minu es, we
3
1. MOTIVATION
seek o elde s wi h ambula o y capabili ies and good cogni i e s a us, so hey can unde s and
and pe o m he exe cises. Following he clinical ad ice o doc o s and physio he apis s om
he Fondazione San a Lucia (FSL), we ha e de ined a p o ocol wi h he inclusion and exclusion
c i e ia along wi h a desc ip ion o he es and expec ed ou comes. In §5we p o ide a de ailed
jus i ica ion o he p o ocol ha can also be consul ed in Appendix §A o he comple e e sion.
I is well-known ha he walking abili y o an indi idual can be a ec ed by i s cogni i e
and/o physical condi ion (Jahn e al. (2010)). One o he p ima y mani es a ions o gai dis-
u bance in old age is he slow gai pace in compa ison wi h he no mal age- ela ed slowing.
Quali a i e abno mali ies o locomo ion, such as dis u bances in he ini ia ion o locomo ion o
balance while walking a e also indica ions o quali a i e impai men o walking. The e o e,
we need o iden i y which will be he mos ele an a iables om he human- obo in e ac ion
able o de e mine gai dis u bance in a g oup o elde ly indi iduals so ha we can de ine a se o
use p o iles. We expec his in o ma ion o be use ul in wo- olds: on he one hand, we aim o
assess he quali y o he in o ma ion ob ained om his human- obo in e ac ion; on he o he
hand, we assume ha his da a will be use ul o ind some pa e ns on walking habi s acco ding
o hese use p o iles. On u u e wo ks, and wi h mo e es s, new con ol s a egies could be
de eloped o ailo he amoun o help a use p o ile will need and imp o e he communica ion
be ween he wo pa ies. I we can ind some pa e ns linking walking and medical pa ame e s,
he i-Walke could be used as an assis i e de ice ha is no only able o assis in elde s mobili y
bu is also able o diagnose a possible heal h decline by moni o ing an indi idual’s ac i i y, o
a leas o con ibu e as a decision suppo sys em o he clinician.
In a i s phase, we will use he i-Walke wi h elde indi iduals p esen ing di e en physical
and cogni i e condi ions, sepa a ing ini ially be ween alle s and non- alle s (i.e. people ha
ha e allen a leas once du ing he las yea ). We ha e also collec ed o he pe sonal cha ac-
e is ics such as he p esence o neu odegene a i e diseases o p os hesis si ua ed in legs and
whe he he pe son is using a adi ional assis i e walking de ice. Once he analysis is done,
we will es he i-Walke again wi h a new se o use s and check i i is possible o co ec ly
classi y he income o new da a and gi e he app op ia e assis ance. We expec he esul s o
his analysis could help clinicians in he diagnosis o cogni i e decline and pe sonalise he aid
ha he i-Walke could o e o he end-use .
As we will see la e , along wi h he ex , he esea ch on sma walke s du ing he las yea s
has been mainly ocused on s udying he mobili y o heal hy young people (o challenged
heal hy indi iduals) and, in a lowe p opo ion, elde ly o blind indi iduals. Li le wo k has
4
1.1 Scope o he hesis
been done in he a ea o people wi h physical and/o cogni i e disabili ies, ha may include
he old. Mo eo e , in gene al, he wo ks p esen ed a e pe o med wi h a sho numbe o
pa icipan s, p obably due o he di icul ies ound when asking o a p o ocol app o al in
hospi als and ca e cen es.
When we hink abou he oppo uni ies ha Assis i e Technologies can p o ide o olde
adul s, se e al ques ions o issues a e aised ega ding na iga ion suppo , human- obo in e -
ac ion and use ’s sa is ac ibili y wi h he whole sys em. Taking in o conside a ion he compo-
nen s and cha ac e is ics o he i-Walke (see §4), he e we e a se o p oblems ha I conside ed
impo an o ake in o accoun o he u u e design o an in elligen suppo sys em o he
i-Walke :
•Whe e a e we? Whe e a e we going? : In e p e a ion o use ’s in en ions
The i-Walke is equipped wi h a se o onboa d senso s ( o ces, odome y and lase ead-
ings) ha could be used o s udy how does each pe son ace di e en na iga ion scena ios
and de ine use p o iles acco ding o hei d i ing skills. This kind o solu ions usually
includes a known indoo en i onmen and he agenda o ac i i ies o he use (Co ´
es
e al. (2010)).
•A e you in ouble? : P edic ion o use ’s in en ions
Depending on he na iga ion si ua ion, a pe son may equi e an ex a amoun o help
(e.g. when he e a e many obs acles a ound his/he pa h). Howe e , i is impo an o
assis he use s only when necessa y and wi h he app op ia e amoun , o he wise he use
ge s used o make a minimal e o , and s/he may e en ually end up losing esidual skills.
•Was I help ul? : Ge ing use ’s eedback
S udies ha e shown ha i a pe son does no eel a ac ed o he assis i e de ice, hey
end o s op using i (Ma ins e al. (2012)). Thus, i use s a e no sa is ied wi h he as-
sis ance, p o ided by he de ice, hey will end o e use i . I he i-Walke can in e p e
he use ’s d i ing in en ions, i could p o ide some guiding assis ance by co ec ing di-
ec ionali y. In U diales (2012), a me hod o e alua ing use ’s disag eemen is p oposed
o implemen a collabo a i e con ol be ween he indi idual and he obo (see §3.2.1).
•Do you unde s and me? : Human-Machine In e ac ion
The senso s onboa d could also be used o enhance he igh in e ac ion be ween he use
and he i-Walke . S a e-o - he-a o he kind o ala m signs ha ha e been used o blind
5
1. MOTIVATION
people o pos -s oke eco e ies has o be done. When dealing wi h people wi h cogni i e
disabili ies, we can no abuse he use o ala ms because hey migh no be unde s ood o
emembe ed. Also, he in e ac ion should be smoo h by making hese ala ms disc e e
enough o a oid s essing he use .
As a esul , he i-Walke would be an assis i e de ice p o iding physical and cogni i e
suppo o impai ed and/o olde adul s. Sa e y and moni o ing a e included in he solu ion,
making he i-Walke a po en ial aid ha would allow a semi-au onomous li e o pe sons wi h
some disabili y, allowing hem o li e longe , wi h an accep able QoL, in he communi y.
Howe e , his is he i s ime ha he ou comes o he i-Walke a e analysed and assessed
om a echnical poin o iew. Un il now, he i-Walke has been only used as a mobili y o
ehabili a ion aid used o be compa ed wi h adi ional assis i e de ices, and he assessmen was
done ega ding clinical scales ega ding he use (see §4.3.1). The e o e, he p ima y objec i e
o his PhD hesis has become o p ocess he da ase ob ained om senso measu emen s o
p o ide a me hodology o da a cleaning and p epa a ion. Since he da a has been collec ed in
di e en si es, he e is a need o uni ica ion and no malisa ion me hods as a p ima y s ep. The
second objec i e is o analyse he da a o p o ide a se ice ha can classi y a use acco ding
o walking and medical pa ame e s and assis him/he in sa e na iga ion. We hypo hesise
ha a usion o senso s’ da a wi h biological da a will allow iden i ying cha ac e is ics o
walking beha iou o di e en g oups o indi iduals, in his case, elde ly people wi h high
isk o alling. We will base ou analysis on he e alua ion me ics ha a e adi ionally used
om a clinical, desc ip i e gai analysis. Taking le e age o he onboa d senso s, we belie e
we will be able o p o ide in o ma ion o he human- obo in e ac ion a s ep o me e de ail,
ins ead o ob aining global me ics. Al hough o he s udies ha e analysed gai cha ac e is ics,
ew a e he wo ks ound on s udying walking beha iou when using a olla o . We expec o
de ine di e en use p o iles acco ding o hei d i ing and walking skills using he i-Walke
bu also ela ed o hei clinical condi ion. The i-Walke could be hen used as a suppo ool
o clinicians when making a diagnose o a new use .
1.2 Plan o he Wo k
The o ganisa ion o his PhD Thesis is as ollows. In §3a e iew on assis i e de ices, and
mo e speci ically sma walke s, is gi en. §2con ains a e iew o gai analysis om a clinical
pe spec i e, om obse a ion o he in ol emen o di e en senso s ha will comple e he
6

1.2 Plan o he Wo k
medical assessmen . We hen p o ide a echnical desc ip ion o he i-Walke in §4, he sma
walke ha is used o pe o m es s wi h eal use s in his hesis, as well as he p e ious p ojec s
in which i has been in ol ed. A ull desc ip ion o he clinical ials in ol ed in his wo k is
gi en in §5. The me hodology used in his PhD p oposal is desc ibed in §6and §7show he
esul s ob ained a he momen . Conclusions and Fu u e wo k a e p esen ed in §8. A de ailed
e sion o he pilo s’ p o ocol is gi en in Appendix §A. The Appendix §Ccon ains he lis o
pape s published so a , as well as he di e en con e ences whe e I ha e assis ed. Finally,
Appendix §Bcon ains an example o a social ne wo k managed by a mul i-agen sys em whe e
he i-Walke p o ides in elligen se ices o he use , ela i es and doc o s.
Resul s on his PhD could b ing a new le el o in o ma ion ha could be used in mHeal h
solu ions like he one p oposed, gene a ing ac i i y epo s, ac ing in dange ous si ua ions and
se e as a pa ien emo e moni o ing ool. Fo his PhD wo k, I ha e s udied he gai cha -
ac e is ics ha a e iden i ied in he clinical li e a u e and he analysis pe o med by se e al
Au ho s. I ha e p oposed a me hodology using AI-based echniques o ansla e hese clinical
concep s in o he domain o he i-Walke , o lea n how o cha ac e ise an indi idual’s walking
beha iou while using a obo ised assis i e de ice. This pe spec i e is complemen ed wi h a
machine lea ning analysis ha p o ides u he knowledge on human mo ion body and how o
ca ego ise i .
7
1. MOTIVATION
8
Chap e 2
A e iew o Gai , Cogni ion and Falls
The abili y o walk no mally is ela ed o se e al bio-mechanical componen s in ol ed in he
gai cycle (also known as s ide), including (i) ee mobili y o join s, pa icula ly in he legs;
(ii) coo dina ion o muscle ac ion in e ms o iming and in ensi y; (iii) no mal senso y inpu ,
such as ision and es ibula sys em (see Rubens ein (2006)). Thus, gai equi es inpu om
he b ain, spinal co d, pe iphe al ne es, muscula powe and join and ca dio ascula heal h.
Because all o hese sys ems a e needed o coo dina e gai , he indi idual’s walking speed is an
indica o o he heal h o many physiological sys ems (see F i z and Lusa di (2009) and Paho
(2006)).
The ela ion be ween gai and cogni ion has been widely analysed om he medical poin
o iew, and we can ind se e al e iews in he li e a u e (see Hagga d e al. (2000), Mon e o-
Odasso e al. (2012) and Rosso (2013)). As people age, hey end o slow hei gai speed, and
hei balance is also a ec ed. Also, he e i emen om he wo king li e and he consequen
educ ion o physical and social ac i i y con ibu e o he inc eased incidence o alls in olde
adul s. Mo eo e , olde adul s su e di e en kinds o cogni i e decline, such as demen ia
o a en ion p oblems, which also accen ua e gai diso de s and i s consequences (Sche de
e al. (2007), Yoge -Seligmann e al. (2007) and Plumme -D’Ama o e al. (2012)). Also, cu -
en concep s in disablemen emphasise he impo ance o iden i ying mobili y impai men s
in ageing humans o enable imely in e en ion and, ul ima ely, p e en disabili y as s a ed by
McGibbon e al. (2001).
I is es ima ed ha a e age 70, 35% o he popula ion p esen gai diso de s due o di e -
en easons ela ed o cogni i e and/o physical decline. Mos o hem associa ed wi h he
ageing p ocess, bu o he ac o s, such as educa ion and li es yle, a e also in luen ial (see
9
2. A REVIEW OF GAIT, COGNITION AND FALLS
Yoge -Seligmann e al. (2007)). One o he mos common and d ama ic consequences o
gai diso de s is alling. O e a hi d o he popula ion, aged 65+ yea s all e e y yea (50%
o adul s aged 80+ yea s). As a consequence, 4-15% o alls cause signi ican inju ies, while
23-40% o inju y- ela ed dea hs in olde adul s a e due o a all O ganiza ion (2015)).
2.1 Gai cha ac e is ics in elde ly popula ion
The capabili y o ge om one place o ano he and success ully each he desi ed des ina ion is
essen ial o e e y animal and he e o e o humans1. Walking is a undamen al pa o e e yday
li e and depends on balance, join mo ion, endu ance, and muscle s eng h (G aham e al.
(2008)). Human locomo ion has been s udied o decades, al hough he pe spec i e, as well as
he ools used o measu emen , ha e e ol ed o hese days. This ield o esea ch includes all
ages o he human being, bu also o he animal domain. This chap e ocuses on gai analysis
as he s udy o he human walking, which analyses he body mechanics and he ac i i y o he
muscles in ol ed in he walking p ocess (Whi le (2007)). Da a used in his ield can bo h come
om clinical assessmen s (physical o cogni i e) and di e en so s o measu emen ools ha
will collec some gai cha ac e is ics.
Gai analysis is a sys ema ic echnique o ecognising nega i e de ia ions in he gai pa -
e n and de e mining hei eason and e ec s (P akash e al. (2016)). As men ioned be o e,
se e al condi ions migh a ec he abili y o walk, and i a ec s mainly o people when ageing.
Fo hese easons, i is essen ial ha clinicians egula ly assess hei gai o diagnose and plan
op imal ea men s o each si ua ion. Al hough i is no possible o p o ide a gene al desc ip-
ion o gai wi hou including all he singula i ies gi en in each pa hology, i is well accep ed
ha a no mal gai in ol es he locomo i e ac ion o he wo legs, al e na ely o p o ide bo h
suppo and p opulsion, ha ing always a leas one oo in con ac wi h he g ound Whi le
(2007). The walking ac i i y is composed o he s ance and swing phases. The as e we walk,
he sho e he s ance phase will be. A gai cycle is hus de ined as he ime in e al be ween
wo successi e occu ences o a epe i i e e en (e.g., he igh leg ini ial con ac wi h he
g ound). Figu e 2.1 shows he di e en phases included in a gai cycle s a ing wi h he igh
leg (in g ey colou ), aken om Whi le (2007). Body equilib ium is a key ac o o co ec ly
al e na e leg displacemen s and sus ains he body weigh .
1Fo he elde ly, walking, s anding up om a chai , u ning, and leaning a e necessa y o independen mobili y.
Gai speed, chai ise ime, and he abili y o do andem s ance (s anding wi h one oo in on o he o he , which
10
Chap e 3
A Re iew on Assis i e Technologies
Olde adul s usually su e om a leas one heal h condi ion, om isual o audi i e impai -
men , muscula weakness, o neu o-degene a i e disease and o he s. Mos o hese si ua ions
lead o a decline in he locomo i e unc ions, which is mani es ed as he di icul y o sol e com-
plex ambula o y si ua ions (e.g., a oid obs acles o u ns o le o igh ), o gai diso de s (e.g.,
dec easing gai speed, uns eady gai ). The combina ion o he men ioned ac o s inc eases he
isk o alling in his popula ion. Assis i e Technologies a e essen ial o he eco e y o e-
placemen o he mobili y unc ions a all ages and enhance he au onomy and quali y o li e o
pa ien s and ela i es.
Th ough his sec ion, I will b ie ly in oduce some esea ch wo ks in he ield o assis-
i e echnologies designed o he mobili y eco e y (e.g., exo-skele ons) and assis ance (e.g.,
Au onomous Robo ic Wheelchai s o canes). Howe e , olla o s and mo e speci ically sma
walke s a e a po en ial ool o s udy, since i can be in ol ed in solu ions o bo h p oblems,
eaching a highe numbe o possible end-use s. This e iew is needed o unde s and and syn-
hesise he backg ound o his hesis. I also in oduces he wo k pe o med wi hin wo EU
unded p ojec s: (i) SHARE-i , whe e di e en assis i e echnology was de eloped o enhance
au onomy and quali y o li e o elde ly people 1; and (ii) I-DONT-FALL, whe e he i-Walke is
used as a ehabili a ion ool o all p e en ion.
1The KEMLG g oup coo dina ed he p ojec a UPC.
17

