Robot for Coaching during Gait Training with Lokomat:Preliminary Experiment with a Multiple Sclerosis Patient

Nathalia CéspedesColombian School of Engineering

Bogotá, Colombianathalia.cespedes@mail.escuelaing.

edu.co

Jonathan CasasColombian School of Engineering

Bogotá, Colombiajonathan.casas@escuelaing.edu.co

Betsy Jaramillo VelasquezMobility Rehabilitation Center

Bogotá, Colombiabetsy8522@gmail.com

Catalina GómezClínica Universidad de la Sabana

Bogotá, Colombiamacatalina@gmail.com

Marcela MúneraColombian School of Engineering

Bogotá, Colombiamarcela.munera@escuelaing.edu.co

Carlos A. CifuentesColombian School of Engineering

Bogotá, Colombiacarlos.cifuentes@escuelaing.edu.co

ABSTRACTPhysical Rehabilitation with Lokomat has been suggested as animportant therapy for gait rehabilitation after neurological disor-ders. Two issues of this approach have been identified: on the onehand, there is a lack of adherence of patients to the programs. Onthe other hand, there are multiple tasks performed by the therapistsuch as: monitoring physiological variables and controlling propergait patterns during sessions. In this context, social robots could bebeneficial as a tool to cooperate with therapists to control the pa-tient’s performance and to provide motivation and encouragementduring the treatment. This work presents a Human-Robot interfacethat monitors and provides feedback to the patient according to thecervical and thoracic posture, cardiovascular performance and Borgscale. A preliminary evaluation of this approach with a multiplesclerosis patient is presented. This study shows the potential ofrobots for coaching in physical rehabilitation in terms of supportand accompaniment tool in Lokomat therapies.

KEYWORDSSocial Assisted Robotics, Physical Rehabilitation, Lokomat

ACM Reference Format:Nathalia Céspedes, Jonathan Casas, Betsy Jaramillo Velasquez, CatalinaGómez, Marcela Múnera, and Carlos A. Cifuentes. 2017. Robot for Coachingduring Gait Training with Lokomat: Preliminary Experiment with aMultipleSclerosis Patient. In Proceedings of HRI (CHICAGO’18). ACM, New York, NY,USA, 5 pages. https://doi.org/

1 INTRODUCTIONAccording to the World Health Organization, around 15% of theworld’s population has some disability, mostly caused by neuro-logical diseases such as stroke and spinal cord injuries [17]. These

Permission to make digital or hard copies of all or part of this work for personal orclassroom use is granted without fee provided that copies are not made or distributedfor profit or commercial advantage and that copies bear this notice and the full citationon the first page. Copyrights for components of this work owned by others than ACMmust be honored. Abstracting with credit is permitted. To copy otherwise, or republish,to post on servers or to redistribute to lists, requires prior specific permission and/or afee. Request permissions from permissions@acm.org.CHICAGO’18, March 2018, Chicago, IL USA© Association for Computing Machinery.

pathologies cause limitations in lower and upper limb motor abil-ities [2] [1]. Physical Rehabilitation (PR) is a continuous processthat involves the evaluation, definition, planning and developmentof different strategies, with the aim of improving the quality oflife and self-reliance of patients who have suffered a neurologicaldisease [17].PR is focused on two main aspects: (1) physiological aspects relatedto the regulation of biological functioning, (2) cognitive aspectsrelated to cognition processes and the processing of informationsuch as language, memory, attention, motivation and perception[18]; these aspects are important to evaluate the evolution of the pa-tient during the programs. Currently, there are several PR methodsbased on conventional therapies and robot assisted gait trainingrehabilitation.Nowadays, Lokomat (Hocoma, Switzerland) is the gold standarddevice in these sessions [21]. The Lokomat is a robotic orthoticdevice that adjusts to the lower limbs of the patient with the gen-eral purpose of retraining the gait through repetitive and intensivesimulation exercises. It uses sensory stimulation fed by propriocep-tive feedback games improving the neuroplasticity for a functionalrecovery [4] [5]. In different studies the positive effects of Lokomathave been demonstrated, increasing muscular tone, balance, motorcontrol and muscular strength [22], compared with conventionaltherapies [7].Two of the most relevant issues presented in PR are: on the onehand, adherence to the programs [10]. Around 42% of the patientsabandoned rehabilitation sessions. Although factors that act as abarrier are not clear, it is believed that some of them could be: lowaerobic performance during exercises, anxiety, low motivation, de-pression and other cognitive conditions [10]. On the other hand,multiple task performed by the therapist during a session [6], forexample, simultaneous measurements and feedbacks of physiologi-cal variables.In this context, Socially Assistive Robotics (SAR) could be beneficialas a tool to cooperate with therapists to control patient’s perfor-mance and to provide motivation and encouragement during thetreatment.