3. A REVIEW ON ASSISTIVE TECHNOLOGIES
3.1 Assis i e De ices
I is in e es ing o obse e he as e olu ion o assis i e mobili y de ices o e he las wo
decades. Al hough we a e used o seeing people a ound us using adi ional assis i e de ices
(and, mo e ecen ly powe ed wheelchai s o scoo e s), solu ions in ol ing obo ic echnology
and A i icial In elligence a e s ill unde esea ch de elopmen . Howe e , he e olu ion o
new echnologies (e.g., diminishing in size and cos while g owing po en ial) a e helping
o c ea e many di e en solu ions ha will someday imp o e ou way o li ing. Di e en
obo ised e sions o each adi ional de ice ha e been ecen ly p oposed, aiming o help in
ei he diagnosis, mobili y o ehabili a ion.
Assis i e de ices can be classi ied in o wo ca ego ies depending on he pe son’s le el o
mobili y: al e na i e o people wi h he o al o empo al incapaci y o mobili y and augmen-
a i e o people wi h emaining mobili y capaci ies. In Ma ins e al. (2012), au ho s ep esen
his classi ica ion as shown in Figu e 3.1.
In he ield o al e na i e assis i e de ices, di e en models o Au onomous Robo ic Wheel-
chai s (ARW) a e p oposed in he li e a u e (Wes on (Hillman e al. (2002)), Wheelesley (Yanco
(1998))), CARMEN (U diales e al. (2011)). ARWs p o ide solu ions o au onomous and as-
sis i e na iga ion, bu hose a e no mally es ic ed o people wi h he o al incapaci y o mo-
bili y. I is ecommended ha people wi h esidual mobili y skills a oid he use o wheelchai s
du ing long pe iods o ime, as i may lead o a loss o capabili ies ( see Ma ins e al. (2012)).
Du ing he las decade esea ch has ocused on he de elopmen o obo ized augmen a i e
de ices, such as ehabili a ion ools o ambula o y- aining o people su e ing om muscu-
loskele al o neu ological diso de s like s okes o spinal inju y (Hap icWalke , (Schmid e al.
(2007)), KineAssis , (Pa on e al. (2008)), LokoHelp, (Swinnen e al. (2010))). The disad-
an age o his kind o de ices is ha hey a e used in-hospi al and equi e he expe ise o a
clinician o physio he apis ha ensu es ha he ools a e being co ec ly applied.
One o he mos impo an augmen a i e de ices is he sma walke (see §4) because o i s
po en ial, no only in mobili y bu also in ehabili a ion. I is well accep ed ha sma walke s
o e enough weigh balance, and hus help o aise sel -con idence and au onomy o pa ien s
wi h locomo ion p oblems. Du ing he las decade, many solu ions ha e been p oposed, being
equipped wi h di e en kinds o senso s o p o ide no only mobili y bu moni o ing as well.
Solu ions also a y in he numbe o wheels, going om he wo-wheeled models o olla o
18
3.1 Assis i e De ices
Figu e 3.1: Ca ego ies o Assis i e De ices (Ma ins e al. (2012))
walke s, wi h ou wheels. Sma walke s a e expec ed o p esen he ollowing unc ionali ies
(F ize a-Ne o e al. (2011)):
•Physical suppo : sma olla o s should p o ide be e gai s abili y.
•Senso ial assis ance: sma olla o s should also collec and p ocess da a om di e en
onboa d senso s o assis in na iga ion and inc ease secu i y o he inal use (use ul o
obs acle a oidance o all p e en ion)
•Cogni i e assis ance: use s ha ing p oblems ela ed o memo y o o ien a ion may need
some guidance and localisa ion sys em
•Heal h moni o ing: used o keep he medical his o y o he use
19
3. A REVIEW ON ASSISTIVE TECHNOLOGIES
•Human-machine in e ace (HMI): di ec ly o indi ec ly, HMIs a e used o communica e
wi h he use h ough a sys em o ale s, ala ms and commands.
Rolla o s allow he pe o mance o a na u al gai pa e n du ing locomo ion. Howe e , hey
a e also conside ed he mos uns able e sion and he isk o alling while using i inc eases in
he si ua ions ha equi e he ull-body weigh suppo o he use (Ne o e al. (2015)).
Rega ding he physical aspec , mos o he sma olla o s ound in he li e a u e a e no
based on a adi ional olla o , so he e is a as a ia ion ega ding physical designs. Fo
ins ance, he numbe o wheels and handle s a ies among he p oposed solu ions, bu also he
deg ee o assis ance ha each o hem o e .
The e a e also di e en applica ions o hese sma walke s ega ding senso ial assis ance:
each solu ion is equipped wi h di e en se s o senso s, usually ocused on objec de ec ion
o use localisa ion. Some o hese solu ions a e designed o assis blind people in na iga ion
(Yu e al. (2003b)). In Glo e e al. (2004) a walke is p o ided wi h na iga ion guidance o
olde adul s who a e cogni i e o men ally ail by lea ning people’s mo ion beha iou s and
p o iding di ec ions h ough a ouch-based in e ace. Sma walke s empowe he mobili y o
he use , p o iding suppo while walking and inc easing he con idence and sa e y pe cep ion
du ing ambula ion.
Rega ding d i ing assis ance, he e a e h ee s a egies o human- obo in e ac ion ha a e
usually ound in he li e a u e, depending on which one has a highe decision con ol. These
s a egies a e mo e commonly de eloped o ARW sys ems since a disag eemen in he d i ing
decision shall no ha e majo physical consequences o he end-use . Howe e , in he case
o sma walke s, a s ong disag eemen could lead he use o lose balance, p o oking some
inju y o e en a all. D i ing con ol s a egies a e desc ibed below:
•Human ull-con ol: The use akes con ol o he assis i e de ice o mo e owa ds a
gi en goal. This solu ion is no sui able o people p esen ing cogni i e disabili ies, as
secu i y issues can be moni o ed bu he obo canno ake any decision.
•Robo ull-con ol: The obo is awa e o he ADLs and he en i onmen o he use ,
and akes all decisions bu , as s a ed be o e, his may lead o a se ious loss o use s
capabili ies and us a ion.
•Sha ed (o collabo a i e) con ol: A combina ion o he p e ious s a egies in which he
obo will p o ide an amoun o assis ance acco ding o use ’s skills. The idea is o help
only when necessa y, so he use does no ge used o he obo doing all he job.
20
3.1 Assis i e De ices
This las s a egy is he mos commonly used in he las decades Cowan e al. (2012) in he
ield o ARWs (see CARMEN U diales e al. (2011), Boy e al. (2002) and Ca lson and Demi is
(2010)). Resea ch on sma walke s has been mo e ocused on na iga ion assis ance as a guide
bu no as a way o con ol. Fo ins ance, he EU FP7 unded p ojec DALi de eloped a po able
mo ion planning using a s anda d olla o equipped wi h a Kinec and a able o guide he use
in c owded en i onmen s. The aim is o elie e he s ess su e ed by people wi h educed
cogni i e o physical abili y, especially olde adul s. The sys em does no ac i ely assis on he
na iga ion, bu i p o ides a b ake con ol when i de ec s ha he use has de ia ed signi ican ly
om he objec i e.
Some examples ha ha e been de eloped in he las decade a e:
•PAMM (Spenko e al. (2006); Yu e al. (2003a)) is a sma obo ic walke , designed a he
Massachuse s Ins i u e o Technology. I aims o p o ide suppo , guidance, and heal h
moni o ing o elde ly use s in o de o delay o ansi ion o nu sing homes. The la es
e sion o PAMM was based on a ou -wheeled s uc u e wi h wo handle (a sma cane
e sion wi h wo wheels and one handle is has also been de eloped). I also con ains a
came a o localisa ion and obs acle a oidance.
•COOL Aide, (Wasson e al. (2008)), buil on a s anda d h ee-wheeled olla o . Fo ce
and momen senso s ha e been added o he handle s, as well as encode s in he wheels
o ob ain he posi ion, eloci y and heading. I also con ains a Hokuyo lase o assis he
use a oiding obs acles. In Huang e al. (2005), au ho s p opose a sha ed con ol s a egy
o help use s dealing wi h possible collisions and eaching sho - e m goals in p ede ined
pa hs.
•iWalke (Kulyukin e al. (2008)), de eloped collabo a i ely by he Ca negie Mellon Uni-
e si y and he Uni e si y o Pi sbu g. I is based on a s anda d ou -wheeled olla o
equipped wi h encode s, RFID senso s and a lase o p o ide au onomous na iga ion and
sel -pa king op ion.
•Sma Walke (Wada e al. (2016)) ha e buil a si ing- ype walke , which is a solu ion
hal -way be ween he wheelchai , whe e people will no equi e any mo ion o ce o
mo e, and he olla o , which some challenged adul s will no be able o use due o he
lack o suppo o ce in a ms o legs. The Sma Walke is equipped wi h an ac i e-cas e
d i ing sys em which will help he use mo ing is gi en si ua ions.
21
3. A REVIEW ON ASSISTIVE TECHNOLOGIES
•UFES (Ne o e al. (2015)) was de eloped unde a esea ch p ojec be ween he Uni e -
si y o Espi i o San o (B azil) and he Uni e isi y o San Juan (A gen ina). I p esen s a
s ic u e o 3 wheels wi h encode s and mo o s, ine ial mo emen senso s and 3D o ce
senso s placed on op o wo o ea m suppo pla o ms. UFES p o ides a con ol a -
chi ec u e which enables an eme gency b aking in unsa e si ua ions. I also collec s gai
pa ame e s wi h addi ional IMU senso s placed on he use ’s body. The o ce senso s
de ec he guidance in en ions
•ASBgo++ (Al es e al. (2017)) is a ou -wheeled mo o ized olla o wi h o ea m suppo
pla o ms, buil a he Minho Uni e si y (Po ugal). I p o ides sa e y na iga ion con ol
and in o ma ion abou he use gai pa e n. I con ains a joys ick which cap u es use s
mo emen in en ions (Ma ins e al. (2014)).
•Cheng and Wu (2017) ha e de eloped a sma olla o wi h p essu e senso s added o he
handle s o cap u e use ’s d i ing in en ion using a suppo ec o machine and AdaBoos
classi ie o iden i y he mo emen ec o s.
One o he objec i es o SHARE-i and also o I-DONT-FALL and FATE, he EU
p ojec s in which he i-Walke has been in ol ed in, was o ensu e ha he assis i e de ice
was use iendly,i.e. a de ice he use is amilia wi h, so s/he will be mo e com o able us-
ing i . The cos o p oduc ion is also educed as well, as we s a ed om a s anda d ame wi h
elec ical componen s embedded ins ead o building he obo om sc a ch. A ull desc ip ion
o he i-Walke is p o ided in Chap e §4.
3.2 SHARE-i
Wi hin he ame o SHARE-i EU, unded p ojec FP6-0450881, a he Uni e si a Poli `
ecnica
de Ca alunya, wo PhD hesis ha e been de ended by C is ian Ba u´
e and C is ina U diales
(see Ba u´
e(2012) and U diales (2012) espec i ely) a he UPC A i icial In elligence PhD
p og amme. These heses in oduced di e en solu ions de eloping and applying a esea ch
1SHARE-i was a h ee yea s p ojec unded by he Eu opean Commission whose p ima y objec i e was o
de elop AT which enable olde adul s o li e independen ly and wi h he high quali y o li e as long as possible.
The aim was o c ea e scalable, adap i e sys ems o add-ons o he senso and AT, mainly ocused on suppo ing
au onomous mobili y, so ha hey can be modula ly in eg a ed in o an in elligen home en i onmen o enhance he
indi idual’s au onomy.
22

3.2 SHARE-i
app oach o imp o e he quali y o li e among indi iduals su e ing some disabili ies. Tha
echnology was u he applied in o he wo EU unded p ojec s I-DONT-FALL and FATE.
The p ima y goal o he SHARE-i p ojec (Co ´
es e al. (2010)) was o con ibu e o he
de elopmen o he nex gene a ion o in elligen and semi-au onomous assis i e de ices o
olde pe sons and people wi h disabili ies (bo h cogni i e and mo o ). When i comes o ageing,
usually diseases do no come alone. When losing mobili y, people’s abili y o be sel -dependen
in pe o ming hei ADLs dec eases and hey end up hospi alised o in a day-ca e ins i u ion,
losing hei social en i onmen . I he pe son also has some cogni i e disabili y, he possibili ies
o ha ing an au onomous li e go educed. The objec i e o SHARE-i was o de elop a scalable,
adap i e sys em o componen s (i.e. senso s and ATs) in eg a ed in o an in elligen home o
enhance indi idual’s au onomy and hus, i s Quali y o Li e.
Rega ding he mobili y aspec , ou di e en assis i e de ices we e de eloped and de-
ployed in eal en i onmen s and wi h eal use s, h ee ARWs (CARMEN (U diales e al. (2011)),
Sphe ik (Ma ´
ınez e al. (2005))and Rolland (Ch is ian e al. (2008)) and a o me e sion o he
i-Walke (Annicchia ico e al. (2008)). The main pu pose was o p o ide mobili y assis ance
o a wide ange o use s wi h di e en capabili ies and needs.
We es ed he mobile pla o ms in an Ambien In elligence en i onmen , a house equipped
wi h a se o domo ic capabili ies and senso s. A mul i-agen sys em con olled all he da a
collec ed and managed he in e ac ion be ween he assis i e de ices and he in elligen house
h ough a se o cogni i e se ices.
In 2009, I de eloped my Bachelo ’s p ojec (PFC) a he Fondazione San a Lucia (FSL)
wi hin he SHARE-i p ojec . The objec i e was o design and pe o m a benchma k wi h eal
in-pa ien s o es he in e ac ion be ween CARMEN and he in elligen home h ough a mul i-
agen sys em. The agen con olling he wheelchai was esponsible o deciding he amoun
o help ha e e y use needed a any gi en ime. A localisa ion sys em allowed he in elligen
house o moni o he use ’s mo emen s; when he use was d i ing in a na ow space (like
a co ido ) o c ossing a doo . The wo agen s in e ac ed o inc ease he amoun o help, i
equi ed, and dec ease he secu i y dis ance be ween he wheelchai and walls o acili a e he
manoeu abili y. We p o ed ha he use s we e able o end up hei asks easie and hei
na iga ion was smoo he hanks o he mul i-agen sys em.
O he agen s should ha e been in ol ed in he benchma k, bu due o ex e nal delays, i
was no possible o es i du ing my s ay a FSL. Howe e , hey we e included in he inal es s
23
3. A REVIEW ON ASSISTIVE TECHNOLOGIES
o he SHARE-i p ojec and esul s we e e y p omising as well. Use s passed a usabili y and
disag eemen es s, and he eedback was also e y posi i e.
The success ul esul s o his p ojec led o he u he de elopmen o he i-Walke , bo h
ega ding design (e.g., educ ion on elec ic componen s, Eu opean homologa ion as he med-
ical de ice) and se ices o mobili y assis ance and eco e y. I has since hen been in ol ed
in h ee EU unded p ojec s (I-DONT-FALL,FATE and ASSAM) and wo Spanish na ional
p ojec s (SiRC and RehAdap a) all o hem ela ed wi h he de elopmen o Assis i e Tech-
nologies. Chap e §4con ains a comple e desc ip ion o he i-Walke used in his hesis, along
wi h he main con ibu ions in esea ch de eloped du ing he las yea s.
3.2.1 CARMEN: an ARW wi h collabo a i e con ol
Wi hin he SHARE-i p ojec , U diales (2012) designed and buil a obo ised wheelchai , CAR-
MEN, wi h a collabo a i e con ol ha aimed o assis a pe son’s mobili y. CARMEN can de ec
how much help a gi en use needs depending on his/he abili ies and cu en condi ion and o
p o ide he equi ed help: no mo e, no less. The concep behind his app oach is o a oid loss
o esidual skills due o excessi e help bu o p o ide, ne e heless, he equi ed assis ance o
achie e mobili y in e e yday en i onmen s wi h a powe ed wheelchai . Following he medical
eam ad ice, as a secu i y measu e, he wheelchai could no be d i en backwa ds.
The wheelchai is only d i en by a joys ick which ansla es use ’s commands in o di ec-
ionali y and p o ides a na iga ion aid by adap ing o he amoun o help equi ed by each
use , acco ding o hei le el o disabili y. A he apis p e iously de e mines he amoun o
assis ance o e e y use .
CARMEN was buil o be d i en in an in elligen , adap ed home ollowing an agenda o
ADLs de ined by a medical eam. I was i s es ed wi h a se o in-pa ien s o he FSL using
a pu ely eac i e con ol o benchma king, which mainly a oids obs acles hanks o he in o -
ma ion ansmi ed by a on al Hokuyo lase . Use s had o es CARMEN in di e en de ined
pa hs (always including a doo o co ido c ossing), i s wi hou any aid (as a con en ional
powe ed wheelchai ) and hen wi h he eac i e con ol.
The sys em ga he ed da a abou he use ’s commands, i s ela i e posi ion in he en i on-
men and he objec s (o po en ial obs acles) su ounding he wheelchai along wi h o he a i-
ables, like he ime equi ed by each use o comple e he pa h. This in o ma ion was used o
measu e how well did he use pe o m a gi en ask. U diales inally o malised he in o ma ion
in o di e en ask me ics, al hough he mos ele an o he es o his wo k whe e:
24
3.2 SHARE-i
Figu e 3.2: Vec o s in ol ed in mo ion command calcula ion.
•di ec ness: he use d i es keeping he goal ahead
•smoo hness: he use ’s na iga ion (in e p e ed h ough he joys ick mo emen s) p esen s
sha p di ec ion changes
•sa e y: he use d i es wi h a sa e dis ance o obs acles
A pe son’s e iciency pe o ming a gi en ask acco ding o hese me ics is hen calcula ed.
Figu e 3.2).a depic s ec o s in ol ed in collabo a i e con ol (VR, VH and VC o obo , hu-
man and collabo a i e); Figu e 3.2).b shows he angles in ol ed in es ima ing smoo hness (R),
sa e y (G) and di ec ness (B): local e iciency a a gi en loca ion can be isually e alua ed om
i s RGB colou . In addi ion o hese me ics, use s had o comple e di e en es s ega ding
medical scales ha measu e cogni i e and physical disabili ies, o usabili y and disag eemen
ques ionnai es. U diales inally es ed i wi h a collabo a i e con ol ha gi es d i ing con ol
o he wheelchai when he use does no each a minimum h eshold o success in achie -
ing he goal o a oiding an obs acle. Resul s we e p omising, as hey showed ha use s we e
able o pe o m be e hei asks (wi hou he na iga ion aid, hey we e o en unable e en o
accomplish hem).
The second phase o he wo k consis ed o c ea ing skill-based wheelchai na iga ion p o-
iles. The objec i e o using all he da a ga he ed om p e ious es s is o sepa a e use s in
p o iles acco ding o hei na iga ion pe o mance and he medical scales o p edic he amoun
25
3. A REVIEW ON ASSISTIVE TECHNOLOGIES
o help equi ed a e e y momen ins ead o ixing i in ad ance. U diales used na iga ional
in o ma ion om bo h heal hy people and pe sons wi h di e en kinds o disabili ies. Acco d-
ing o Minguez e al. (2004), in obo na iga ion, we can deal wi h six di e en si ua ions o
achie e collision a oidance in oublesome scena ios, depending on he le el o sa e y o he
obo wi hin he en i onmen (secu i y zone), and he eachabili y o he goal ( ee walking
a ea). Figu e 3.31:
•Sa e y c i e ion: High Sa e y (HS) o Low Sa e y (LS) ep esen s he absence and p es-
ence o obs acles espec i ely.
The i s c i e ion is applicable o e e y si ua ion. The ollowing c i e ia co espond o HS
si ua ions.
•Goal wi hin he ee walking a ea c i e ion: Co esponds o he High Sa e y Goal in
Region (HSGR), which means ha we a e in HS and he goal loca ion is wi hin he ee
walking a ea
•F ee walking a ea wid h c i e ion: When he goal is no in he ee walking a ea, wo
new si ua ions a e de ined depending on whe he he a ea is wide o na ow: High Sa e y
Wide Region (HSWR) and High Sa e y Na ow Region (HSNR)
Finally, hey ob ain h ee new si ua ions in LS.
•Goal wi hin he ee walking a ea c i e ion: This c i e ion is simila o he second one,
bu his ime he ee walking a ea p esen s some obs acles. The esul ing si ua ion is
Low Sa e y Goal in Region (LSGR).
•Dange ous obs acle c i e ion: Two possible si ua ions may occu applying his c i e ion.
The i s one, Low Sa e y 1 Side (LS1), is gi en when he e a e obs acles wi hin he
secu i y zone, bu only on one side o he discon inui y (closes o he goal) o he ee
walking a ea. The la e one, Low Sa e y 2 Sides (LS2) means he e a e obs acles wi hin
he secu i y zone on he wo sides o he discon inui y
In U diales e al. (2013) au ho s used all he in o ma ion ga he ed du ing he es s desc ibed
abo e, adding he da a o heal hy people ha pe o med he same asks, o ob ain a sample o
100 people. Thei baseline use p o ile is buil in a h ee-s ep clus e ing p ocess.
1This igu e is ex ac ed om Minguez e al. (2004) shows hese six possible si ua ions which a e de ined
acco ding o he ollowing c i e ia
26
3.3 I-DONT-FALL
need o be s able pa ien s, wi h a leas one mon h om las acu e e en . The p esence o a ca e-
gi e du ing he sessions is manda o y. On he o he hand, pe sons p esen ing aphasia and/o
neglec o majo beha iou al dis u bances we e excluded om he s udy. Also, people in ol ed
in ehabili a i e aining canno pa icipa e in he pilo due o sa e y easons. Pa icipan s who
sa is ied inclusion c i e ia we e assigned andomly o one o hese ou g oups:
•MOTOR: 125 pa icipan s in walking aining ha pe o m a se o exe cises wi h a
physio he apis using he i-Walke .
•COGNITIVE: 125 pa icipan s in cogni i e aining ha pe o m a se o exe cises using
he SOCIABLE pla o m.
•MIXED: 125 pa icipan s in he combined aining (mo o + cogni i e aining).
•PLACEBO: 125 pa icipan s as a con ol g oup (placebo ac i i y o con ol he subjec -
expec ancy e ec ).
The walking aining p ocess is execu ed in 2 sessions pe week o 12 weeks (24 sessions).
Each aining session akes 1h o du a ion. Volun ee s a e ee o d op ou o he s udy i he
subjec ails o pa icipa e in he aining o mo e han wo consecu i e weeks (4 sessions),
hen s/he is conside ed as a d opou . Each aining session is dedica ed o 1/2 o balance and
1/2 o gai exe cises a e a b ie session on wa m-up exe cises.
•Walking aining session o 30 minu es:
–3 o wa m-up exe cises (exe cises selec ed om wa m-up pool)
–15 o balance (exe cises selec ed om balance pool)
–15 o gai (exe cises selec ed om gai pool)
•Walking aining session o 60 minu es:
–3 o wa m-up exe cises (exe cises selec ed om wa m-up pool)
–15 o balance (exe cises selec ed om balance pool)
–15 o gai (exe cises selec ed om gai pool)
–5 pause
–15 o balance (exe cises selec ed om balance pool)
33

3. A REVIEW ON ASSISTIVE TECHNOLOGIES
–15 o gai (exe cises selec ed om gai pool)
The cen al objec i e o his s udy is o educe he numbe o alls in he expe imen al g oup
espec o he con ol g oup. As a consequence, i is also expec ed o minimise he isk o alling
(measu ed by he Tine i es ) and he ea o alling (measu ed by he Fall E icacy Scale FES
es ). We also aim o e alua e he usabili y o he sys em and use sa is ac ion conce ning he
applied aining p og am (i.e., i hey ha e pe o med walking and cogni i e aining o jus
one o he wo). The seconda y ou comes o his s udy a e he imp o emen o mobili y using
use s balance and gai (measu ed by Tine i es , 6 minu es Walking Tes and 10 me e s Walking
Tes ) (see Tine i e al. (1986)), as well as he imp o emen o he QoL and he unc ional and
cogni i e abili ies. We will also analyse he dynamic ela ionship be ween pushing o ces and
walking dynamics, c ossed wi h he use medical p o iles. Ou comes will be e alua ed h ough
a mul idimensional assessmen ha will be adminis e ed wo imes: be o e he aining pe iod
s a s (T0) and a e he aining pe iod ends (T1).
34
Chap e 4
The i-Walke
The assis i e de ice p esen ed in he e is based on he p o o ype de eloped on EU unded p ojec
SHARE-i (see SHAREi and Co ´
es e al. (2010)) a he Uni e si a Poli `
ecnica de Ca alunya
(UPC). The consolida ed e sion, he one used o his wo k, is based on a s anda d 4-wheeled
Rolla o AD-100 wi h a se o embedded senso s and ac ua o s, aiming o assis o he mobili y
and he ehabili a ion o pe sons wi h physical and/o cogni i e disabili ies and moni o ing
hei ac i i ies (see Annicchia ico e al. (2008)). In he end, he i-Walke looks like a adi ional
olla o . I is designed o p o ide po en ial use s wi h su icien ambula o y capabili y in an
e icien , cos -e ec i e way. I ollows he ISO equi emen s o walking aids manipula ed by
bo h a ms (see ISO) and has ob ained he EU app o al as a medical de ice o clinical esea ch.
The ac ual e sion o i-Walke has been used in he I-DONT-FALL ,FATE and ASSAM EU
p ojec s unded by he Compe i i eness and Inno a ion F amewo k P og amme (CIP) o he
Eu opean Union (see I-DONT-FALL,FATE and ASSAM espec i ely). Fo his PhD, pa o
he da a gene a ed by he i-Walke in he I-DONT-FALL p ojec has been used, which has been
desc ibed in §3.3.
In his chap e , a ull desc ip ion o he sma walke ( he i-Walke , see Figu e 4.1) used in
his PhD is gi en, including an in oduc ion o i s main componen s and he ole hey play in he
sys em; he eac i e con ol ha has been de eloped o p o ide compensa ion and sa e y in he
ambula o y ac i i ies in indoo and ou doo en i onmen s; a possible app oach o sha ed con ol
na iga ion. This chap e con ains also an o e iew o he i-Walke ’s assis i e en i onmen ,
desc ibing i s ole in di e en esea ch p ojec s: h ee mas e hesis de eloped a UPC, whe e
da a collec ed om he i-Walke in di e en scena ios was analysed wi h di e en pe spec i es
35
4. THE I-WALKER
and aims. This sec ion also p o ides a summa y o he I-DONT-FALL p ojec esul s om a
clinical poin o iew.
In Appendix §Bin oduces he esul s o in eg a ing he i-Walke as an in elligen se ice
wi hin a Social Ne wo k (Ba u´
e e al. (2015)). This wo k was a pa o a p elimina y s udy o
assess he plausibili y o lea ning om he in e ac ion o se e al i-Walke s and hei espec i e
use s and ca egi e s. In ha wo k, he i-Walke in e ac ed in he Social Ne wo k as one mo e
agen in a Mul i-Agen sys em.
4.1 Main componen s
The i-Walke is a dis ibu ed mic o-con olle a chi ec u e which d i es he sys em and eco ds
and p o ides s uc u ed in o ma ion o he apis s. All he elec onics a e embedded inside he
handle s (1) and ea wheels (9) o he i-Walke . A box (3) unde he sea (2) con ains he
compu ing powe onboa d (a Raspbe y Pi) and a se o senso s ha will p o ide in o ma ion
abou mo emen and il . Fo his wo k, we ha e also added a on al Hokuyo lase o de ec
obs acles and a oid possible collisions. Table 4.1 summa ises all he di e en a iables cap-
u ed and ga e a b ie desc ip ion o each one. The names appea ing in pa en hesis will be he
ones used along he es o his documen .
Va iable Desc ip ion
Le Hand Fo ce X (lh x) Longi udinal (Fo wa d-Backwa d) pushing
o ce exe ed by he use on he le handle-
ba
Le Hand Fo ce Y (lh y) T ans e sal (Le -Righ ) pushing o ce ex-
e ed by he use on he le handleba
Le Hand Fo ce Z (lh z) Ve ical (Up-Down) pushing o ce exe ed by
he use on he le handleba
Righ Hand Fo ce X ( h x) Longi udinal (Fo wa d-Backwa d) pushing
o ce exe ed by he use on he igh handle-
ba
Righ Hand Fo ce Y ( h y) T ans e sal (Le -Righ ) pushing o ce ex-
e ed by he use on he igh handleba
Righ Hand Fo ce Z ( h z) Ve ical (Up-Down) pushing o ce exe ed by
he use on he igh handleba
36
4.1 Main componen s
Le No mal Fo ce (ln ) Le Rea Wheel No mal Fo ce. This is he
o ce ha he loo exe s on he wheel. When
he alue is below a gi en posi i e h eshold,
i means ha he i-Walke is losing i s con ac
wi h he loo
Righ No mal Fo ce ( n ) Righ Rea Wheel No mal Fo ce. This is he
o ce ha he loo exe s on he wheel. When
he alue is below a gi en posi i e h eshold,
i means ha he i-Walke is losing i s con ac
wi h he loo
Til ( il ) Angle o e he la e al axis o he i-Walke
Roll ( oll) Angle o e he longi udinal axis o he i-
Walke
Hand B ake Le (hbl) S a e o he b eak: 1 i blocked, 2 i manually
ac i a ed
Hand B ake Righ (hb ) S a e o he b eak: 1 i blocked, 2 i manually
ac i a ed
Es ima ed Pose X (epx) Pose es ima ed om he beginning o he ex-
e cise in Y axis (s a ing om poin [0,0] in a
Euclidean space)
Es ima ed Pose Y (epy) Pose es ima ed om he beginning o he ex-
e cise in Y axis (s a ing om poin [0,0] in a
Euclidean space)
Es ima ed Pose O ien a ion
(psi)
Es ima ed o ien a ion o he i-Walke ega d-
ing he ini ial o ien a ion ( he o ien a ion
ha he i-Walke had when he exe cise has
s a ed)
Le Wheel Speed (ls) Le wheel speed
Righ Wheel Speed ( s) Righ wheel speed
Table 4.1: i-Walke a iables and de ini ions.
Handleba s
The o iginal handleba s ha e been eplaced by ano he model also used in s anda d olla o s.
These new handleba s (1) measu e he use ’s o ce exe ed along he longi udinal, la e al and
e ical di ec ions (X,Yand Z espec i ely) h ough embedded o ce senso s. Heigh g adu-
a ion is main ained in his new design (8)1. By adding hese o ce senso s in he g ips o he
handleba s, we can egis e a any ime he o ce exe ed by he use . The handleba s can also
moni o he s a es o he b ake le e s, ha ing a manual b ake (7) wi h wo di e en s a es:
1This is use ul in pos -s oke pa ien s, as hey may need di e en heigh g adua ion on each hand
37
4. THE I-WALKER
Figu e 4.1: The I-DONT-FALL e sion o he i-Walke .
38

4.1 Main componen s
•Pa king b ake: a mechanical b ake can be ope a ed by pushing down he b ake le e
(b aked s a e). This ac ion is ecommended when he use wan s o s op in he middle o
a slope o wan s o es in he sea .
•Dynamic b ake: he b ake le e can be pulled up o slow down he speed o he walke .
This will inc ease he pushing o ces exe ed by he use , gi ing him/he a highe sense o
secu i y. This unc ionali y is use ul in downhill, as he use does no ha e o be pulling
he i-Walke and, hus, a oiding his one o go away oo as (a mo e de ailed desc ip ion
is gi en in §4.2).
All he in o ma ion p o ided by he handleba s is collec ed and s o ed so he medical s a
can con ol i he use is doing a good use o he de ice.
The new handleba s include some ea u es ha aim o imp o e he in e ac ion be ween he
i-Walke and he pe son using i . On he one hand, a b igh mul icolou ed ligh ing ing is used
o indica e he di e en s a es o he i-Walke : calib a ion s a us and ba e y le els. Handleba s
also embody a ib a o ha can be used as a hap ic de ice o enhance he in e ac ion be ween
he use and he i-Walke , e.g. when he use is ge ing oo close o a wall o an obs acle. This
ligh ing ing can be handy o use s wi h isual o cogni i e impai men s. In bo h cases, he in-
ensi y and du a ion o he lashes o ligh ning/ ib a ions can be de ined o e e y ype o signal.
Rea wheels
The ea wheels (10) ha e been modi ied by adding a mo o and elec onics in each one. The
ex e nal s uc u e o he wheel is main ained, bu he in e nal pa is edesigned, changing he
axis o o a ion o couple wi h he mo o . The ea wheels collec odome ic in o ma ion om
h ee Hall-e ec senso s in eg a ed inside each engine. The ga he ed da a is use ul o calcula e
he (X,Y) use ’s posi ion, as well as he longi udinal and o a ional speeds. In he end, hese
mo o s wo k by helping he use o mo e a ound sa ely, especially in up and down slopes.
The mo o s can wo k in ou di e en ope a ing modes desc ibed in Table 4.2. The eac i e
con ol (see §4.2) uses hese di e en modes depending on he amoun o pushing/pulling help
i is equi ed in e e y momen .
Cen al box
The cen al box con ains a Raspbe y Pi, he ba e y and h ee new senso s ha ha e been added
o ha e a comple e Ine ial Measu emen Uni (IMU):
39
4. THE I-WALKER
Ope a ing Mode Desc ip ion
FREE No ac ion applied on mo o . The wheel u ns eely,
like a con en ional wheel. I is use ul when we wan
he wheel o u n eely, bu i also main ains access
o in o ma ion abou kinema ics, empe a u e, e c
AS (Ac i e Speed con-
ol)
The mo o is d i en by a se poin o angula speed.
Use ul when any use is no using he i-Walke bu
we wan i o beha e like an au onomous ehicle.
AC (Ac i e Cu en con-
ol)
The mo o is d i en wi h an ac ual se poin p opo -
ional o a o que se poin . Use ul o compensa e
he o ces equi ed om he use .
AB (Ac i e B aking con-
ol)
The wheel is b aked by se ing a cu en se poin
ha opposes o he mo emen .
Table 4.2: i-Walke Mo o ope a ing modes.
•Gy oscope: de ec s he h ee angula speeds o o a ion. This in o ma ion s eng hens he
da a ga he ed om odome y, ha ing mo e p ecise moni o ing o he use ’s mo emen s.
•Accele ome e : de ec s i he use is walking on a slope and/o he i-Walke is being
accele a ed.
•Magne ome e : de ec s he p ojec ion o he magne ic ea h iled on he in eg a ed ci cui .
4.2 Reac i e con ol
The mos inno a i e ea u e o he i-Walke is he design o a eac i e con ol, de eloped a
he Au oma ic Con ol Depa men (ESAII-UPC), which p o ides mobili y aid o he use . The
amoun o helping o ce and b aking o ce in each hand ough bo h o be de e mined p e iously
by a clinician.
The i-Walke pla o m p o ides ou p ima y se ices. Th ee a e ela ed o elde /impai ed
assis ance; we use he ou h o da a logging. A physio he apis should plan all he suppo
gi en o a use . Se ices p o ided a his momen a e:
•Ac i e mo o assis ance o compensa e lack o muscle o ce on climbs.
•Ac i e b ake assis ance o compensa e lack muscle o ce on descen s.
40
4.2 Reac i e con ol
Figu e 4.2: Example o le -hand o ce compensa ion.
•Ac i e di e en ial assis ance o compensa e unbalanced muscle o ce.
•Reco ding o senso measu emen s and ac ua o s ac i i ies o la e e alua ion (le and
igh -hand o ces, no mal o ces, il and odome y)
Desc ibed s a egies a e no exclusi e: we can ha e he use pushing he i-Walke going down-
hill and a he same ime he i-Walke elie ing him om pa o he necessa y pulling/pushing
o ce o mo e a ound. Fo sa e y easons he i-Walke au oma ically s ops when he use e-
leases he handle s, ha is when no o ces a e de ec ed on hem.
Hand o ce compensa ion
The eac i e con ol wo ks wi h wo p ima y a iables: λand ν. The o me ep esen s he
amoun o helping o ce ha he use ecei es, while he la e one is a helping b ake o ce. The
combina ion o bo h pa ame e s allows he he apis s o c ea e a pa ien ’s ailo ed con igu a ion.
These pa ame e s a e se du ing he ini ial se up o he i-Walke by a physio he apis .
41
4. THE I-WALKER
Figu e 4.3: Hand o ce compensa ion s a egies in an uphill scena io.
When a use is going uphill wi h he i-Walke , he λpa ame e will elease he use om
pa o he pushing o ce s/he has o do. In Figu e 4.2, ed a eas ep esen he longi udinal o ce
ha he use should be exe ing in he gi en scena io. The g een a eas co espond o he o ce
compensa ion p o ided by he i-Walke . Du ing he uphill, he olla o will compensa e he lack
o pushing o ce; when going downhill, i will assis he use by sending ac i e b eak signals
o he ea wheels. This compensa ion will a oid excessi e e aining o ce (i.e., nega i e
longi udinal o ce), and hus a possible all. The aid p o ided in downhill is independen o he
alue o λ, as i is a sa e y mechanism ha aims o p e en alls.
The i-Walke allows o de e mine di e en amoun s o compensa ion s a egies in each side
o he body acco ding o he use ’s dys uncions. Figu e 4.3 shows ha he con igu a ion o λ
and νpa ame e s is independen on each handleba (e.g., an indi idual wi h hemipa esis needs
di e en help on each hand). In his case, he use ecei ed helping o ce on he le hand
( he di e ence be ween he pink a ea and he do ed line is he con ibu ion o he i-Walke ’s
42
4.3 The i-Walke ’s Assis i e En i onmen
Figu e 4.5: Dend og am ep esen ing he clus e ing applied o Righ Hand Fo ce X (longi udi-
nal o ce h x).
simila me hodology wi h a mul i- a iable app oach o iden i y signi ican da a usion use ul o
lea n new concep s on an indi idual’s walking beha iou . Mo eo e , i would be o pa icula
in e es o iden i y a iables ha oge he can de ine di e en gai dis u bances and associa e
hem wi h he pa hologies a ec ing elde ly people.
A simila a me hodology was used o classi y use s in o wo g oups o age. This ime, he
selec ion o he s udy included 42 indi iduals o ages be ween 22 and 94, which consis ed in
pe o ming a 3-minu e walking es in an indoo co ido o 40 me es. The wo k has been
de eloped by Ojeda (2018) as he Mas e hesis, which I ha e co-di ec ed, and esul s will be
p esen ed in Ojeda e al. (2018).
The me hodology uses he Bag-o -SFA-Symbols (BOSS) model o he ep esen a ion and
49

4. THE I-WALKER
Figu e 4.6: BOSS Model is used o indexing and ep esen a ion and ans o ms he ime se ies
in o BOSS his og ams
indexing. This me hod ans o ms he nume ic ime-se ies in o a bag o wo ds ep esen a ion
o la e c ea e a k-dimensional ma ix. The BOSS model desc ibes ime se ies as an uno de ed
se o subs uc u es using Symbolic Fou ie App oxima ion (SFA) wo ds. The BOSS low
model is ep esen ed as ollows (see Figu e 4.6): pa ame e s de ini ion, windowing, SFA and
his og ams agg ega ion and educ ion (Sch¨
a e (2015)).
Rep esen a ion and indexing
The goal o his s ep is o ans o m each ime se ies o a new ep esen a ion based on a ocab-
ula y ex ac ed om he beha iou o he da a in he equency domain.
This ans o ma ion has a se o pa ame e s ha need o be uned in o de o ind he op imal
con igu a ion: leng h o he windows, he numbe o Fou ie coe icien s o a window and he
numbe o le e s o he wo d ep esen a ion.
Fo each ime se ies, a se o ixed-size windows a e gene a ed using a sliding window o
leng h w. The i s window begins a 0 and ends in he posi ion w, he second one will o se
one posi ion ending a posi ion w+1 and so on un il he end o he se ies is eached. The esul
will be ans o med in o a ocabula y whe e each window ep esen s a wo d. The leng h o he
wo ds a e de ined by he quan iza ion o he Fou ie coe icien s o he window as explained in
he ollowing pa ag aphs.
The ans o ma ion o he windows in o a ocabula y uses he SFA me hod, which aims
o simpli y in o ma ion by emo ing he unnecessa y da a and keeping only he mos ep e-
50
4.3 The i-Walke ’s Assis i e En i onmen
sen a i e cha ac e is ics. The SFA pe o ms h ee s eps o achie e i s goal: app oxima ion,
quan iza ion and wo ds compu a ion.
The SFA uses a disc e iza ion me hod based on he Momen a y Fou ie T ans o m (MFT),
o he app oxima ion p ocess. The MFT a ge s o keep he c i ical da a ex ac ing he Fou ie
coe icien s o he signal (Alb ech e al. (1997)). This me hod allows o inc emen ally compu e
he i s Fou ie coe icien s o a sliding window in a se ies in e icien manne .
The idea behind his disc e iza ion is o decompose he ime se ies in o wo basic ype o
unc ions (Sch¨
a e and H¨
ogq is (2012)). The o me consis s in iden i ying slow changes in
he da a, while he la e iden i ies apid changes. Fo he app oxima ion, hose wi h slow
changes a e enough o a ai desc ip ion o he signal. These ypes o unc ions also p o ide a
smoo he signal wi h low pass il e ing.
The decomposi ion ep esen s a ime se ies by i s Fou ie coe icien . The magni ude o he
coe icien ep esen s he ampli ude o he signal. The p oposed app oxima ion uses only he
i s coe icien s. The i s Fou ie coe icien can be op ionally disca ded because i s ands
o he mean alue o he signal, ob aining his way o se in a iance.
The quan iza ion s ep educes he g anula i y o he da a by di iding he alues o he
Fou ie coe icien s in o a his og am o equal equency bins (Gu ajada and S i as a a (1991))
and mapping each coe icien o i s ange. To de ine hese his og ams, a numbe o bins is
de ined as a pa ame e , ep esen ing he le e s o he disc e iza ion alphabe ha will be used
o compu ing he wo ds ep esen ing he windows ex ac ed om he ime se ies. Each posi-
ion o he Fou ie coe icien s is disc e ized sepa a ely. This is done by compu ing Mul iple
Coe icien Binning (MCB), and i aims o minimize he los in o ma ion when pe o ming he
disc e iza ion p ocess.
The disc e iza ion wo ks as a map ha con ains in e als o nume ic alues pe each coe -
icien . The ou come o he SFA is a wo d o le e s pe window and a se o wo ds pe ime
se ies. In o de o a oid a bias due o la ge pe iods o s able signal, he BOSS model educes
he leng h o he ocabula y by nume osi y educ ion, emo ing iden ical consecu i e wo ds.
The inal s age o he BOSS model is o ans o m he ocabula y o a ime se ies in o a
his og am o ela i e equencies o wo ds. This ans o ms he se ies in o a ec o , so we can
compa e di e en se ies using di e en simila i y measu es. The main ad an age o his inal
ans o ma ion is o allow he compa ison o ime se ies o di e en leng hs.
Simila i y Measu e and Space Embedding
The nex s ep is o e alua e how simila a e he se s o ime se ies. F om he possible simila i y
51
4. THE I-WALKER
unc ions ha can be de ined, he one ha be e i s he model is he cosine simila i y measu e
because his me ic conside s only he o ien a ion o a ec o and no he magni ude (S einbach
e al. (2000)); his cha ac e is ic is use ul when wo king wi h symbols ins ead o numbe s. The
ou pu o his p ocess is an a ini y ma ix ha will be u he ans o med be o e clus e ing he
da a.
Fo his model, a Spec al Embedding (S ange and Zwiggelaa (2014)) is applied o he
a ini y ma ix o embed he da a in a me ic space and o enhance he ele an cha ac e is ics
by using a non linea ans o ma ion. This embedding ans o ms he a ini y ma ix in o a
k-dimensional ma ix. The numbe o dimensions kis pa o he con igu a ion pa ame e s. Fo
his pape , he ans o ma ion was limi ed a mos o h ee dimensions.
Clus e ing
A Bayesian Gaussian Mix u e Model (BGMM) wi h Di ichle p io s (Li and Mihaylo a (2017);
Sanjay-Gopal and Hebe (1998)) is used o pa i ioning he da a ep esen ed by he k-dimensional
da a ma ix. A Gaussian mix u e model is a p obabilis ic model ha assumes all he da a poin s
a e gene a ed om a mix u e o a ini e numbe o Gaussian dis ibu ions wi h unknown pa-
ame e s. Each clus e is o med wi h a se o poin s ha shape a Gaussian dis ibu ion using a
Expec a ion Maximiza ion (EM) algo i hm. The use o a Di ichle p io includes he de e mi-
na ion o he numbe o clus e s in he op imiza ion p ocess.
The EM o mix u e models consis s o wo s eps. The i s s ep calcula es he expec a ion
o he componen o each da a poin gi en by he model pa ame e s. The second phase maxi-
mizes he expec a ions calcula ed in he p e ious s ep conce ning he model pa ame e s. Then
hese wo s eps a e epea ed un il he esul con e ges.
E alua ion
To assess he quali y o he clus e ing, i is necessa y o apply some e alua ion echniques. This
app oach conside s wo echniques: (i) he adjus ed and index (ARI, de Va gas and Bed ega
(2013)) o measu e he s abili y o he clus e ing o andom ini ializa ion and (ii) he Silhoue e
index (SI, Ve ma e al. (2015)).
Scena ios
The wo k is based on he ecogni ion o he pa e ns o he o ces applied by he use s o he
i-Walke while walking. F om he o ces eco ded by he i-Walke , he e ical ones seem o
be he mos ela ed o he indi iduals0compensa ion s a egies. The e o e, his s udy examines
wo scena ios. The i s model is p o ided wi h all he o ces o ind i he ans e sal and
longi udinal o ces add ele an in o ma ion. The second one conside s only he e ical o ces
52
4.4 Summa y
o s udy whe he he ans e sal and longi udinal o ces add noise ins ead o con ibu ing o he
esul s.
Resul s
This me hodology has been es ed wi h di e en ini ial se ing pa ame e s o c ea e he ocab-
ula y. A e inding he sui able combina ion o pa ame e s, he clus e ing was applied o he
wo abo e-men ioned scena ios. Resul s show ha he p oposed app oach is able o di ide
pa icipan s by age using he applied o ces o he i-Walke . Mo eo e , i can be obse ed how
he s anda d de ia ion o he leaning o ces inc eases wi h age, which migh be an indica o
o he loss o balance ha people p esen s as hey age. Figu e 4.7) shows ha he a iabili y
is di ec ly ela ed wi h he magni ude o he e ical o ce. The esul s ob ained a e cohe en
wi h he li e a u e: (i) gai eloci y is educed wi h ageing, implying a highe isk o alling;
(ii) olde adul s wi h di e en so s o pa hologies p esen an abno mal gai ha is iden i ied in
a pa icula clus e and, mo eo e , is ela ed o unbalanced use o o ces, which migh lead o
dis unc ional gai . In addi ion, clus e ing esul s we e mo e accu a e when combining he he
h ee esul ing o ces om he handle senso s han using he e ical o ce as single- a ia e
clus e ing.
4.4 Summa y
Resea ch p o o ypes, like he i-Walke , a e beginning o achie e he pe o mance needed o
make a di e ence in he daily li e o he elde ly socie y. By he momen , he ma ke s ill o e s
only limi ed solu ions o subs an ially p olonging he ime ha olde adul s can li e indepen-
den ly a home, bu hese should be suppo ed wi h ele an heal h and social ca e se ices in
an in eg a ed manne (see Ba u´
e(2012), Wasson e al. (2008)). Olde adul s a e becoming
a p edominan aspec o ou socie ies and a e expec ed o subs an ially a ec he economy a
wo ld-wide le el, changing he pa adigm o public heal h sys ems i o de o main ain hem as
sus ainable as possible. As such, solu ions bo h e icacious and cos -e ec i e need o be sough
(Sun e al. (2014)). In pa icula , he e is a need o imp o e mo e he cos - o-bene i a io o
obo -assis ed he apy s a egies and hei e ec i eness o ehabili a ion he apy.
I is clea ha he i-Walke is an assis i e de ice ha o e s many possibili ies o clinicians,
pa ien s and ela i es, and i could become a powe ul ool o he design o he incoming
digi al heal h solu ions. I has been designed o collec da a o long- e m pe iods, which could
be use ul in solu ions in ol ing he emo e moni o ing o elde people pe o ming hei ADLs.
53
4. THE I-WALKER
Figu e 4.7: Compa ison o he e ical o ce and i s a iabili y pe clus e and age.
54