CHICAGO’18, March 2018, Chicago, IL USA N. Céspedes et al.

2 RELATEDWORKOver the past years, SAR have been introduced in different scenarios.Mataríc et al. used SAR for post-stroke rehabilitation, demonstrat-ing the novel use of an autonomous mobile platform with severalfeedback and monitoring functions, showing potential therapeuticbenefits and patient’s positive response in achieving different goals[14]. Jung et al. examined the functional utility of an embodied ro-bot agent (uBot-5) as an interactive medium in stroke rehabilitation.The study was performed by one male patient during 8 sessions,the subject demonstrated improvement in the average number oftrials accomplished per minute during upper-limb exercises [11]. C.Kidd and C. Breazel in 2008, studied the long-term effects of human-robot interaction in coaching with the aim of reducing the ratesof overweight and obesity. In this case, the robot asked patientstheir diet goals in terms of burning calories during the exerciseand data related to the food consumed during the day. The resultsshowed that the participants assisted by the social robot were moreinterested in knowing the calories consumption and exercise per-formed than those who used other methods [12]. In 2012 a socialrobot was designed to engage elderly users in physical exercise.During the session, the robot asked to the user to perform simpleseated arm gestures and the user was able to communicate with therobot. The study consisted of two conditions: relational condition,where the robot employed social interaction and personalizationapproaches, and nonrelational condition where the robot only pro-vided instructional exercises. The results of the study show a strongusers preference for the relational robot, the exercise system waswell received and effective in motivating physical exercises [8].NAO robot also has been a tool to assist conventional physiother-apy practices, by performing 9 exercises selected since the frequentmistakes or inaccuracies in the execution that patients usually do inthe execution of the exercise, for example: abduction of the shoulder.This study shows that patients improve the movement’s technicalquality when the robot perform the exercises according to the syn-chronization speed during the human- robot interaction [19].This work presents a Human-Robot interface that monitors andprovides feedback to the patient according to patient’s performance,which is in order to deal with motivational and engagement issues.Additionally, this interface intends to reduce repetitive therapist’stask during the treatment with Lokomat, such as corrective tasksfocus on, both, rectifying patient’s gait pattern, and monitoringcardiovascular performance and Borg scale. The proposed interfaceintegrates: (1) a Heart Rate monitor sensor to continuous monitorcardiac activity, (2) two Inertial Measurement Units (IMU) to es-timate cervical and thoracic posture, and (3) a Joystick to choosethe Borg Scale (perceived exertion scale). A NAO robot (AldebaranRobotics, France) is used to give feedback to the user with verbalgestures according to on-line measurements.

3 MATERIALS AND METHODSThis work was developed in two stages: first, a description of theHuman-Robot interface, and second, a protocol carried out in thepreliminary study where SAR system helps the therapist to assist apatient in order to measure his performance during the session.

3.1 Human-Robot InterfaceThe proposed Human-Robot interface is based on a previous studydeveloped in [16]. The authors performed a survey on a group thera-pist and patients, who are undergoing Lokomat therapy at MobilityRehabilitation Center and Clínica Universidad de la Sabana, whichare two institutions involved in this research. This work includesa set of variables relevant to be measured to develop a robot forcouching during gait training in therapy with Lokomat (Fig 1), asit is described below.