4.4 Summa y
Au ho s in Ba u´
e e al. (2015) ha e al eady shown he alue o elde s’ ac i i y ecogni ion and
moni o ing since i migh be an ea ly symp om o some decline ha mus be p e en ed as soon
as possible.
The i-Walke can be easily in eg a ed wi h o he s akeholde s o assis i e de ices o sha e
in o ma ion, gene a e epo s o ale s and p ocess speci ic da a (see Appendix §B). I has also
been designed o o e cons an mobili y assis ance o he use , p o iding sa e y and ensu ing
compensa ion on he uppe limbs o ces.
The objec i e o his PhD is o combine di e en ou pu s o he i-Walke ’s senso s and
biological da a ob ained by clinicians h ough cogni i e and physical assessmen s o ex ac
some cha ac e is ics o hei gai and y o iden i y di e en use s p o iles. This analysis will
iden i y he s ides pe o med by indi iduals while walking and y o ind g oups o people who
walk simila ly. The exis ence o hose p o iles could allow he i-Walke o p o ide ailo ed
assis ance while i lea ns new ea u es o he pa ien . The main di e ence wi h he esea ch
p esen ed in his chap e is ha ins ead o analysing exe cises as a whole, i will use he o ce
used a each s ide pe o med du ing an exe cise o lea n how do people walk and use he i-
Walke . In §6we i s desc ibe he me hodology used in his PhD p oposal o s udy he walking
beha iou esul ing om he in e ac ion be ween he use and he i-Walke . Then a model o
all isk p edic ion is p esen ed and es ed wi h a g oup o pa icipan s o he I-DONT-FALL
p ojec (see §5.3.1).
55
4. THE I-WALKER
56
Chap e 5
Clinical Tes s: design and
implemen a ion o a pilo p o ocol
Heal h esea ch has become an essen ial mains ay o o imp o e he ac ual s a e o he a in
i s di e en ields, such as epidemiology, biomedicine, heal h se ices, pe sonalised medicine
as well as i s socio-economic impac on socie y. One o he mos common o ms o heal h
esea ch is he clinical ial, whe e olun ee indi iduals pa icipa e in s udies o assess a new
medical p oduc , de ice o ea men . Mos o his esea ch is done h ough collec ing di e en
ypes o pe sonal da a, om biological cha ac e is ics o physical and/o cogni i e assessmen s
among o he s. Du ing he las yea s, i has also included new da a coming om senso s ha
will in e ac di ec ly wi h he indi idual (e.g., h ough bio-me ic o wea able senso s) o
en i onmen al senso s. Al hough heal h esea ch aims o p omo e indi iduals’ heal h and well
being, as well as o imp o e ca e and social se ices, i s managemen migh also ep esen a
isk o he socie y due o he ype o da a ha is s o ed and p ocessed.
E hics is an essen ial dimension o human philosophical esea ch, conside ed bo h as disci-
pline and p ac ice. Philosophe s oday usually di ide e hical heo ies in o h ee gene al subjec
a eas: me ae hics, no ma i e e hics, and applied e hics. Fo clinical esea ch, e hically jus i ied
c i e ia o he design, conduc , and e iew o clinical in es iga ion can be iden i ied by obli-
ga ions o bo h he esea che and human subjec (see Ai a and Riche (2005); Gu aya e al.
(2014); Wo ld Medical Associa ion (2001)). In o med consen 1, con iden iali y, p i acy, p i i-
1In o med consen e e s o an e hical and legal doc ine based on he unde s anding ha all in e en ions
(diagnos ic, he apeu ic, p e en i e, o ela ed o scien i ic s udies) in he medical ield should only be pe o med
a e a pa icipan has been in o med abou he pu pose, na u e, consequences, and isks o he in e en ion and has
eely consen ed o i , see Glickman e al. (2009).
57
5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
leged communica ion, and espec and esponsibili y a e c i ical elemen s o e hics in esea ch.
The Eu opean Commission has published se e al documen s ha desc ibe, om an e hical
and egula o y poin o iew, he di e en aspec s ha should be espec ed when wo king in a
clinical ial wi h human subjec s (see CAREGIVERSPRO-MMD (2016)).
The da a collec ed and analysed du ing his PhD wo k comes om di e en hospi als and
ca e cen es dis ibu ed in Spain and I aly. As a esul , we buil ou da ase s, al hough we
excluded he i s one om he analysis. The espec i e E hical Commi ee app o ed each pilo
es ollowing he Clinical T ial Di ec i e 2001/20/EC (see Council o Eu opean Union (2001)),
and was suppo ed by clinicians in he pa icipan s’ selec ion and ec ui men and he execu ion
o physical and cogni i e assessmen s speci ied in each p o ocol.
5.1 De ini ion o a p o ocol
Acco ding o he In e na ional Con e ence on Ha moniza ion (see ICH (1996)) a p o ocol is a
documen ha desc ibes he objec i e(s), design, me hodology, s a is ical conside a ions, and
o ganisa ion o a ial. A clinical ial p o ocol should include:
•Gene al In o ma ion abou in es iga o s and sponso s (i.e., names and con ac s).
•Backg ound In o ma ion o he es ed p oduc (i.e., i s po en ial isks and bene i s, he
desc ip ion o he a ge popula ion o e e ences o he li e a u e ha a e ele an o he
ial).
•T ial Objec i es and Pu pose.
•T ial Design should include: (i) p ima y endpoin s and he seconda y endpoin s, i any,
o be measu ed du ing he ial; (ii) a desc ip ion o he ype o ial (e.g., double-
blind, placebo-con olled o pa allel design); (iii) a desc ip ion o he measu es aken
o minimize/a oid bias (e.g., andomiza ion o blinding); (i ) a desc ip ion o he ial
ea men ; ( ) he expec ed leng h o he subjec ’s pa icipa ion and a desc ip ion o he
du a ion o all ial pe iods and ollow-up.
•Selec ion and Wi hd awal o Subjec s de ines he subjec inclusion and exclusion c i e ia,
as well as he wi hd awal c i e ia and p ocedu e.
•Assessmen o he p oduc includes me hods, iming and scales o assessing, eco ding
and analysing he e icacy and sa is ac ion pa ame e s.
58
5.2 P o ocol design
Figu e 5.1: Map o he Fondazione San a Lucia en i onmen and he wo i s d i ing es s.
Figu e 5.2: Map o he Fondazione San a Lucia en i onmen and he wo las d i ing es s.
65

5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
Figu e 5.3: Map wi h he h ee scena ios o MAD pilo .
c ea ed con usion among some pa icipan s. On he o he hand, he s aigh lines we e oo sho
o ex ac any signi ican gai ea u e. Fo his eason, i was decided o imp o e he ambula o y
exe cises hanks o he expe ience ob ained du ing he FSL pilo .
This is why we decided o subs i u e his es wi h he 10MWT, which is a well-accep ed
pe o mance measu e used o assess walking speed in me e s pe second (m/s) o e a sho
dis ance. 10MWT can also be employed o de e mine unc ional mobili y, gai and es ibula
unc ion. As i is a widely used measu e, i is easy o ind in o ma ion abou gai speed acco d-
ing o di e en p o iles (i.e., heal hy adul s, hip ac u e, s oke, e c.)1. Figu e 5.3 shows he
upda ed pa hs ha will be used in u u e pilo s.
5.2.4.1 Ten Me e Walking Tes
The adi ional 10 Me e Walk Tes (10MWT) is a pe o mance measu e used o assess walking
speed in me e s pe second o e a sho dis ance. I can be employed o de e mine unc ional
mobili y, gai and es ibula unc ion (see Adell e al. (2013)). Conclusions a e made om an
expe ’s obse a ion.
The in ended popula ion is anged as: p eschool child en (2-5 yea s), child en (6-12 yea s),
adolescen s (13-17 yea s), adul s (18-64 yea s), elde ly adul s (65+) wi h a ange o diagnoses
including, among o he s:
1h p://www. ehabmeasu es.o g/
66
5.2 P o ocol design
•Acqui ed B ain Inju y
•Ge ia ics
•Hip F ac u e
•Lowe Limb Ampu a ion
•Mo emen Diso de s
•Mul iple Scle osis
•Pa kinson’s Disease
•Spinal Co d Inju y
•S oke
•T auma ic B ain Inju y
The 10MWT is pe o med by an indi idual who is ins uc ed o walk wi hou assis ance
in a dis inc pa hway whe e a s aigh line o 10 me e s is ma ked o . Addi ional miles ones
we e placed a me e s 2 and 8. The clinician indica es o he use when s/he can s a walking
and will measu e he ime used o comple e he dis ance. Usually, he clinician will only ake
in o accoun he ime spen du ing he in e media e ma ked 6 me e s o allow some space o
accele a ion and decele a ion (see Figu e 5.4). The S a and Finish poin s a e indica ed o he
use be o e he es execu ion, bu no lines a e ma ked on he loo as a guide. The imed zone
co esponds o he six cen al me e s o he exe cise. The use o assis i e de ices is pe mi ed
bu mus be kep consis en and documen ed o each es . The 10MWT only assesses walking
speed and does no conside he amoun o physical assis ance equi ed de ices o endu ance.
The 10MWT is no app op ia e i he indi idual needs physical assis ance o ambula e.
Figu e 5.4: The 10 Me e Walk Tes measu emen
In he 10MWT he subjec has o walk 10 me e s in a maximal s aigh way, as abo e said
he only measu emen ool is a s opwa ch (see Ali and Raad). In addi ion o his, he i-Walke
can measu e he a elled dis ance, he du a ion ime, he maximal speed, he maximum o he
la e al de ia ion and he p essu e on he handle s. F om he addi ional pos -p ocessing, i is
possible o ex ac new indica o s.
O he esea che s ha e been using obo ic olla o s (see o example Wang e al. (2014) and
Balles e os e al. (2015)) and/o o he au oma ic means o measu e use s’ pe o mance in he
67
5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
10MWT (see Yo ozu e al. (2015)). In o he cases, kinema ic da a om in a- ed came as a e
eco ded o s udy he walking beha iou o indi iduals (see Nooijen e al. (2009)).
The adi ional 10MWT is measu ed by he ime du a ion o he pe o mance and he e-
sul ing a e age speed o he pa icipan . I is conside ed ha gai speed can be used as a use ul
ou come o assess he physical condi ion o olde adul s as well as o p edic he isk o alling,
o he heal h de e io a ion o e en su i al (see S udenski e al. (2011)). The e is a gene al belie
ha gai speed is ela ed wi h ageing p ocess (such as muscle weakness o balance loss), al-
hough esul s a e unce ain on his heo y due o biological and beha iou al di e ences among
he popula ion (see Shimada e al. (2010)).
5.2.4.2 Timed Walking Tes s
Timed walking es s measu e he dis ance an indi idual can walk du ing a gi en ime (usually 6
minu es, named 6mWT) on a la su ace a sel -paced. The 6mWT i s applied in ail elde ly
pa ien s 60-90 yea s o age e e ed o a ge ia ic hospi al, and i a ge s communi y-dwelling
ail elde s. Howe e , he es has been used in he s udy o a a ie y o ch onic disease adul s
(Annega n e al. (2012)) o heal hy adul s (Ha ada e al. (1999)). In his las case, au ho s show
ha ac i e olde use s ob ained be e pe o mances o 6mWT han non-ac i e heal hy use s.
Mo eo e , i could be used o p edic mo bidi y and mo ali y.
The 6mWT was used in he I-DONT-FALL p ojec as an assessmen along wi h o he me -
ics o measu e pa icipan s isk o alling a p e and pos - ea men . The objec i e o his es
was o e alua e he e ec i eness o he di e en ea men s (mo o , cogni i e, mixed, placebo)
o dec ease he numbe o alls, isk o alling and ea o alling.
Fo he las phase o da a collec ion, a new g oup o adul s pe o med a educed e sion o
he 6mWT, he 3mWT. This exe cise was pe o med in an indoo co ido o 40 me e s leng h
as depic ed in Figu e 5.5. The main ins uc ions gi en o he pa icipan s o pe o m he es
we e:
•Walk o h ee minu es along he co ido .
•Do no d op he handle s while pe o ming he exe cise.
•Tu n when eaching he end o he hallway o con inue wi h he walking ajec o y.
•I he ime inishes and he pe son is in he middle o he co ido keep walking un il
e u ning o he s a ing poin .
68
5.3 Pilo s
Figu e 5.5: The 6 minu es Walk Tes measu emen .
The S a and Finish poin s a e indica ed o he use be o e he execu ion o he es , bu
no lines a e ma ked on he loo as a guide. The imed zone co esponds o he six cen al
me e s o he exe cise. This es is expec ed o p o ide u he knowledge on gai a iabili y
and walking pa e ns due o i s highe ime and e o consump ion in compa ison wi h he
10MWT (Hausdo (2005b)).
5.3 Pilo s
In his sec ion, we p o ide some highligh o each pilo si e. The an h opome ic cha ac e is ics
o all he pa icipan s in ol ed in his s udy a e depic ed in Table 5.2. Since we we e collec ing
da a om hospi als and ca e cen es, i was ha d o ind a balance in age, being mos o he
pa icipan s om he baseline aged o e 80 yea s. Besides women we e, in gene al, mo e will-
ing o pa icipa e. Hence hei ep esen a ion is highe . A he ime o designing he p o ocol,
we aimed o collec balanced handedness da a, al hough his was almos impossible. I was
quickly ema ked ha a he beginning o he las cen u y, schools end o co ec le -handed
child en, hus nowadays mos o he old age people a e igh -handed.
5.3.1 IDF Pilo
The I-DONT-FALL da ase is composed o h ee o he pilo s in ol ed in he p ojec : FSL, HGG
and SERMAS om I aly and Spain. As men ioned be o e, pa icipan s in hese pilo s sha e he
cha ac e is ic o ha ing su e ed a leas one all du ing he yea p e ious o he expe imen al
phase. Pa icipan s o he I-DONT-FALL p ojec whe e assessed physically and cogni i ely
be o e he beginning o he h ee-mon hs aining (mo o , cogni i e, mixed o placebo) and
a he end o his pe iod (see §3.3.3). The gene a ed da ase includes biological and clinical
69
5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
Cha ac e is ics
IDF
(N= 85)
N(%) o
Mean ±SD
Baseline (MAD)
(N= 60)
N(%) o
Mean ±SD
Baseline (FSL)
(N= 30)
N(%) o
Mean ±SD
Age
<80 yea s
≥80 yea s
To al
32 (37.64%)
53 (62.36%)
82.53(±8.47)
8 (13.33%)
52 (86.67%)
87.58(±5.79)
29 (96.67%)
1 (3.33%)
70.97(±5.53)
Gende
Male
Female
26 (30.58%)
59 (69.41%)
21 (35%)
39 (65%)
13 (43.33%)
17 (56.67%)
MMSE 25 (±3) 25.79 (±2.97) 28.5 (±1.59)
Tine i 17 (±4) 16.56 (±3.90) 24.47 (±3.77)
Ba hel 79 (±19) 81.95 (±14.77) 96.33 (±6.15)
Table 5.2: An h opome ic Cha ac e is ics o he S udy Pa icipan s.
70

5.3 Pilo s
C i e ia Va iable Mean S d De
Elde ly Age 74.7 7.9
Fo mal Educa ion Yea s 9.9 4.2
High Risk o Fall POMA o al 19.8 5.5
High Risk o Fall P e ious Falls 1.33 1.16
Non-demen ed MMSE 25.8 4.6
Table 5.3: Inclusion C i e ia o he I-DONT-FALL inal da ase . Resul s unde 21 o he
POMA analysis along wi h mo e ha one all in he ecen yea a e c i e ia used o de e mine
high isk o alling.
da a along wi h he i-Walke measu emen s o he 10MWT exe cises pe o med a p e and
pos - ea men phases.
Du ing he cleaning p ocess, we had o exclude se e al pa icipan s who passed he in-
clusion c i e ia bu had inaccu a e senso eadings (p obably due o a bad calib a ion a he
beginning o he exe cise, a w ong exe cise agging o a ailu e in he communica ion sys em
o he i-Walke ), emaining only 87 in o al om he h ee pilo s. Table 5.3 shows he inclu-
sion c i e ia pa ame e s o he I-DONT-FALL p o ocol and he mean and s anda d alues o he
s udied popula ion.
5.3.2 MAD Pilo
This pilo ook place in Ma ch 2015 du ing wo weeks a Los Nogales (MAD) cen e, in
Mad id. People pa icipa ing in his pilo we e olde adul s li ing in a esiden ial ca e cen e,
wi h educed mobili y bu good cogni i e condi ions, o ela i es. A o al o six y indi iduals
pa icipa ed in he s udy, wi h only one d op-ou . Al hough he objec i e was o collec da a
only om heal hy olde adul s wi h no alls, inally 15 o hem p esen ed a all du ing ha yea
(see Figu e 5.6 o an o e iew ep esen a ion o MAD popula ion). The exe cises included in
he inal s udy we e he 10MWT pe o med wi h no helping pa ame e s (λ= 0). The clinical
eam a Los Nogales p o ided he same assessmen esul s o comple e he i-Walke da a.
As i can be eadily obse ed in Table 5.2, pa icipan s om he MAD pilo a e appa en ly in
a be e cogni i e condi ion and p esen highe sco es in independen li ing despi e being olde
han he IDF popula ion. Mo eo e , al hough mos o he MAD pa icipan s did no su e any
all, hey p esen a highe isk o alling in he Tine i Scale han indi iduals om IDF. In he
case o FSL. Al hough hal o he popula ion had a p e ious all, esul s om physical and
71
5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
Figu e 5.6: Dis ibu ion o MAD pa icipan s ep esen ed demog aphically by age and gende ,
bu also clinically by numbe o alls and isk o alling
cogni i e scales a e much be e han he o he g oups. This is p obably ela ed o age ac o
since only one pa icipan om FSL was olde han 80 yea s old. The educ ion o au onomy
and mobili y ha olde adul s usually ace in esiden ial cen es could explain his dis ibu ion.
5.3.3 CV I Pilo
This las pilo was collec ed in July 2017 du ing wo weeks a he Cen e de Vida Independen
(CVI) in Ba celona. People in his pilo a e qui e he e ogeneous ega ding age, which goes
om 22 up o 94 yea s old. This da ase is composed o 42 pa icipan s, 18 o hem p esen ing
ca diac p oblems o inju ies caused by a all. The clinical eam ha suppo ed us assessed his
g oup bo h physically and cogni i ely, bu he es o he olun ee s we e conside ed as heal hy
indi iduals wi h no isk o alling (all o hem we e au onomous non- e i ed adul s, wi h no
p e ious alls o neu ological no physical issues).
This ime, he exe cise pe o med in his s udy was a sho e sion o he 6mWT. The du-
a ion o he exe cise educed by ecommenda ion o he clinical eam a CVI o h ee minu es
(3mWT). The objec i e was o collec mo e ex ended bags o s ides o each pa icipan since
he 10MWT exe cises whe e oo sho o ob ain a good p o ile o each pe son. We execu ed
he pilo in a long indoo co ido o mo e han 40 me es; once pa icipan s eached he end o
he hallway, hey we e indica ed o u n a ound and keep walking o he s a ing poin un il he
ime ends. The co ido had an a ea wide enough o allow ound u ns wi h a olla o .
72
5.4 Summa y
5.4 Summa y
In his sec ion, we ha e desc ibed he an h opome ic cha ac e is ics o he pa icipan s in he
di e en clinical ials. We aimed o c ea e a baseline wi h simila c i e ia han in I-DONT-
FALL since i is essen ial o isola e an h opome ic e ec s on gai analysis (P akash e al.
(2015)). In ou case, he only di e en ia ion among ials is he numbe o alls. Pa icipan s
in bo h clinical es s used an i-Walke wi h he e ical posi ion o handle s adap ed o hei
heigh . A b ie desc ip ion o he pilo (s a and end poin , numbe o epe i ions and led ligh s
unc ion) was gi en be o e he olun ee s signed he p o ocol ag eemen .
Du ing he es s, we ied o le hem na iga e as independen as possible, al hough some
indi iduals p esen ing isual impai men s ecei ed some indica ions when u ning and ge ing
in o a oom. We also had some issues wi h indi iduals p esen ing hea ing impai men s, es-
pecially when we needed hem o s op walking. In some occasions, hese exe cises whe e
epea ed. In he case o people s opping la e han equi ed, da a was conse ed bu hen p o-
cessed o exclude unnecessa y in o ma ion (see §6).
Al hough I al eady had pa icipa ed in o he clinical ials be o e his wo k, being in ol ed
a all s ages, om he design o he p o ocol o he es execu ion and la e p ocessing has
been challenging bu has also p o ided me new expe ise in he design and managemen o
his kind o expe imen al esea ch ield. Wi h e e y clinical es , new lessons we e lea n
om p e ious mis akes, which helped o une he p o ocol. Also, when dealing wi h a obo ic
de ice, pa icula a en ion mus be gi en o he p ecision o he execu ion o he exe cises o
he calib a ion o he senso sys em. Many in e nal o ex e nal ac o s migh a ec he p ope
collec ion o da a and u he esul s and analysis.
A o al o 632 indi iduals ha e used he i-Walke , al hough only 168 o hem will be in-
cluded in he analysis p oposed in his hesis. The size o he da ase is conside ably good o
alida e he me hodology p esen ed he e, bu i would be sui able o inc ease i in he u u e,
especially i he objec i e is o cha ac e ise gai pa e ns by pa hologies. In ha case, he p o o-
col p esen ed should be applied o speci ic g oups o popula ion, e.g., people diagnosed wi h
Alzheime Disease, Pa kinson, pos -s oke, e c.
73
5. CLINICAL TESTS: DESIGN AND IMPLEMENTATION OF A PILOT PROTOCOL
74
6.1 Gai Analysis based on Human-Rolla o In e ac ion
•wand pco espond o he wid h and p ominences o he peaks and a e e u ned as ec o s
as well. The p ominence o a peak is he minimum e ical dis ance ha he signal mus
descend on ei he side o he peak be o e ei he climbing back o a le el highe han he
peak o eaching an endpoin . On he o he hand, he wid h o each peak is compu ed as
he dis ance be ween he poin s o he le and igh o he peak whe e he signal in e cep s
a e e ence line.
•<Name,Value >op ions speci y op ional comma-sepa a ed pai s o Name, alue a -
gumen s ha will help adap ing he local maxima iden i ica ion wi h il e ing op ions.
Name is he a gumen name, i.e., he il e applied o he signal, and Value is he co e-
sponding alue o his il e . We will use hese op ions o de ine peak il e ing, which can
be done by minimum p ominence heigh o wid h and minimum peak dis ance. O he
op ions include il e ing by a h eshold, by he maximum numbe o peaks o by peak
so ing.
•xand Fs a e op ional pa ame e s ha speci y a loca ion ec o xo a sample a e Fs o
he da a. When using his a gumen , wand locs a e gi en in e ms o xo con e ed o
ime uni s o he second case.
Ano he app oach o gai de ec ion is o use dis ance as a e e ence ins ead o ime and o ce
p ominence. In his case, he xop ional a gumen is used, which will be a ec o con aining
he inc emen al posi ion o he i-Walke du ing he exe cise. Figu e 6.4 show a Fxdi signal
il e ed by minimum peak dis ance.
Wo king wi h aw da a implies ha he signal is no smoo h and hus, he sea ch o local
maxima peaks needs o be il e ed o a oid alse posi i es. In his case, we use he dis ance
as e e ence o he indpeaks unc ion. Depending on he physical condi ion o heigh o he
pe son using he i-Walke , he dis ance be ween s eps (and hus, he numbe o s eps) a ies
signi ican ly, needing se e al s eps o il e ing o include all ypes o beha iou s in he model.
A e ying di e en app oaches, he a iable ha bes il e ed he s eps posi ions was he
a e age speed, and he o al walked dis ance o in e an es ima ed leng h o he s ep. As be o e
men ioned, he de ec ed peaks a e s eps pe o med wi h he igh oo ; be ween wo peaks o
igh s eps, he e is always a nega i e peak, which indica es when he le oo s ep has aken
place. The e o e, we can conside ha he ac ion be ween wo de ec ed posi i e peaks is a
s ide ha begins wi h he igh leg (a igh s ide om now on).
81

6. METHODOLOGY
0 1 2 3 4 5 6 7 8 9 10 11 12 12.3
Time (s)
-6
-4
-2
0
2
4
6
8
10
12
14
lh x (kg)
-5
0
5
10
15
h x (Kg)
Pushing Fo ces
lh x
h x
0 1 2 3 4 5 6 7 8 9 10 11 12 12.3
Time (s)
-15
-10
-5
0
5
10
15
h x - lh x (Kg)
Combined Pushing Fo ces
Figu e 6.3: Iden i ica ion o s eps in a 10 Me e Walk Tes exe cise. a) Indi idual longi udinal
hand o ces; b) Resul ing pushing ec o Fxdi .
A his s age, a ocabula y o s ides has been gene a ed, whe e each ins ance is a ime
se ies con aining he o ce applied a each sample ime along he s ide (i.e., e e y 100ms).
Also, he spa io empo al in o ma ion o each igh s ide is also comple ed wi h iden i ie s o
he use , exe cise and s ep numbe wi hin he exe cise, he o e all s ide leng h and ime, and
he a e age pushing o ce exe ed du ing he s ide. This se o cha ac e is ics allows o ming
a dic iona y (o ocabula y) o s ides, aiming o emula e he dic iona y o wo ds used in he
bag-o -wo ds me hodology. This echnique is well-known in ex and documen classi ica ion,
whe e he equency o each wo d is used as a classi ie ea u e. The bag-o -wo ds has also
been used in compu e ision applied o image classi ica ion, whe e image ea u es a e ea ed
as wo ds. This app oach is also known as bag-o -X due o i s adap abili y o o he ields. In
gene al, his echnique elies on iden i ying ele an key-wo ds and analysing hei equency o
appea ance. In his hesis, he s ide ocabula y will be used o cha ac e ise ypes o exe cises
82
6.1 Gai Analysis based on Human-Rolla o In e ac ion
Figu e 6.4: Fxdi signal il e ed by minimum peak dis ance. The inc emen al posi ion ec o
has been used as loca ion e e ence o he peaks.
acco ding o he occu ences o each ype o s ide. In his case, a bag-o -s ides will be used
o s udy he esul s o he clus e ing p ocess. The concep o bag-o -s eps was in oduced in
Pla e al. (2017), whe e au ho s used i o p edic he ehabili a ion leng h and discha ge da e o
a pa ien using insole o ce senso s. O he han he inal applica ion, hese wo me hodologies
di e on he ocabula y gene a ed (s ides s s eps) and he ools used o collec hem.
The es o he app oach p esen ed is de eloped wi h he aim o de e mine whe he hese
bags-o -s ides con ain in o ma ion ha allows eme ging new knowledge on walking pa e ns
on s ides o g oups o indi iduals wi h simila walking pa e ns.
6.1.3 Clus e ing Time Se ies
A ime se ies is a sequence o obse a ions collec ed a a gi en pe iod in ch onological o de .
A ime se ies da ase Dis hus a collec ion o ime se ies TS wi h he same ime base, al hough,
he numbe o obse a ions can a y among he ins ances o he da ase and is ep esen ed as D
={TS1,TS2, ..., T Sn}. T adi ional ime se ies analysis ocus on smoo hing, decomposi ion and
o ecas ing, bu hey a e la e used also o sol e clus e ing and classi ica ion p oblems. Time
se ies clus e ing is becoming popula in ecen s udies since i allows explo ing la ge amoun s
83
6. METHODOLOGY
o da a in mo e complex da a mining algo i hms such as ule disco e y, indexing, classi ica ion
and anomaly de ec ion. Time se ies clus e ing has been applied o di e en ields, such as
inances o medical esea ch ( he mos commonly used example would be in ECG analysis).
The inc easing use o senso s in many so s o s udies has also enhanced he in e es in empo al
analysis. Time se ies can be analysed on equency o ime base, depending on he expec ed
ou comes and he ype o analysis applied.
Usually, ime se ies clus e ing has wo s eps: he i s s ep consis s o wo king ou an ap-
p op ia e dis ance/simila i y measu e and hen, a he second s ep, o apply an exis ing unsupe -
ised pa i ioning echnique, such as k-means, hie a chical clus e ing, densi y-based clus e ing
o subspace clus e ing. As a esul , he ime se ies da ase Dwill be di ided in o C={C1,
C2, ..., Ck}clus e s, in such a way ha homogeneous ime se ies a e g ouped oge he based
on a speci ic simila i y measu e (Aghabozo gi e al. (2015)). The p ima y challenge on ime
se ies clus e ing elies on he high dimensionali y o da ase s and he selec ion o he simila i y
measu e. Also, ime se ies a e na u ally noisy and migh con ain ou lie s, hence i is equi ed
o apply some signal il e ing o ob ain a use ul simila i y ma ix.
Time se ies clus e ing can be applied (i) as a whole, whe e all he indi idual ime se ies a e
clus e ed conce ning hei simila i y; (ii) as a subsequence o ime se ies ob ained wi h a sliding
window; (iii) as a ime poin clus e ing, whe e ime se ies a e segmen ed. In he case o he
s udy p esen ed he e, a ime se ies clus e ing is applied o bo h baseline and alida ion da ase s.
The size is no excessi ely la ge due o he sho ime o exe cise, bu he i-Walke can wo k o
se e al hou s and gene a e la ge amoun s o da a. Mo eo e , since da a is b oken in o s ides,
each ins ance o he da ase will be p esumably sho and o di e en size. This migh p esen
a compu a ional challenge when compa ing he simila i y among hem, bu solu ions can be
ound in he li e a u e o ackle his. In adi ional clus e ing, he dis ance be ween objec s o a
da ase is calcula ed based on exac ma ch, bu his is no possible in ime se ies clus e ing due
o he na u e o ime se ies objec s, whe e sample in e als and leng hs can be i egula .
As men ioned, he e a e di e en dis ance measu es applied o ime se ies, such as Haus-
do dis ance, Dynamic Time Wa ping, Euclidean dis ance o Longes Common Sub-Sequence
among o he s. The decision on which one is mo e app op ia e elies on he clus e ing objec-
i es, i.e., he ype o simila i y among objec s ha is being s udied. Th ee p ima y goals ha e
been iden i ied: simila i y in ime, in shape o change. Finding simila i ies in ime means o
look o he simila i y a each s ep ime and is usually based on Euclidean dis ances. Howe e ,
84
6.1 Gai Analysis based on Human-Rolla o In e ac ion
due o he dimensionali y o ime se ies da ase s, i is ecommended o apply i s a ans o -
ma ion o educe he compu a ional cos . Simila i ies in change look o simila s uc u al
di e ences be ween ime se ies, i.e., ime se ies wi h iden ical co ela ion s uc u es. This
app oach is no ecommended o sho ime se ies like he ones used in his hesis.
In he case o his hesis, i aims o ind simila s ides by he shape o he pushing o ces
along he s ide. A popula me hod o compa e and iden i y pa e ns o ime se ies is he
Dynamic Time Wa ping (DTW om now on, Deza and Deza (2013)). DTW inds op imal
alignmen be ween wo- ime se ies, ega dless o he ime poin s ha each one con ains. In
ac , DTW has al eady been used in he con ex o gai analysis. Fo example, in Ba h e al.
(2015) au ho s p opose a me hodology o au oma ic single s ide segmen a ion om walking
exe cises. Sequences we e ex ac ed om ine ial mo emen senso s loca ed a he pa icipan s’
oo wea . In Boulgou is e al. (2004) au ho s use ideo sequences o gai ecogni ion and
use DTW o compa e es and e e ence gai cycles. De awi e al. (2010) p esen s ano he
me hodology using ideo sequences and wea able senso s o ecognise gai cycles and DTW is
again he simila i y measu e o compa e hem.
In his hesis, a DTW dis ance is calcula ed o each pai o s ides om he p e iously
gene a ed ocabula y o s ides, gene a ing a M×Mdis ance ma ix, whe e Mis he numbe
o objec s in he ime se ies da ase . This was pe o med in RS udio using he d w and dis
lib a ies; each DTW dis ance ma ix ook se e al hou s o be gene a ed. The esul ing dis ance
ma ix is hen used as inpu o he clus e ing, ins ead o he o iginal ime se ies, which will
sa e a massi e amoun o ime when compu ing he ime se ies clus e ing analysis. Since he
lea ning algo i hm used is unsupe ised, he clus e ing p ocess will equi e se e al epe i ions
un il inding he igh pa i ion; once he dis ance ma ix is gene a ed, he compu a ional ime
o ying di e en alues o k(pa i ions) is educed o ew seconds. In his case, he pa i ion
a ound medoids algo i hm (Kau man and Rousseeuw (1990)), also known as k-medoids, was
applied in Rs udio using clus e and p oxy lib a ies. The e m medoid e e s o an obse a-
ion wi hin a clus e o which he sum o he dis ances be ween i and all he o he membe s
o he clus e is a minimum.
The main challenge o wo king wi h an unsupe ised lea ning echnique is o de e mine
he numbe o clus e s (k) ha be e ep esen s he analysed da a. Un o una ely, he e a e no
de ini i e c i e ia o de e mine ha alue, bu i is somewha subjec i e and depends on he
me hod used o measu ing simila i ies and he pa ame e s applied o he clus e ing algo i hm.
The pa i ion a ound medoids algo i hm can be e alua ed wi h some well-known echniques
85
6. METHODOLOGY
such as he wi hin-clus e sums o squa es o he a e age Silhoue e among o he s (A belai z
e al. (2013)). The la e is an ou pu o he pam algo i hm a ailable in RS udio when pe o ming
k-medoids. I e u ns an a e age alue o each objec o he da ase , ep esen ing how well
does i lay wi hin i s clus e . Values can be ei he nega i e o posi i e, bu he highe he alues
a e, he be e is o he clus e ing quali y. Thus, he op imal numbe o clus e s kis he one
ha maximises he a e age silhoue e o e a ange o possible alues o k.
Howe e , in his case, he Silhoue e me hod migh no be en i ely sui able o e alua ing
DTW dis ances since i is based on Euclidean simila i y. The e o e, clus e ing will be execu ed
wi h di e en alues o kand assessed in he second pa o he gai analysis.
A his s age, he ime se ies clus e ing e u ns a clus e ing ec o ha assigns each s ide
o he ime se ies da ase o a g oup o s ides, ega dless o he use , hey belong o. This
app oach applies o any exe cise pe o med wi h he i-Walke wi hin a s aigh line and will
be applied o bo h he 10MWT exe cises om IDF and MAD pilo s and o he s aigh lines
ex ac ed om he 3mWT collec ed a he CVI pilo .
6.1.4 Exe cises as bags-o -s ides
Once he op imal k( ypes o s ides) is de ined, he ou pu s o he pa i ioning can be g ouped
by clus e s o s udy he dis ibu ion o s ides om he same exe cise in he di e en ob ained
kg oups. As a esul , a new da ase is c ea ed, wi h as many objec s as exe cises had he
o iginal da ase s (IDFMAD and CVI), whe e each objec ep esen s a pai {use , exe cise}
and he ep esen a ion o he s ides ex ac ed om ha exe cise in e ms o clus e s i.e., he
p opo ion o s ides o an exe cise ha all in each s ides clus e . In o he e ms, we ob ain
a his og am o each pai {use , exe cise} ep esen ing he equency o appea ance o each
s ide clus e by exe cise. A se o his og ams H={h1,h2, ..., hn}is gene a ed, whe e each hk
is composed by a numbe o bins which sum up o one. Clus e ing his og ams ha e become
popula hanks o he bag-o -wo ds ca ego isa ion me hod (Nielsen e al. (2014)). This p o ides
a i s isual o e iew o he dis ibu ion o each exe cise, bu he e a e s ill oo many objec s
o analyse. The e o e, a second k-medoids pa i ioning is applied o his new da ase , using wo
di e en simila i y me ics (Deza and Deza (2013)):
•Euclidean dis ance is he simila i y measu e used by de aul in pa i ional clus e ings.
•Kullback-Leible (KL) Di e gence: also called ela i e en opy, measu es how one p ob-
abili y dis ibu ion di e ges om a second expec ed p obabili y dis ibu ion.
86

6.2 Spa io- empo al Analysis
The symme ised KL Di e gence is used o ob ain a dis ance ma ix o he da ase . P e ious
esea ch has shown ha i has be e pe o mance han he Euclidean dis ance. This second
pa i ion will be execu ed wi h bo h dis ances, al hough only KL Di e gence will be conside ed
o he inal analysis. Again his clus e ing p ocess will be pe o med wi h di e en alues o
k o de e mine he mos sui able pa i ion. The esul o his clus e ing will be a se o bag-o -
s ides ha ep esen exe cises cha ac e ised by a gi en ocabula y o s ides. In Chap e §7.2
a g aphical ep esen a ion o hese bag-o -s ides is gi en as well as he analysis o he con en
o he clus e ing esul .
6.1.5 Clus e s abili y
In his second clus e ing, i was no iced ha each execu ion o he pam algo i hm e u ned a
di e en clus e ing ec o . The e o e, each <pam,k0>, whe e k0={2,3,4,5,6}, was execu ed
20 imes in o de o de e mine he op imal pa i ions. Then he no malised mu ual in o ma ion
(NMI, Wagne and Wagne (2007)) index was applied o measu e he mu ual dependence be-
ween each pai o clus e ing ec o s ob ained. As a esul , o each <pam,k0>, a 20 ×20
ma ix was gene a ed wi h he NMI esul s. The pa i ion is conside ed s able when all he
esul s om he 20 ×20 NMI ma ix e u ned 1 (i.e. all he clus e ing ec o s a e equal). In
case any combina ion e u ns 1 o all he epe i ions, hen he mos s able should be chosen,
conside ing he numbe o 1 appea ing in he 20 ×20 ma ix and he alue o he o he cases
( he highe , he be e ).
This p ocess was applied using bo h he Euclidean and he KL di e gence as a dis ance
measu e o he k-medoids. The pai s <pam,k0>whe e NMI was equal o 1 we e he pa -
i ions selec ed o he inal analysis. Table 6.1 shows he combina ions ha will be ob ained
and p esen ed in Chap e §7. This able only includes he clus e s o exe cises wi h he KL
di e gence, as i is mo e used in he li e a u e when clus e ing his og ams. In addi ion, he
combina ions ob ained wi h he Euclidean dis ance we e pa o he KL esul s as well.
6.2 Spa io- empo al Analysis
In his sec ion, we will desc ibe in de ail he di e en pa ame e s ha we e p oposed in P akash
e al. (2015) and ela e i o he co esponding i-Walke ou comes. Al hough pa ame e s a e
he e p esen ed sepa a ely, we will la e combine hem o ob ain a be e unde s anding o ou
s udied popula ion and he di e ences hey p esen in ela ion o he i-Walke da a.
87
6. METHODOLOGY
Pilo S ide Clus e ing Exe cise Clus e ing Name
IDFMAD k = 4 k’ = 5 id mad4k5k’
CVI k = 3 k’ = 3 c i3k3k’
k = 5 k’ = {3,5}c i5k3k’
c i5k5k’
k = 6 k’ = {3,4}c i6k3k’
c i6k4k’
Table 6.1: Resul ing combina ions o s ides pe exe cises o be analysed in di e en scena ios
Spa io- empo al pa ame e s p o ide he simples o m o objec i e gai e alua ion in e ms
o ime and dis ance. This sec ion desc ibes he me ics ha a e commonly used in he li e -
a u e o assess indi iduals’ walking pe o mances. Rela ion o hese me ics o he ex ac ed
a iables om he i-Walke aw da a is also p o ided, al hough he p e ious sec ion al eady
showed how hese we e ob ained. To e iew he de ini ions o s ep, s ide and gai cycle, see
Chap e §2.
6.2.1 Desc ip i e Gai Pa ame e s
Mos o he human gai esea ch ocus on he same se o me ics o assess he gai quali y
o he use . In no mal walking pa e ns, a gai cycle is he con inuous epe i ion o s ides o ,
as de ined in Pappas e al. (2001), a succession o 4 phases o a gi en oo : s ance, heel-o ,
swing and heel-s ike. Conside ing a s ep as he mo emen o one oo in on o he o he ,
and a s ide (o gai cycle) as a s ep om one oo ollowed by a s ep om he o he oo , we
can de ine he ollowing quan i a i e me ics:
•S ep leng h: dis ance (in me e s o cen ime es) be ween he heel-o and heel-s ike o
one oo .
•S ep ime: du a ion (in seconds) be ween he heel-o and heel-s ike o one oo .
•Numbe o s eps: o al numbe o s eps du ing an exe cise.
•Numbe o s ops: o al numbe o s ops du ing an exe cise
•S ide leng h: dis ance (in me e s o cen ime es) o a gai cycle.
88
6.2 Spa io- empo al Analysis
•S ide ime: du a ion (in seconds) o a gai cycle (o he ime be ween s eps o he same
oo )
•Cadence: Numbe o s eps pe minu e
The me ics included in his hesis a e he s ide leng h and ime, he numbe o s eps and
cadence. In his wo k, he numbe o s ides was collec ed ins ead, and he hy hm was modi ied
o ep esen he numbe o s ides pe minu e. Al hough he p esen ed me hodology allows also
ex ac ing he in o ma ion a s ep le el, hese we e no conside ed in he analysis. S ide- o-
s ide luc ua ions ha e been adi ionally used in li e a u e o s udy gai a iabili y, which is
a complemen a y way o e alua e human locomo ion and i s change wi h age o disease. I
has also been closely ela ed o gai diso de s, leading he esul s o ca ego ise indi iduals
pa icipa ing in hese s udies in e ms o ail y o isk o alling (Hausdo (2005a)).
As men ioned be o e, s eps (and hus, s ides) a e ob ained by local maxima o he Fxdi
pushing o ce. We use he walking dis ance as he e e ence. The e o e, i is easy o ex ac
he dis ance be ween peaks o local maxima. In addi ion, we know ha da a is collec ed a a
pe iodical and ch onological ime, so he ime be ween peaks can be ob ained by coun ing he
numbe o obse a ions and con e ing i o seconds. We p o ide he numbe o s ides a he
end o he dic iona y o s ides gene a ion, and he cadence is a simple con e sion om he
numbe o s ides along he exe cise (usually ew seconds) o minu es.
6.2.2 Gai Veloci y
The e a e se e al app oaches ha use gai eloci y (e.g.,Mon e o-Odasso e al. (2012)) as an
assessmen me ic o cogni i e decline and isk o alling, al hough one o he mos commonly
used is he 10 Me e Walking Tes . T adi ionally, his es is measu ed by di ec obse a ion
o he pe o mance and a simple ime measu emen h ough a ch onome e . Mo e ecen s ud-
ies include he use o wea able senso s (usually accele ome e s placed a w is o ankles) o
walking pla o ms o oo acking (see §2). Less equen a e he s udies whe e assis i e de-
ices a e in ol ed. Howe e , au ho s like Wang e al. (2014) and Balles e os e al. (2015) ha e
p oposed di e en me hodologies using di e en e sions o obo ised olla o s. This so o
s udies ough o be e y ca e ul wi h he a ge popula ion selec ed since he olla o is no
adequa e o speci ic pa hologies, e.g., people wi h hemipa esis who ha e no la e al balance
gene ally due o he consequences o a s oke. As men ioned in Chap e §4.3.1, he i-Walke
can p o ide a helping o ce ha would gi e he equi ed balance o walk sa ely. Howe e , his
89
6. METHODOLOGY
was no conside ed in his hesis due o he di icul y o pass he p o ocol in each pilo and o
ge access o people wi h his pa hology.
Acco ding o Mon e o-Odasso e al. (2005), olde adul s can be di ided in h ee g oups
based on gai eloci y (GV): high GV (>1 m/s), medium GV (0.7-1 m/s) and low GV (<0.7
m/s). I is conside ed ha people no belonging o he high GV g oup, and especially hose
belonging o he low GV g oup, ha e mo e p obabili ies o su e om ad e se e en s due o
physical o cogni i e decline. We di ided pa icipan s acco ding o his me hodology, and i
will be used o obse e he ends in di e en pe iods o ime o he I-DONT-FALL popula ion.
We will pe o m an analysis on Gai Veloci y o bo h he IDF as an independen da ase ,
he baseline da ase s (IDFMAD) and he alida ion da ase (CVI) o s udy he dis ibu ion o
ou popula ion. Pa icipan s will be classi ied acco ding o he app oach p oposed by Mon e o-
Odasso e al. (2005), and we will de e mine whe he ou popula ion ollows he clinical hypo h-
esis men ioned du ing his wo k: (i) poo pe o mances o gai eloci y a e ela ed o age, and
cogni i e s a us; (ii) physical and/o cogni i e aining helps o imp o e gai eloci y and, hus,
educing gai a iabili y; (iii) aining will also help o educe he isk o alling and ea o
alling. Fo he IDF da a, we will compa e he pe o mance be o e and a e he h ee-mon hs
aining (T0 and T1 espec i ely), and we will s udy he e olu ion o each o he aining
g oups (mo o , cogni i e, mixed and placebo).
6.3 Use D i ing Skills
The e a e se e al me hods o e alua ing he human- obo in e ac ion while d i ing. U diales
(2012) p o ides me hods o (i) analyse use d i ing skills and ob ain use p o iles acco ding
o hei pe o mance, and (ii) in e p e use disag eemen in a collabo a i e con ol d i ing
s a egy. In ou case he i-Walke wo ks wi h a pu ely eac i e con ol o now, as no con ol
s a egies ha e been added o assis in he e ec i eness o he d i ing skills o he use . Hence,
i will ac as a sensing pla o m ha collec s da a om di e en senso s du ing he exe cise a
a a e o 10Hz.
In his hesis, we will ollow he same philosophy han in U diales (2012) o assess human
d i ing skills, which has been ully desc ibed in §3.2.1, bu adap ing i o he cha ac e is ics
o he i-Walke a chi ec u e. The main di e ence is ha he i-Walke equi es human body
mo ion o mo e, i.e., use s mus use angle join s such as legs o c ea e mo ion and a ms o
exe a o ce on he handle s o he olla o . By analysing how does an elde pe son use he
90
6.5 Modelling Fall Risk Assessmen
6.5 Modelling Fall Risk Assessmen
One o he objec i es o he I-DONT-FALL p ojec was o educe he isk o alling in elde ly
popula ion h ough di e en ypes o aining. As men ioned be o e, da a ob ained om clinical
scales a baseline was compa ed o he pos - aining esul s o e alua e he e ec i eness o he
sys em.
We used he i-Walke da a gene a ed du ing he 10MWT exe cises be o e and a e he
aining and applied a machine lea ning echnique o he p edic ion o he isk o alling o
an indi idual. We selec ed se e al a iables o he 10MWT exe cises om he s udy sample
using he i-Walke . These a iables combined aw da a wi h o he calcula ed a iables. They
co espond o he a e age alues om he o ce senso s (X,Yand Zdi ec ions) o bo h hands
and he a e age speed. The da ase has a o al o 20 a iables. The a e age was compu ed
o all he 10 me e s o he exe cise and also o he alues disca ding he i s and las wo
me e s ( he cen al pa o he exe cise). The eason o d opping alues om he beginning
and he end o he exe cises was o educe noise om accele a ion and decele a ion phases and
ocusing on he s able egime o he exe cise.
The exe cises we e di ided in o wo da ase s. The i s co esponds o he exe cises pe -
o med be o e ea men (T0) and he second o he exe cises pe o med a e ea men (T1).
The da a be o e ea men was used as a aining se o ob ain a p edic i e model o he isk o
alling. The second da ase is used as es o de ec i he popula ion has changed hei s a us as
an e ec o he ea men .
To ob ain he model o all isk, a logis ic eg ession was pe o med using L1 egula i-
sa ion, i s wi h all he a iables, and hen selec ing only he ele an ones, by disca ding all
a iables ha we e assigned ze o weigh by he egula ised logis ic eg ession. F om all he
a iables only 10 we e used by he model ha included a iables o all he exe cise (a e age
speed, mean o ce on he Xdi ec ion, igh hand o ce on he Xdi ec ion and le hand o ces
on he h ee di ec ions) and only o he cen al pa (le hand o ces on he Xand Zdi ec ions
and igh hand o ces on he Yand Zdi ec ions). A Suppo Vec o Machine (SVM) model
wi h linea ke nel was also compu ed, ob aining iden ical esul s. I is clea ha he in o ma-
ion collec ed om he o ce exe ed by he use o he i-Walke is essen ial o cha ac e ise i s
walking beha iou .
This model was no applied o he es o da ase s since hey did no ollow he same p o o-
col. The es o pilo s execu ed he es in one day, and hence no empo al compa ison o hei
97

6. METHODOLOGY
e olu ion is possible.
6.6 Summa y
The i-Walke o e s se e al sou ces o in o ma ion ha can be ela ed o he human walking
beha iou . The main idea is o ansla e da a senso eadings in o eadable and unde s andable
epo s, aiming o complemen he obse a ions o clinicians while he use is pe o ming an
exe cise. Fusing he expe knowledge on human body mo ion wi h he da a ob ained om
olde adul s using i-Walke would allow he design o messages o ale o in o m he use
abou a si ua ion (see a i s app oach o he implemen a ion o his se ice in Appendix B).
Mo eo e , he i-Walke would be able o ake decisions in dange ous si ua ions and assis he
use in his/he mobili y.
One o his PhD wo k aims is o ind hose known gai pa ame e s ep esen ed in da a
ob ained om he i-Walke o he selec ed a ge popula ion. Fo his, we ha e i s ocused on
a desc ip i e analysis o he s udied pa ame e s. We ha e also p oposed a model o p edic he
isk o alling o he I-DONT-FALL popula ion. Finally, p e ious esul s on applying clus e ing
echniques o iden i y indi iduals wi h alls a e p omising (see Chap e §4.3.3), al hough hey
conside exe cises as a whole. Thus, i canno p ocess in o ma ion on use ’s in e ac ion a
each momen . We expec ha by clus e ing use s’ s ides, we will be able o iden i y gai
dis u bances associa ed wi h non-heal hy olde adul s.
As a esul o he me hodology p esen ed he e, a bag-o -s ides and a bag-o -exe cises a e
c ea ed o each da ase (IDFMAD and CVI), con aining he spa io empo al me ics abo e-
men ioned and he exe ed o ces by he human o he i-Walke a s ide le el while pe o ming
he exe cise. We a e con iden as we ha e obus se s o da a coming om I-DONT-FALL and
o he campaigns.
98
Chap e 7
Resul s
In his chap e , we p esen he esul s ob ained om he p oposed me hodology and he al eady
in oduced da ase s. To desc ibe hese esul s we will discuss he ele ance o each o he gai
cha ac e is ics ex ac ed om he in e ac ion be ween he pa icipan s and he i-Walke .
Fo his analysis, we ha e selec ed some e alua ion me ics ha can be ound in he li-
e a u e o human gai esea ch (see Chap e §6) ha will be e alua ed by means o he da a
ob ained om he i-Walke senso s. Fi s , he me hodology will be applied o he baseline
popula ion (bo h alle s and heal hy olde adul s) ega ding gai eloci y while pe o ming he
10 Me e Walk Tes . The ele ance o his measu e in he ea ly de ec ion o he decline in
olde adul s has al eady been explained in his documen . The objec i e is o obse e how
a e he pa icipan s dis ibu ed ega ding gai eloci y and whe he signi ican di e ences a e
obse able be ween g oups. Second, he esul s o he spa io empo al analysis a e shown,
ocusing on he cha ac e is ics o an elde indi idual’s gai (leng h, ime, eloci y). Following,
a s udy use s’ d i ing skills conce ning di ec ionali y and la e ali y is gi en. Then, he esul s
om he clus e ing app oach o gai analysis a e p esen ed o bo h he baseline and alida ion
da ase s (i.e., obse a ions o 10MWT and 3mWT). Finally, he esul s o a eg ession model
o all isk p edic ion applied o he IDF da ase a e in oduced in his chap e .
7.1 Spa ioTempo al Analysis
In his sec ion, we will show p elimina y esul s on he ime and dis ance pa ame e s o human
gai h ough i s in e ac ion wi h he i-Walke .
99
7. RESULTS
7.1.1 Desc ip i e Gai Pa ame e s
We ha e seen ha s eps can be ep esen ed by equency o by dis ance, bu bo h equi e he
use o il e s o disca d noise om human oscilla ions. Once s eps a e de ec ed and loca ed
in he da a signal, i is possible o ep esen i in di e en unde s andable ways as shown in
Figu e 7.1. He e he signal has been il e ed by dis ance, i.e., by a minimum o walked dis-
ance be ween s ides based on he a e age walking speed. Acco ding o ou in e p e a ion,
exe cises s a when he i s igh s ide is de ec ed. Wi h a simila p ocess, we can ex ac he
s ides pe o med wi h he le leg. Fu he mo e, his igu e adds a new laye o in o ma ion,
in oducing he concep s o p ominence and wid h de ined in §6.1.2. In his case, he wid h
ep esen s he ime in seconds be ween peaks, i.e., he a e age ime he use has spen push-
ing be ween s eps. This concep can also be in e p e ed in e ms o dis ance a elled be ween
peaks. Howe e , o he es o he analysis, he wid h was ep esen ed in e ms o dis ance, as
al eady explained in §6.1.
As abo e said, he posi i e pa s o he signal co esponding o hose momen s whe e he
use was exe ing mo e o ce wi h i s igh hand. I is expec ed ha in an exe cise like he
10MWT, whe e he ajec o y is assumed o be a s aigh line, he signal will be balanced. A
pe son d i ing o one side while d i ing will p esen less ze o- ime momen s since he inc ease
in he opposi e side will no be high enough o compensa e he mo emen . The compa ison o
he o ce p ominence o bo h a ms aims o illus a e how does he indi idual in e ac wi h he
i-Walke and associa e i wi h i s physical o cogni i e dys unc ion.
Figu e 7.2 ep esen s he e olu ion o he s ep leng h o each oo while pe o ming he
10MWT. Usually, 10MWT assessmen s dis ega d he beginning and he end o he exe cise (2
me e s on each side) since hey co espond o he accele a ion and decele a ion phases and he
gai pace is no egula . In Figu e 7.2.a, we can obse e his beha iou o bo h ee . Mo eo e ,
i s eng hens he hypo hesis ha he indi idual has s a ed he exe cise wi h he igh oo and
inishes i wi h he le one. Figu e 7.2.b ep esen s he inc emen s o each oo ’s s ides in
e ms o dis ance (me e s). As we obse e, in his case, he indi idual does bigge s ides wi h
he igh oo . The same analysis and ep esen a ion can be gi en in e ms o s ide ime.
A e applying he bag-o -s ides app oach desc ibed in he me hodology, a da abase is
gene a ed whe e, o each s ide, he ollowing in o ma ion is being s o ed:
•use ID, gende , age ange, all isk (Tine i), alle (yes/no)
100
7.1 Spa ioTempo al Analysis
0 1 2 3 4 5 6 7 8 9 10 11 12
Time (s)
-15
-10
-5
0
5
10
15
FxDi (Kg)
signal
peak
p ominence
wid h (hal -p ominence)
Figu e 7.1: Righ leg s ides, pushing o ce inc emen s and a e age pushing ime wi hin peaks
• imes amp and numbe o exe cise (being T0-T1 o he IDF da ase , o ound numbe in
he case o MAD and CV I)
•s ide ID (inc emen al wi hin he same exe cise)
• o al numbe o s ides in exe cise
•dis ance a elled, du a ion o he exe cise, a e age speed and cadence (in me es, sec-
onds, me es/seconds and s ides/minu e espec i ely)
•s ide leng h and ime (in me es and seconds espec i ely)
•a e age s ide leng h and ime in exe cise(in me es and seconds espec i ely)
•a e age pushing o ce along he s ide in e ms o Fxdi
This will allow depic ing he dis ibu ion o ou s udied popula ion om di e en pe spec-
i es ha will be p esen ed in he ollowing sec ions. Table 7.1 shows he main spa io empo al
cha ac e is ics o he IDFMAD pilo popula ion.
7.1.2 Gai Veloci y
The Gai Veloci y has been calcula ed o bo h he IDF and MAD pilo s o he 10 Me e
Walk Tes . I was also applied o he CVI pilo , whe e each s aigh line was conside ed as an
101
7. RESULTS
Le Righ
Fee
0
1
2
3
4
5
6
7
8
9
10
Dis ance (m)
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25
S eps
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
S ide Leng h (m)
Righ
Le
Figu e 7.2: E olu ion o an indi idual’s s ide leng h du ing he 10MWT: (a) Es ima ed ee
posi ion; (b) S ide leng h o each oo .
independen exe cise. In he case o MAD and CVI, gai eloci y was ob ained by calcula -
ing he a e age walking dis ance and walking ime o all he exe cise pe o med by he same
pa icipan ; his echnique is known in he li e a u e as he es - e es . Fo he inal analysis,
some pa icipan s we e excluded due o senso ailu e o d op-ou . A a i s s age, pa icipan s
om MAD wi h alls we e excluded, bu we inally decided o ea IDF and MAD as a single
baseline da ase .
The dis ibu ion o each pilo is depic ed in sepa a ed ables and commen ed below. The
Spa io Tempo al Cha ac e is ics
Biological
Cha ac e is ics N Cadence Speed sLeng h sLeng h CV sTime sTime CV FxDi
Age M 85 49,88 0,80 0,93 42,45 1,04 54,31 1,72
Gende F 49 49,14 0,80 0,94 42,63 1,05 55,37 1,36
Gende M 36 50,89 0,79 0,91 42,21 1,02 52,87 2,19
Age O 200 45,83 0,61 0,76 42,06 1,28 55,98 1,66
Gende F 140 45,56 0,57 0,71 42,12 1,35 55,91 1,60
Gende M 60 46,48 0,71 0,87 41,91 1,12 56,16 1,81
To al gene al 285 47,04 0,67 0,81 42,18 1,21 55,49 1,68
Table 7.1: Spa io- empo al cha ac e is ics o he IDFMAD popula ion.
102

7.1 Spa ioTempo al Analysis
Cha ac e is-
ics
Low GV
(N= 52)
N(%) o
Mean ±SD
Median GV
(N= 27)
N(%) o
Mean ±SD
High GV
(N= 6)
N(%) o
Mean ±SD
Age
<80
≥80
To al
14 (26.9%)
38 (73.1%)
85.3 (±6.8)
15 (55.5%)
12 (44.5%)
79.2 (±9.4)
6 (100%)
0 (0%)
75.3 (±5.2)
Gende Female
Male
36 (69.2%)
16
18 (66.7%)
9
4 (66.7%)
2
T ea men
Mix
Mo o
Cogni i e
Placebo
15
8
12
17
4
6
7
10
4
1
1
0
Risk
Low
Medium
High
1
11
40
3
17
7
4
2
0
Table 7.2: Dis ibu ion o he IDF pa icipan s be o e 3-mon hs aining (IDF-T0) pe o ming
he 10MWT. Da a is ep esen ed by Gai Veloci y g oups. These esul s a e used o e alua e
he e ec i eness o he IDF solu ion a e he aining phase.
dis ibu ion has been obse ed in e ms o age, gende , ea men and o iginal isk o alling.
Table 7.2 and Table 7.3 p o ide he same s uc u e o he IDF popula ion. Tables 7.4 and Table
7.5 show he same in o ma ion o he MAD and CVI pilo s espec i ely, excep o he ea -
men in o ma ion which only ega ds he IDF p o ocol. In hese wo la e pilo s, pa icipan s
epea ed se e al imes he same exe cise. Thus he walking speed has been ob ained om he
a e age walking speed o each exe cise pe o med by he same use , as be o e-men ioned.
E en hough he MAD popula ion is supposed o be heal hie han he one om IDF, hey
a e also olde ( he p opo ion o people aged 80+ is signi ican ly highe han in he IDF pilo ,
which co esponds o he ypical age o people li ing in ca e cen es). Hence, only h ee in-
di iduals (5%) eached he High GV g oup and, su p isingly, hey p esen a medium o high
isk o alling. Un o una ely, he Low isk o alling is poo ly ep esen ed, and hus no de ini e
conclusions can be ex ac ed om his esul . The es o g oups a e cohe en wi h he li e a-
103
7. RESULTS
Cha ac e is-
ics
Low GV
(N= 50)
N(%) o
Mean ±SD
Median GV
(N= 22)
N(%) o
Mean ±SD
High GV
(N=12)
N(%) o
Mean ±SD
Age
<80
≥80
To al
15 (30%)
35 (70%)
84.8 (±6.6)
10 (45.5%)
12 (55.5%)
82.7 (±9.3)
11 (91.7%)
1 (8.3%)
73.3 (±6.9)
Gende Female
Male
37 (74%)
13
13 (59.1%)
9
7 (58.3%)
5
T ea men
Mix
Mo o
Cogni i e
Placebo
13
11
10
16
4
3
8
7
6
1
2
3
Risk
Low
Medium
High
0
24
26
4
17
1
7
4
1
Table 7.3: Dis ibu ion o he IDF pa icipan s a e 3-mon hs aining (IDF-T1) pe o ming
he 10MWT. Da a is ep esen ed by Gai Veloci y g oups and is used o assess he usabili y o
he all isk p e en ion sys em.
u e: mos o he people wi h highe isk o alling a e classi ied in he low GV g oup. A e
con ac ing Los Nogales clinical eam, i was possible o e ie e upda ed assessmen s one yea
a e he pilo ook place o hose pa icipan s ha we e s ill li ing in he ca e cen e. I was
obse ed a gene al imp o emen in e ms o Tine i scale esul s, especially o hose ha ing
a be e cogni i e s a us (measu ed wi h he MMSE scale). Thus, i is p esumed ha people
wi h a pa icula cogni i e condi ion and abili y (o willingness) o keep an ac i e li ing we e
able o imp o e hei mobili y and hus educe hei isk and/o ea o alling. This da a will
no be sha ed o e hical issues, as i was no included in he consen ed in o m signed by he
pa icipan s a he ime o he pilo es ing.
In he case o he CV I pilo , he age ange is expanded, including now he Young ca ego y
(≤65 yea s old). In his es , 13 pa icipan s had some ca diac pa hology o physical inju ies
esul ing om a all and we e li ing in he ca e cen e. The es o people a e heal hy people,
mos o hem a e s ill employed o e i ed in he las yea , and we e li ing independen ly. The
dis ibu ion o he GV g oups by age ollows he same end han in he es o pilo s. I also
104
7.1 Spa ioTempo al Analysis
Cha ac e is-
ics
Low GV
(N= 34)
N(%) o
Mean ±SD
Median GV
(N= 21)
N(%) o
Mean ±SD
High GV
(N= 3)
N(%) o
Mean ±SD
Age
<80
≥80
Mean
3 (8.8%)
31 (91.2%)
88.3 (±5.8)
4 (19%)
17 (77.78%)
86.9 (±6.2)
0 (0%)
3 (100%)
87.6 (±5.8)
Gende Female
Male
27 (79.4%)
7
9 (42.8%)
12
1 (33.3%)
2
Risk
Low
Medium
High
1
8
21
1
10
10
0
2
1
Table 7.4: Dis ibu ion o he MAD pa icipan s a e a 10MWT es - e es ep esen ed by Gai
Veloci y g oups.
shows ha 8 o he 13 challenged pa icipan s all in he Low GV g oup; hese people whe e
e en he eldes om his challenged g oup, which was he expec ed. The o he i e use s a e
classi ied in he Median GV g oup co espond o middle-aged people wi h p ema u e ca diac
pa hologies. I has also been no iced ha young pa icipan s change hei walking beha iou
when using an assis i e de ice, educing hei walking speed, which migh a ec he esul s o
he clus e ing p esen ed in §7.2. I was also expec ed ha pa icipan s (especially challenged o
elde ) would show some end in he walking speed among ounds (i.e., elde people pe o m-
ing each s aigh line slowe han he p e ious one), bu his is me ely obse able. People end
o walk a a simila speed be ween each ound, especially he younges and eldes pa icipan s;
people in middle age a e he ones ha show mo e a ia ions, bu ne e mo e han 0.2 m/s, and
usually always wi hin he same GV class.
Resul s on Table 7.2 and Table 7.3 show ha he g oup o pa icipan s a e dis ibu ed as
expec ed in e ms o age and gai eloci y: he younge hey a e, he as e hey walk. Mo e-
o e , people unde 80 yea s old end o imp o e hei pe o mance in e ms o eloci y a e
he 3-mon hs aining. On he o he hand, men walk as e han women on a e age, aking
in o accoun he p opo ion o gende ep esen a ion, al hough his could easily be ela ed o
105
7. RESULTS
Cha ac e is-
ics
Low GV
(N= 8)
N(%) o
Mean ±SD
Median GV
(N= 18)
N(%) o
Mean ±SD
High GV
(N= 16)
N(%) o
Mean ±SD
Age
<65
65-80
≥80
Mean
0 (0%)
0 (0%)
8 (100%)
87 (±4.2)
8 (44.4%)
3 (16.7%)
7 (38.9%)
63.2 (±24.4)
9 (56.3)
1 (6.2)
6 (37.5%)
52.5 (±19.1)
Gende Female
Male
7 (87.5%)
1
14 (77.8%)
4
13 (81.2%)
3
Risk
Low
Medium
High
0
1
7
11
6
1
14
2
0
Table 7.5: Dis ibu ion o he CV I pa icipan s pe o ming he 3mWT. Da a is ob ained by
ex ac ing he s aigh lines o he exe cise, which go om 25 o 35 me es long, and applying
he es - e es me hod o ob ain he a e age speed. Da a is ep esen ed by Gai Veloci y (GV)
g oups.
an h opome ic pa ame e s, such as heigh . Finally, as i was expec ed in he ou comes o he
I-DONT-FALL p ojec (see I-DONT-FALL), people ha ing been assigned o he Mixed T ain-
ing g oup ha e ob ained be e pe o mances han he es o he g oups. Howe e , he Mo o
T aining g oup did no p esen he expec ed imp o emen s in e ms o eloci y, al hough he
sample is small o ex ac a eliable conclusion wi h his single me ic. In gene al e ms, esul s
a e cohe en wi h hose exis ing in he e iewed li e a u e: (i) gai eloci y is ela ed o some
biological cha ac e is ics o he indi idual; (ii) i is essen ial o p omo e ac i e, heal hy ageing
bo h in physical and cogni i e asks; (iii) gai eloci y can be used as an ea ly indica o o
decline in olde adul s.
Resul s we e also ep esen ed g aphically o obse e he dis ibu ion among he di e en
pilo s. Figu e 7.3 shows he di e ence in gai eloci y be ween T0 and T1. The ed a ea
ep esen s he MAD pilo , which is concen a ed a ound he ze o alue (i.e., no di e ences
be ween T0 and T1). The h ee s udied pilo s om IDF (HGG in g een, FSL in blue and
SERMAS in yellow) show di e en dis ibu ions in gai eloci y. A eas on he igh side o he
106
7.2 Clus e ing Resul s o he Gai Analysis
Bag-o -s ides in Exe cise Clus e 2
1 2 3 4
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 2
1 2 3 4
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 3
1 2 3 4
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in exe cise Clus e 4
1 2 3 4
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 5
1 2 3 4
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Figu e 7.5: Scena io 1: id mad4k5k’. Clus e ep esen a ion o bags-o -s ides wi h ou ypes
o s ides, g ouped in i e so s o exe cises.
we aimed o ocus on a common pa hology (in his case, ca diologic dys unc ion) o he elde
g oup o pa icipan s. Howe e , i was no possible o each he minimum numbe o olun ee s.
Thus we had o inc ease he numbe o young pa icipan s, ob aining an unbalanced pilo in
e ms o biological da a. Resul s a e p esen ed in he ollowing sec ion.
7.2.2 CV I pilo
The CVI da ase p o ided di e en combina ions o s able clus e ing esul s, which a e p e-
sen ed in Table 6.1. Each combina ion will be s udied in his sec ion and jus i ica ions o he
selec ed op ion will be gi en a he end. In p inciple, he mo e clus e s we ob ain, he mo e
accu a e will hei in o ma ion be. Howe e , we mus obse e he ype o g oups ha ha e been
o med in o de o de e mine hei in a simila i y in e ms o gai shape and an h opome ic
cha ac e is ics o he people belonging o a same bag-o -s ides.
113

7. RESULTS
Figu e 7.6: Righ de ia ion du ing a 10MWT
7.2.2.1 Fi s Scena io
The i s scena io s udies he esul s o ca ego ising s ides in h ee clus e s. The bags-o -
s i des o each exe cise a e hen dis ibu ed h ee ypes o exe cises. Figu e 7.8 depic s he
dis ibu ion o he bag-o -s ides o each kind o exe cise and Table 7.7 con ains some de ails
abou he an h opome ic and spa io- empo al cha ac e is ics o he pa icipan s belonging o
each clus e , along wi h he gene al dis ibu ion. Use s’ in o ma ion is classi ied by age ange
(Middle, Old, Young) and hen by gende (Female, Male). The isk o alling has no been
included since i is highly co ela ed o he age g oup in his da ase .
Figu e 7.8 depic s ha each g oup o exe cises has a gai shape which p e ails o e he
es , and in each case, i is a di e en s ide ype. In he i s ype o exe cise, use s pe o med
mainly s ides o ype 1, wi h o e a 70% o ep esen a ion. In he second g oup o exe cises,
again o e 70% o s ides belong o he second clus e o s ides. Finally, mo e han 80% o
he s ides pe o med in he hi d kind o exe cise belong o he hi d clus e o s ides. As i
can be obse ed in Figu e 7.7, he shape o hese s ides sha e some common cha ac e is ics a
i s glance. The wo i s clus e s ha e simila ime leng h, while s ides in he hi d g oup a e
114
7.2 Clus e ing Resul s o he Gai Analysis
Clus e 1 2 3
Age N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h
Age M 6 1,17 61,58 1,18 6 1,09 59,81 1,14 26 1,05 55,20 1,18
F 6 1,17 61,58 1,18 6 1,09 59,81 1,14 16 0,99 53,24 1,14
M 10 1,14 58,34 1,25
Age O 7 0,95 54,16 1,11 24 0,75 53,03 0,82 16 0,56 52,82 0,65
F 2 0,70 56,22 0,77 19 0,73 51,12 0,80 16 0,56 52,82 0,65
M 5 1,05 53,33 1,24 5 0,86 60,31 0,86
Age Y 37 1,05 56,62 1,13 7 1,20 59,48 1,28 24 1,05 55,87 1,18
F 37 1,05 56,62 1,13 7 1,20 59,48 1,28 16 1,07 56,91 1,18
M 8 1,00 53,80 1,18
To al 50 1,05 56,87 1,14 37 0,89 55,35 0,96 66 0,93 54,87 1,05
Table 7.7: An h opome ic and spa io empo al gai cha ac e is ics o Scena io 1
sligh ly longe . The o he di e ence elies on he amoun o pushing o ce applied. The second
g oup di e s om he o he ha ing he bigges di e ence in he esul ing pushing o ce: use s
in his clus e uses a la ge pushing o ce on he pa o he body ha is making he s ep. Mos
o he exe cises om his clus e a e pe o med by pa icipan s aged 85+ yea s old wi h se e al
como bidi ies; i is p esumable ha hese indi iduals equi e mo e suppo on he i-Walke o
pe o m he exe cise and hus p esen his gai pa e n ha clea ly shows he oo ansi ion
in he swing phase. The o he wo g oups emain in posi i e alues e en when he le s ep is
aking place, which means ha he compensa ion be ween bo h sides o he body while walking
is mo e unbalanced. This si ua ion is mo e ema kable in he hi d clus e which is he mos
he e ogeneous g oup in age and gende ep esen a ion. Finally, Exe cise clus e 1 has mainly
classi ied emale pa icipan s, especially young women.
7.2.2.2 Second Scena io
The second scena io conside s i e ypes o s ides. The co esponden bags-o -s ides a e hen
g ouped in ei he h ee o i e so s o exe cises, as ep esen ed in Figu e 7.10 and Figu e 7.11
espec i ely. Gai shapes o his clus e ing a e depic ed in Figu e 7.9. In he i s case, i can
be obse ed ha each g oup o exe cises has one o wo dominan ypes o s ides, and in each
case, hese di e . In he i s g oup, he second s ide ype p e ails wi h o e 65% o he s ides
pe o med in he exe cises clus e ed in his g oup. This g oup o s ides is also he one wi h
mo e obse a ions, including one- hi d o he o al amoun o hem. People in his g oup a e
115
7. RESULTS
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i3k-75-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i3k-3862-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i3k-1230-medoid and neighbou s in clus e
Figu e 7.7: Gai shapes in CVI da ase wi h h ee clus e s o s ides
Bag-o -s ides in Exe cise Clus e 1
123
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 2
123
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 3
123
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Figu e 7.8: Scena io 1: c i3k3k’. Clus e ep esen a ion o bags-o -s ides wi h h ee ypes o
s ides, g ouped in h ee so s o exe cises.
116
7.2 Clus e ing Resul s o he Gai Analysis
hose p esen ing a minimal o ce a ia ion. The gai shape indica es ha a signi ican igh -
hand o ce was exe ed du ing he phases ega ding he igh s ep, and hen he o ces applied
om he wo pa s o he body ge balanced o pe o m he le s ep. This clus e is again he
mos he e ogeneous in age and gende and coincides wi h he one ob ained in he p e ious
scena io.
In he second g oup o exe cises, almos 80% o he s ides a e ep esen ed by wo ypes,
s ides 1 and 3, which a e also he sho e ones. As depic ed in Table 7.8, his g oup is mainly
o med by exe cises pe o med by women, especially young ones, and one single man om
he olde age g oup. The s ide leng h is s ongly ela ed o he indi idual’s heigh , which is
gene ally lowe in women. In his case i appea s o be he main disc imina o , along wi h
he amoun o o ce employed, o ca ego ise mos women oge he . Finally, he hi d clus e
o exe cises is mos ly ep esen ed by he wo las ypes o s ides om Figu e 7.9. The shape
o he ou h s ide e eals ha he o ce compensa ion be ween bo h pa s o he body du ing
he le s ep a e mo e balanced, since hey a e close o ze o. The shape o he i h s ide is
he one wi h mo e a ia ion be ween igh and le hand o ce, bu bo h s a and end he gai
cycle (i.e., he igh s ep phases), using a conside able amoun o o ce wi h he igh hand.
As p e iously obse ed, his is a pa e n in exe cises pe o med by he olde olun ee s, which
we e diagnosed wi h se e al como bidi ies (mos ly ela ed o ca diological p oblems). Mos
o he people execu ing his ype o o ce along he gai cycle, and especially hose alling in
his las Exe cise Clus e , a e hose p esen ing mo e como bidi ies, and hus ha e p esumably
mo e dis u bances while walking. Table 7.8 shows ha his clus e is composed by hal o he
exe cises pe o med by he olde adul s, bu has also some ep esen a ion om he young and
middle-age g oups. Du ing he execu ion o his pilo , i was obse ed ha some people om
he con ol g oup (i.e., no needing a mobili y suppo de ice) modi y hey walk when using
he i-Walke 1, which could explain he walking beha iou o hose young people appea ing in
his Exe cise clus e .
I we ake a look a he second case o his scena io (Figu e 7.11), whe e bags-o -s ides a e
ca ego ised in o i e so s o exe cises, i can be obse ed ha in gene al, a pa e n is eme ging
in wo clus e s. In his case, he i s Exe cise clus e co esponds o he p e iously explo ed
clus e s ha we e mainly o med by young women (in his case 28 exe cises pe o med by
1This ema k is gi en unde a pe sonal, empi ical e idence. I was obse ed ha some young and middle age
pa icipan s could cause some alse lead du ing he analysis: some changed hei walking speed du ing he exe cise,
o o he s we e e y ocused on he in e ac ion hey we e ha ing wi h he handle s.
117
7. RESULTS
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i5k-29-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i5k-1230-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i5k-318-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i5k-1297-medoid and neighbou s in clus e
5 10 15 20 25 30 35 40 45 50
S ide Time (#ins ances in imese ies)
-6
-4
-2
0
2
4
6
FxDi Fo ce (in N)
c i5k-3797-medoid and neighbou s in clus e
Figu e 7.9: Gai shapes in CVI da ase wi h i e clus e s o s ides
Bag-o -s ides in Exe cise Clus e 1
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 2
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 3
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Figu e 7.10: Scena io 2.1: c i5k3k’. Clus e ep esen a ion o bags-o -s ides wi h i e ypes
o s ides, g ouped in h ee so s o exe cises.
118

7.2 Clus e ing Resul s o he Gai Analysis
Clus e 1 2 3
Age N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h
Age M 26 1,05 59,52 1,18 5 1,18 67,46 1,18 7 1,09 63,84 1,14
F 16 0,99 57,32 1,14 5 1,18 67,46 1,18 7 1,09 63,84 1,14
M 10 1,14 63,04 1,25
Age O 13 0,50 54,13 0,58 11 0,89 57,80 1,02 23 0,76 55,71 0,82
F 13 0,50 54,13 0,58 7 0,74 58,25 0,80 17 0,74 53,15 0,82
M 4 1,17 57,01 1,41 6 0,81 62,96 0,82
Age Y 20 1,01 59,26 1,15 40 1,05 61,37 1,14 8 1,27 66,42 1,32
F 12 1,01 59,84 1,14 40 1,05 61,37 1,14 8 1,27 66,42 1,32
M 8 1,00 58,39 1,18
To al 59 0,91 58,25 1,04 56 1,03 61,21 1,12 38 0,93 59,46 0,99
Table 7.8: An h opome ic and spa io empo al gai cha ac e is ics o Scena io 2.1
young emale and wo om a middle-age). The second Exe cise clus e cha ac e ises he mixed
g oup, whe e now all he exe cises execu ed by male olun ee s appea oge he . Su p isingly,
he exe cises coming om women ha e he wo s a e age pe o mances o all age g oups and
clus e s (i.e., women in his clus e pe o med a he lowes a e age gai eloci y in compa ison
o women in same age ca ego y bu di e en clus e s). The second Exe cise clus e is again
he one con aining mo e ins ances, being mos o hem o middle-aged people (26 ou o 58).
Since mos o he s ides a e o ype wo, i is likely o say ha i is he mo e gene al gai shape
o all he s ides collec ed in his da ase . This clus e o s ides alls in he middle o s ide
leng h, being some o hem e y sho (a ound 2.5 seconds). The o ce compensa ion in he
swing phase o he gai cycle is qui e balanced, wi h a gene al end o applying mo e igh
o ces, especially when eaching he end o he pe iod.
The o he h ee clus e s could be unde s ood as a spli ing o he g oup o med mainly by
old ladies in scena io 2.1. As i can be obse ed in Figu e 7.11, he s ep ype has los ele ance
in he h ee i s Exe cise clus e s o become mo e signi ican in he las wo classes o exe cises.
I is especially impo an in he i h Exe cise clus e since people in his g oup combine he
wo s eps wi h majo o ce a ia ion du ing he swing phase, which should be analysed by an
expe eye o de e mine i his is a possible iden i ie o gai dis u bance o all isk.
7.2.2.3 Thi d Scena io
The hi d scena io p esen ed wi h he CVI da ase is he one wi h six ypes o s ides. Once he
bags-o -s ides a e gene a ed, he clus e ing o he his og ams e u ns wo possible solu ions
119
7. RESULTS
Clus e 1 2 3
Age N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h
Age M 2 1,16 64,61 1,16 26 1,05 59,52 1,18 2 1,03 65,22 1,04
F 2 1,16 64,61 1,16 16 0,99 57,32 1,14 2 1,03 65,22 1,04
M 10 1,14 63,04 1,25
Age O 13 0,50 54,13 0,58 11 0,88 61,33 0,93
F 13 0,50 54,13 0,58 8 0,85 60,00 0,91
M 3 0,97 64,88 0,97
Age Y 28 1,05 61,38 1,14 19 1,02 59,25 1,17 5 1,18 65,22 1,23
F 28 1,05 61,38 1,14 11 1,02 59,87 1,16 5 1,18 65,22 1,23
M 8 1,00 58,39 1,18
To al 30 1,06 61,59 1,14 58 0,91 58,22 1,04 18 0,98 62,84 1,02
Clus e 4 5
Age N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h
Age M 6 1,16 65,50 1,20 2 1,12 65,78 1,14
F 6 1,16 65,50 1,20 2 1,12 65,78 1,14
M
Age O 12 0,84 57,46 0,97 11 0,68 50,26 0,75
F 5 0,70 55,68 0,80 11 0,68 50,26 0,75
M 7 0,95 58,74 1,09
Age Y 7 1,23 64,84 1,31 9 1,02 60,80 1,10
F 7 1,23 64,84 1,31 9 1,02 60,80 1,10
M
To al 25 1,03 61,46 1,12 22 0,86 55,98 0,93
Table 7.9: An h opome ic and spa io- empo al gai cha ac e is ics o Scena io 2.2
wi h h ee and ou so s o exe cises, which a e ep esen ed in Figu es 7.14 and 7.15 espec-
i ely. As a gene al obse a ion om hese wo igu es, and in compa ison wi h he o he wo
scena ios, i is clea ha he mo e ypes o s ides we ha e, he mo e di icul i will be o ob ain
a ype o s ide which s ongly ep esen s an Exe cise clus e . In his hi d scena io, he mo e
p e alen ypes o s ides each up o hal o he s ides appea ing in ha clus e , while in he
i s scena io his could go up o 80%. Also, he mo e ypes o s ides we ha e, he ha de i is
o ind signi ican di e ences among hem. Obse ing Figu es 7.12 and 7.13, we can see ha
he i s s ides clus e ep esen he sho e s ides in ime and clus e 5 has he longe ones;
he es o clus e s ha e a mix o sho o long s ides, going om 3 o 5 seconds pe s ide.
In he i s case (Figu e 7.14), om he h ee Exe cise clus e s, only he second has a
p e alen ype o s ide. This clus e is again mainly composed by a mix o young and middle-
aged people bu also has a signi ican ep esen a ion o he mo e elde ly popula ion: 25% o
120
7.2 Clus e ing Resul s o he Gai Analysis
Bag-o -s ides in Exe cise Clus e 1
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 2
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 3
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 4
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Bag-o -s ides in Exe cise Clus e 5
12345
Clus e s o s ides
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
S ides dis ibu ion
Figu e 7.11: Scena io 2.2: c i5k5k’. Clus e ep esen a ion o bags-o -s ides wi h i e ypes
o s ides, g ouped in i e so s o exe cises.
he exe cises pe o med by people aged 80+ all in his clus e , despi e being hose wi h he
lowes gai eloci y (0.5 m/s) wi hin he olde people. As i can be obse ed, people in his
g oup apply mo e igh -hand o ce since alues o Fxdi a e always posi i e, al hough hey
emain balanced du ing he swing phase ( he e a e no sha p changes in he gai shape du ing
he gai cycle). The hi d Exe cise clus e is mainly ep esen ed by s ides ype 1 and 4 (which
a e ba ely p esen in he o he wo exe cise clus e s). These s ides a e he sho es in ime.
The s ide shape also indica es ha he o ce applied o he i-Walke a ies a each phase o he
cycle: when he igh s ep is aking place, he pushing o ce is mainly gi en by he igh o ce
and changes o he le o ce a he le s ep. The amoun o o ce applied in each case is qui e
simila ; hus in ha sense, hey a e compensa ed. In his case, 38 ou o 46 exe cises alling in
his clus e we e pe o med by young people, and in gene al, he gai eloci y was high, e en
o he h ee exe cises pe o med by old adul s. I is cohe en ha s ides ha a e sho in ime
belong o hose people who walk as e o o hose wi h sho e legs, as s a ed be o e.
The i s Exe cise clus e is he mos homogeneous in s ide dis ibu ion, excep o he
121
7. RESULTS
Clus e 1 2 3
Age N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h N Speed Cadence S ide
Leng h
Age M 7 1,09 66,04 1,14 26 1,05 61,68 1,18 5 1,18 69,97 1,18
F 7 1,09 66,04 1,14 16 0,99 59,36 1,14 5 1,18 69,97 1,18
M 10 1,14 65,39 1,25
Age O 32 0,78 57,16 0,87 12 0,52 56,50 0,59 3 0,82 58,56 0,96
F 23 0,72 54,71 0,81 12 0,52 56,50 0,59 2 0,68 60,92 0,73
M 9 0,94 63,42 1,01 1 1,11 53,83 1,41
Age Y 10 1,27 69,51 1,33 20 1,01 61,46 1,15 38 1,04 63,34 1,13
F 10 1,27 69,51 1,33 12 1,01 61,97 1,14 38 1,04 63,34 1,13
M 8 1,00 60,69 1,18
To al 49 0,93 60,95 1,00 58 0,92 60,53 1,05 46 1,04 63,75 1,12
Table 7.10: An h opome ic and spa io empo al gai cha ac e is ics o Scena io 3.1
s ide ypes h ee and ou which ha e almos no ep esen a ion: hese wo ypes a e he wo
mo e ele an in he o he wo clus e s o exe cises. S ides can be g ouped in wo ca ego ies,
acco ding o he amoun o o ce equi ed du ing he igh s ep phase: in he i s wo s ides,
people employ a ound 2N o pushing o ce in Fxdi , which is a easonable quan i y. The
wo las ypes o s ides exe a ound 6N o o ce in he same phases, which means ha use s
pe o ming hese s eps ely mo e on he suppo o e ed by he olla o . In addi ion, hese s ides
can also be di e en ly g ouped by shape: he second and ou h s ide ep esen gai cycles
whe e body o ces we e compensa ed du ing he le s ep pa o he cycle, wi h an inc easing
amoun o igh o ce applied du ing he igh s ep phases. The i s and las gai shapes p esen
a highe oscilla ion o o ces du ing he swing phase, i.e., pa icipan s pe o ming hese s ides
equi e highe suppo o bo h pa s o he body while walking. In Table 7.10, we can see ha
his g oup is mainly composed o olde adul s, wi h some ep esen a ion o he younge and
middle-age g oups. The s ides mo e ele an om he k-medoids esul belong o challenged
olde adul s wi h ascula demen ia o hype ension diagnosed, among o he s.
The second case, wi h ou ypes o exe cises, is qui e simila o he p e ious one: clus e s
1 and 4 ha e a p e alen s ide each one, which is he same han Scena io 3.1, bu also hose
ound in he second scena io. Clus e 1 is mainly composed o he younge emale pa ici-
pan s (al hough en ha e mo ed o o he clus e s) bu s ill con ains ew exe cises pe o med
by women om di e en age anges. Clus e 4 is once mo e he he e ogeneous one, wi h a
simila ep esen a ion o all g oups o ages bu sligh ly highe in middle-age indi iduals. This
clus e also has g ouped oge he all he exe cises pe o med by men aged 80- yea s old. As
122
7.3 CVI Clus e Explana ion om he Spa io-Tempo al Gai Cha ac e is ics
y = Ec1 y = Ec2 y = Ec3
Weig h Fea u e Weig h Fea u e Weig h Fea u e
1.49 h x 0.83 h x 2.18 h x
0.41 h z 0.66 h z 0.51 lh y
0.31 sLeng hCV 0.59 lh y 0.48 lh x
Table 7.14: Mos ele an ea u es o SVM in Scena io 1
o ep esen a ion. In he case o he SVM, Table 7.14 shows he mos ele an ea u es by
Exe cise clus e (which co espond o each ycolumn). The weigh is an absolu e alue and i
mainly ep esen s he impo ance o a ea u e o e he o he s: he la ge a e he weigh alues,
he mo e signi icance has hei associa ed a iable.
In his i s scena io, Exe cise clus e 1 con ains mainly ins ances o young emale ha , as
p e iously desc ibed based on Table 7.7 and Figu e 7.7, ha e he sho es s ide shape in leng h.
The pushing o ce applied in hese exe cises is qui e balanced. This is cohe en wi h he ele an
ea u es ob ained in Table 7.14. Mo eo e , his is he only class con aining a spa io- empo al
gai cha ac e is ic ( he coe icien o a iance o he s ide leng h) among i s mos ele an ones.
This suppo he obse a ions made in he p e ious sec ion ela ed o he common pa e n o
he s ide shape o his g oup. The second Exe cise clus e ep esen s hose s ides wi h highe
o ce a ia ion du ing he gai cycle. In his case, he SVM conside ed almos as impo an he
pushing and leaning o ces, bu i also gi es ele ance o he la e al o ce. This class con ains
mos exe cises pe o med by olde adul s wi h ca diologic pa hologies. This again backs up
he so s o s ides ha ep esen his Exe cise clus e : people equi ing o walking assis ance
show a highe a iance in he pushing o ce, a pa e n ha is p obably p esen in he es o
applied o ces. The hi d class conside s h x as he mos ele an ea u e, a om he sco es
ob ained by he es o a iables.
7.3.2 Second Scena io
The second scena io has i e so s o s ides and p esen s wo possible op ions o he numbe
o Exe cise clus e s, sepa a ed in scena ios 2.1 and 2.2.
In his case, he RF ob ains i s wo s accu acy (0.73) while SVM has i s bes pe o mance
( ied wi h he i s scena io esul , see Table 7.12). Compa ing Tables 7.14 and 7.15, we obse e
ha he accu acy esul s a e qui e simila . On he one hand, one o he Exe cise clus e s ha e
129

7. RESULTS
p ecision ecall 1-sco e suppo
Exe cise Clus 1 0.98 0.87 0.92 67
Exe cise Clus 2 0.89 0.77 0.83 65
Exe cise Clus 3 0.37 0.67 0.47 21
a g / o al 0.86 0.80 0.82 153
Table 7.15: Accu acy esul s o he SVM in Scena io 2.1
y = Ec1 y = Ec2 y = Ec3
Weig h Fea u e Weig h Fea u e Weig h Fea u e
1.85 h x 1.71 h x 0.74 h z
0.52 speed 0.36 speed 0.70 lh z
0.50 lh x 0.26 h z 0.55 lh y
Table 7.16: Mos ele an ea u es o SVM in Scena io 2.1
e y low p ecision sco es (0.43 in he i s scena io, 0.37 in his case). On he o he hand,
ano he class has ob ained a 0.98 o p ecision in bo h scena ios.
The exe cises alling in each case a e e y simila . The clus e wi h be e pe o mances
co esponds o he p e iously named as he mixed g oup, wi h highe gende and age balance
(i is also he clus e wi h mo e obse a ions). On he o he hand, he hi d clus e is mainly
composed by olde pa icipan s and a ew young and middle-aged women. Almos hal o he
o iginal exe cises o his clus e a e classi ied in he second clus e (which includes mos o he
exe cises pe o med by young women). I is possible ha people pe o ming he ou h s ide
ype depic ed in Figu e 7.9, which has a simila shape han he second s ide ype, co esponds
o he younge indi iduals ound in he hi d Exe cise clus e .
The mos ele an ea u es o he RF a e again h x,lh z and h z wi h weigh alues o
0.38, 0.16 and 0.14 espec i ely. In he case o he SVM, Table 7.16 depic s how he pushing
o ce along wi h he speed a e equally ep esen ed in classes 1 and 2, while he hi d class
gi es mo e impo ance o he leaning o ces. I we obse e he i h s ide ype in Figu e 7.9,
we can conclude ha indi iduals in his clus e ely mo e on he i-Walke as a suppo de ice
while walking. The amoun o exe ed pushing o ce can only be compensa ed by a signi ican
leaning o ce in o de o compensa e he body balance du ing he gai cycle. Mo eo e , he le
la e al o ce is also ele an in his class. I seems ha he combina ion o he applied o ces is
di e en han in he o he age anges, which could be an indica o o a pa hological gai .
130
7.3 CVI Clus e Explana ion om he Spa io-Tempo al Gai Cha ac e is ics
Scena io 2.2, wi h i e ypes o s ides and o exe cises, di e s om he p e ious one in
he accu acy de ia ion o he SVM algo i hm. This case is he i s ime ha he RF C oss
Valida ion accu acy excels he SVM. In he e, he esul s in accu acy a e be e in a e age
compa ed o he p e ious scena ios, as depic ed in Table 7.17, being he wo s sco e 0.52 o he
ou h Exe cise clus e . The bags-o -s ides belonging o his class a e he mos he e ogeneouly
dis ibu ed as depic ed in Figu e 7.11, wi h almos 40% o he s ides belonging o he i s ype
and he es equally dis ibu ed (excep o he second s ide which has almos no ele ance
in his g oup o exe cises). I is p essumable ha he walking pa e n o hese people show
cha ac e is ics o gai a iabili y. In addi ion, based on he spa io- empo al da a depic ed in
Table 7.9, people in his g oup ha e he longes s ide leng hs, independen ly o hei age ange.
Table 7.18 shows he h ee mo e ele an coe icien s pe class. In he case o he ou h
clus e , p e iously commen ed, he SVM gi es mo e ele ance o spa io- empo al cha ac e is-
ics such as he walking speed o he s ide leng h. The ins ances om his clus e ha ha e
allen in he i h class belong o and elde male adul wi h high isk o alling, wi h a s ide
leng h and walking speed signi ican ly below he pe o mances o o he male adul s in his
class. In addi ion, hese cha ac e is ics a e mo e simila o hose ound in he i h Exe cise
clus e . Howe e , due o he lack o male ep esen a ion in his da ase , we canno be su e o
he accu acy o his a i ma ion.
The hi d class, which con ains exe cises ha ha e a simila age dis ibu ion and s ide
leng h mean alues, also elies in spa io empo al ea u es o he classi ica ion (walking speed
and he coe icien o a iance o he s ide ime). Howe e , hese wo classes ha e ob ained he
wo s p ecision and ecall sco es; hey a e also hose paying less a en ion o he hand o ces
espec o he o he s. In gene al, we can obse e ha he mo e s ide and exe cise ypes we
ha e, he mo e ele an become he spa io- empo al ea u es, al hough he pushing o leaning
o ces a e s ill p esen a each class. In ac , RF keeps conside ing hem he mos ele an
ea u es, gi ing 0.38 o impo ance o h x, 0.14 o h z and 0.11 o lh z. We can say ha he
in e ac ion o he human wi h he i-Walke is e y impo an o classi y co ec ly he exe cises,
al hough esul s a e be e when he walking speed is also conside ed.
7.3.3 Thi d scena io
The hi d scena io conside s wo possible numbe s o exe cises o bags-o -s ides composed
by six ypes o s ides.
131
7. RESULTS
p ecision ecall 1-sco e suppo
Exe cise Clus 1 0.80 0.73 0.76 33
Exe cise Clus 2 0.97 0.92 0.94 61
Exe cise Clus 3 0.67 0.75 0.71 16
Exe cise Clus 4 0.52 0.65 0.58 20
Exe cise Clus 5 0.82 0.78 0.80 23
a g / o al 0.82 0.80 0.81 153
Table 7.17: Accu acy esul s o he SVM in Scena io 2.2
y = Ec1 y = Ec2 y = Ec3 y = Ec4 y = Ec5
Weig h Fea u e Weig h Fea u e Weig h Fea u e Weig h Fea u e Weig h Fea u e
2.79 h x 5.53 h x 4.471 sTimeCV 2.21 sLeng h 6.96 h x
2.63 sLeng hCV 4.41 speed 2.93 speed 1.90 speed 5.19 lh x
1.58 lh x 3.83 h y 2.36 lh z 1.28 h x 2.98 cadence
Table 7.18: Mos ele an ea u es o SVM in Scena io 2.2
The i s case ep esen s h ee clus e s o exe cises. The accu acy ob ained wi h C oss
Valida ion is 0.79 o RF and 0.75 o SVM. The mos impo an ea u es in RF a e again
h x, h z and lh z wi h 0.41, 0.19 and 0.14 o he o al ep en a ion. In his case, he le
leaning o ce is he one disc imina ing he obse a ions o he second Exe cise clus e , while
h z is mo e ele an o he o he wo clus e s. I we go a s ep u he in he decision ee,
we ind ha he cadence and he s ide ime a e s ongly co ela ed o he classi ica ion o he
Exe cise clus e s 1 and 3; mo e speci ically, small alues o hese a iables a e dis inc i e o
he exe cises classi ied in clus e 3. This is cohe en wi h he dis ibu ion o bags-o -s ides in
Exe cise clus e one (see Figu e 7.14) and wi h he p e alen s ides o his his og am (s ides
1 and 4, depic ed in Figu es 7.12 and 7.13 espec i ely). This is also ep esen ed in Table 7.20,
whe e he s ide ime appea s as one o he mos aluable coe icien s o he hi d class.
The i s Exe cise clus e has ob ained he wo s p ecision sco e, wi h mo e han hal o he
exe cises dis ibu ed in he o he wo clus e s, as shown in Table 7.19. The pa e n is simila o
he ones desc ibed in he p e ious scena ios. The mos ele an a iables o his clus e acco d-
ing o he SVM explana ion in Table 7.20 a e h z and lh z, which is cohe en wi h he s ide
shapes ep esen ed in his Exe cise clus e (see Figu es 7.12 and 7.14). I is p essumable ha
he SVM has classi ied people wi h a mo e la en o ce in e ac ion wi h he i-Walke oge he ,
aking he younge pa icipan s o o he clus e s. Tha means ha he indi iduals showing his
132
7.3 CVI Clus e Explana ion om he Spa io-Tempo al Gai Cha ac e is ics
p ecision ecall 1-sco e suppo
Exe cise Clus 1 0.47 0.74 0.58 31
Exe cise Clus 2 0.98 0.85 0.91 67
Exe cise Clus 3 0.85 0.71 0.77 55
a g / o al 0.83 0.78 0.79 153
Table 7.19: Accu acy esul s o he SVM in Scena io 3.1
y = Ec1 y = Ec2 y = Ec3
Weig h Fea u e Weig h Fea u e Weig h Fea u e
0.94 h z 1.86 h x 1.42 h x
0.68 lh z 0.50 lh x 0.55 h z
0.59 lh z 0.48 lh y 0.38 sTime
Table 7.20: Mos ele an ea u es o SVM in Scena io 3.1
in e ac ion wi h he i-Walke a e hose wi h highe dependence on he suppo ing de ice and,
hus, hose wi h highe isk (o ea ) o alling.
Finally, he second case o his hi d scena io consis s o six ypes o s ides and ou clus-
e s o exe cises. The RF has a mean accu acy sco e o 0.8 while he SVM e u ns he wo s
mean accu acy among all (0.70). This ime, he h ee mos impo an ea u es o RF a e h x
(0.33), h z (0.23) and lh z (0.09, which is closely ollowed by he s ide ime coe icien o
a iance), i.e., almos 60% o he da a can be explained wi h hese a iables. Su p insingly, he
igh leaning o ce does no appea among he mos ele an coe icien s o SVM, as shown in
Table 7.22. These a iables a e specially ele an o Exe cise clus e 4, which is he one wi h
bes p ecision pe o mance and is also he one wi h mo e classi ied objec s, which co esponds
o al eady men ioned he mixed g oup (see Table 7.11). Acco ding o Table 7.22, i s mos im-
po an a iables a e he walking speed, he igh pushing o ce h x and also he s ide leng h.
This clus e con ains mos o he exe cises pe o med by male pa icipan s, which p essumably
a e s onge and alle han women. Thus i combines spa io- empo al cha ac e is ics wi h he
o ce in e ac ion be ween he indi idual and he i-Walke .
On he o he hand, he i s and h id Exe cise clus e s a e hose wi h wo s esul s on p eci-
sion and ecall, al hough hey ha e signi ican ly imp o ed compa ed o he esul s ob ained in
he i s case o his scena io. I s mos ele an ea u es a e h x and lh z and h y; howe e ,
i has been p e iously men ioned ha his clus e is mainly o med by young women, which
main gai cha ac e is ic is he sho s ide leng h. I is likely o hink ha he i s case o his
133
7. RESULTS
p ecision ecall 1-sco e suppo
Exe cise Clus 1 0.69 0.76 0.72 29
Exe cise Clus 2 0.85 0.74 0.80 47
Exe cise Clus 3 0.61 0.78 0.68 18
Exe cise Clus 4 0.95 0.92 0.93 59
a g / o al 0.83 0.82 0.82 153
Table 7.21: Accu acy esul s o he SVM in Scena io 3.2
y = Ec1 y = Ec2 y = Ec3 y = Ec4
Weig h Fea u e Weig h Fea u e Weig h Fea u e Weig h Fea u e
2.56 h x 6.47 sLeng h 14.75 el 9.18 el
1.94 lh z 6.35 h x 12.56 sLeng h 6.11 h x
1.44 h y 6.17 el 4.53 sLeng hCV 5.92 sLeng h
Table 7.22: Mos ele an ea u es o SVM in Scena io 3.2
scena io is mo e accu a e in his sense, since i also conside s gai cha ac e is ics o ob ain a
mo e gene al desc ip ion o a gai cycle.
7.4 Modeling Fall Risk
As al eady explained in Chap e §3.3, he main objec i e o he I-DONT-FALL p ojec was
o design a physical and cogni i e ea men ocused on olde adul s a high isk o alling o
ha ing su e ed a all along he las yea . This ea men aimed o educe, no only he numbe
o alls, bu also he isk and ea o alling. The IDF da ase used, in oduced in §3.3.3 and
Table 5.1, was also used o de elop a model able o p edic he all isk o hose indi iduals a e
he ea men pe iod. The me hodology has been in oduced in Chap e §6.5 and he ollowing
esul s we e published in Co ´
es e al. (2016).
The bes accu acy ob ained om he model measu ed using 10- old c oss alida ion was
0.88 wi h s anda d de ia ion 0.1. The accu acy o he model o all he da a was 0.95. Table
7.23 p esen s he esul s om he model o each class.
The esul ing model was applied o he T1 da ase , di iding he da a be ween he indi iduals
ha we e submi ed o ea men and he indi iduals ha pe o med no ea men (placebo). The
goal is o es i hei all isk s a us had changed o no .
134

7.4 Modeling Fall Risk
p ecision ecall 1-sco e suppo
Low Fall Risk 0.91 0.77 0.83 13
High Fall Risk 0.96 0.99 0.97 72
a g / o al 0.95 0.95 0.95 85
Table 7.23: Accu acy esul s o he logis ic eg ession model o he aining se .
The placebo g oup is o med by 27 indi iduals, 24 o hem had high all isk, and 3 had
low all isk. The model ob ained wi h he T0 da a i s well o his da a, p edic ing he co ec
label o all he subjec s excep o wo high all isk indi iduals ha a e classi ied as low all
isk.
The ea men g oup is o med by 68 indi iduals, 48 o hem had high all isk, and 10 had
low all isk. The model p edic s he same label o 43 indi iduals (36 high all isk and se en
low all isk). The emaining 20 indi iduals ha e changed hei s a us.
This means ha his pa o he da ase has de ia ed om he ini ial model and he ea men
has had some e ec on he all isk s a us o he indi iduals. Speci ically, 12 high- isk all
indi iduals a e now conside ed low isk. S ill, i seems ha many indi iduals ha e no had
enough bene i om he ea men in e ms o all isk, and jus a mino i y ha e imp o ed hei
condi ion. Also, some indi iduals ha e wo sened hei s a us, p obably because o he na u al
cou se o hei medical condi ions.
Mos o he obse a ions in his da ase belong o people a high isk o alling. Mo eo e ,
i is ha d o obse e e iden imp o emen s in people a his age, since hey con inue o de elop
como bidi ies ha in he end a ec hei locomo ion and o he abili ies. I would be in e es ing
o ex end he da ase wi h a b oade age ange and isk all. I could also be ins umen al
o enla ge he ea men pe iod and he pos - ea men e alua ion o s udy he e ec s o such
ehabili a ion a long- e m pe spec i e. Possible u u e wo k could combine he p edic ed isk
o alling wi h he ca ego isa ion o he walking pe o mance o complemen he s a egies o
con ol o p o ide assis i e and sa e na iga ion.
135
7. RESULTS
136
Chap e 8
Conclusions
Olde adul s a e becoming he p edominan demog aphic g oup in mos de eloped socie ies.
The p ocess o his ageing socie y is al eady a ec ing, o will soon a ec , bo h de eloped
and de eloping coun ies. Senio s usually su e om one o mo e diseases and disabili ies
ela ed o age, causing a loss o esidual skills and, hence, au onomy. This o en poses a ba ie
o he elde ly, limi ing hei access o ela i es, iends and social ac i i ies, which may lead
o isola ion, dep ession and se e ely impac s hei QoL. Ageing also challenges he abili y
o pe o m he ac i i ies o daily li ing, such as d essing, ba hing o sel - eeding, bu also
o he unc ional mobili y. As such, solu ions bo h e icacious and cos -e ec i e need o be
sough wi h wo main objec i es: (i) imp o e, o a leas main ain, he QoL o senio s and hei
ela i es, and (ii) e hink public heal h sys ems, aking le e age o echnological solu ions ha
allow a emo e and ubiqui ous heal h managemen (see §3.2,§Bo Ba u´
e e al. (2017)).
The In e ne o Medical Things is a new concep eme ging ecen ly ha akes in o accoun
his cu en si ua ion. I aims o enable he machine o machine in e ac ion and eal- ime in-
e en ion solu ions o c ea e a ne wo k o connec ed de ices con inuously collec ing and/o
p ocessing da a. Concep s al eady explained in his documen , such as sma wea able o assis-
i e de ices, home-use medical de ices o mobile heal hca e applica ions, will be he p incipal
componen s ha will allow communica ing wi h medical expe s emo ely. This sys em could
no only be used o moni o ing, bu also o p e en ion, heal hy li es yle p omo ion o emo e
in e en ion in eme gency si ua ions. This is also empowe ed in Eu ope by he H2020 und-
ing p og ammes, which aim o include he new gene a ion o mHeal h solu ions ha will help
use s o manage hei heal h and communica e wi h ela ed s akeholde s o he ca e p ocess
(e.g., doc o s, in o mal ca egi e s, ela i es, e c.). The i-Walke is a de ini e candida e o play
137
8. CONCLUSIONS
a ole in his heal h sys em a chi ec u e as i mee s he equi emen s needed o con ibu e o i s
achie emen .
Th ough his wo k, he s ong ela ionship be ween cogni ion and mobili y has been shown,
as well as he impo ance o being ac i e while ageing o main ain, no only ou skills bu also
ou au onomy in communi y-dwelling. In he case o mobili y eco e y and assis ance, we ha e
e iewed se e al solu ions de eloped du ing he las wo decades ha in ol e di e en assis i e
echnologies and senso s and in e ac a a ious le els wi h end-use s and heal h p o essionals.
Wi hin his ield, sma walke s o e an exci ing oppo uni y o he senio popula ion as a
suppo ool o gai and balance. Howe e , hey also p esen some limi a ions o speci ic
use s wi h physical o cogni i e impai men s depending on he numbe o legs and wheels,
such as pos -s oke use s (see §4.3.1 and Giuliani e al. (2012), Mo one e al. (2016)).
The i-Walke has been p esen ed in his documen as a sma walke wi h a sys em o
senso s and ac ua o s ha is al eady able o assis people in di e en ypes o en i onmen s
wi h a eac i e con ol, p o iding sa e y and sel -con idence o he use . We ha e also e iewed
p e ious esea ch wo ks in which he i-Walke has an essen ial ole as a ehabili a ion ool
(Giuliani e al. (2012)) and as an in elligen se ice in eg a ed in o a medical social ne wo k
(Ba u´
e e al. (2015)).
One o he main challenges when wo king wi h a obo ic ool wi h senso s is he in e -
p e a ion o aw da a. In he p e ious wo ks de eloped wi h he i-Walke (see Chap e §4.3
and Annex §B). In his PhD, we ha e been wo king in u ning hese da a o eadable and un-
de s andable in o ma ion o clinicians in di e en o ma s such as ac i i y epo s, g aphics
showing he e olu ion o a senso eading h ough an exe cise o messages ia social ne wo k
communica ion channels. The esul s ob ained in he wo ks as men ioned abo e we e s udied
om a clinical pe spec i e, whe e scales and pe o mances a e compa ed be ween pe iods. In
§4.3.3, we ha e also p esen ed wo app oaches o da a analysis using unsupe ised lea ning
echniques o ca ego ise exe cises by ype (s aigh s, u ns) o by pa icipan s’ age. Howe e , i
is also in e es ing o analyse he human- obo in e ac ion om a biomechanical poin o iew,
i.e., by in e p e ing he da a ex ac ed om he onboa d senso s and ela e i o he s udy
o gai cha ac e is ics o old-age indi iduals wi h high isk o alling. The i-Walke o e s a
unique oppo uni y o lea n he e ec o o ces exe ed di ec ly by a pe son a e e y momen
while walking and o alida e ou hypo hesis ha wi h his in o ma ion i is possible o de e -
mine pa e ns o use in e ac ion wi h he i-Walke as well as assessing he isk o alling in a
nea - u u e. Th ough his documen , we ha e men ioned di e en echniques o gai analysis
138