Figure 1: Interface architecture diagram for physical reha-bilitation.

Cardiovascular parameters: tomonitor heart function during theexercises guided by the Lokomat[3], a heart rate monitor, ZephyrHxM BT (Medtronic, USA) was located with an elastic band in thelower part of the patient’s sternum.Spinal posture parameters: IMU node (Fig 1) reports the degreesto estimate the spinal posture. Since it is an affected variable duringthe sessions and is constantly corrected by the therapist [9, 20], tworeal-time IMUs (Bosch, Germany) were used to measure cervicaland thoracic inclination angles at pitch, roll and yaw rotations. Onesensor was located in the user’s forehead and the other one betweenT6-T7 spinal segments adjusted with velcro straps [13].Cognitive parameters:The borg scale which represents the pa-tient’s perception of the effort during physical activity [15] and itis controlled by a joystick sensor that allows the patient to insertthe number related to the scale. Besides, motivational feedback ismade by a social robot NAO using randomly verbal phrases.All of these sensors have different transmission rates and samplingfrequencies which represent an issue for the on-line system. Tosolve this issue, the processes of each sensor are made in differentnodes, each node is used as a driver to obtain the raw-data of thesensors and perform an initial processing used for the transmission.

Robot for Coaching during Gait Training with Lokomat: Preliminary Experiment with a Multiple Sclerosis PatientCHICAGO’18, March 2018, Chicago, IL USA

As can be observed in Fig 1, this data is acquired by the Index Man-ager, which performs a downsampling (with a sampling frequencyof 1Hz) in order to manage and join all nodes, including the NAOmodule, graphical user interface and the database.Currently, the robot interacts with the patient through the feedbackof the user’s performance in two ways: (1) in the spinal posturecase, NAO gives a verbal advice if the angles do not correspond toan upright posture. (2) NAO provides feedback according to theBorg scale selected by the user. Additionally, the patient has accessto a graphical user interface feedback by means of visual reports ofthe sensors data.

3.2 Preliminar Study DesignAn experimental study was designed with the main objective ofobserving the functionality and usability of the Human-Robot in-terface. As shown in Fig 2 one patient was randomly chosen to do atherapy assisted by a social robot, a multiple sclerosis male patient(1.83 m, 60Kg, 62 years old) active in Lokomat therapies. For thisprotocol the exclusion criteria included cognitive pathologies thatcould affect the understanding of the protocol.The Lokomat settings are: treadmill speed at 1.5 m/s, distance of652.1 m, 29.2% of body weight support, therapy time of 30 minutes,based on the therapist experience. Finally the subject stood in acool down phase. The measurable parameters were: spinal posture,hear rate and Borg Scale.

Figure 2: Experimental setup atMobilityRehabilitationCen-ter

4 RESULTSAccording to the protocol described in the methods and materialssection, the interface was on-line for 30 minutes. Fig 3 and Fig 4show the evolution of postural behavior. The most relevant inclina-tions of the body were observed on the sagittal plane corresponding

to the pitch axis of the IMU sensor, taking into account the move-ments in the other planes are controlled by the Lokomat’s features .Bad posture described by a threshold occurs when angles increasein a range of 10-15 degrees (blue and red horizontal line in Fig 3and Fig 4), over the initial position, in this case 0 degrees.

Figure 3: Cervical posture registered by pitch, yaw and rollangles during 30 min of Lokomat session

Figure 4: Thoracic posture registered by pitch, yaw and rollaxis during 30 min of Lokomat session

Comparing spinal postures in pitch axis, cervical posture reachedmore frequently the threshold than the thoracic posture, due to thecharacteristics of the Lokomat, which has a support that restrictsmovements in all the lumbar section and a part of the thoracicsection. In this cases the robot interacted with the patient throughphrases that encourage the patient to improve the posture, for moti-vational purposes, three of the main phrases are : " Your head has abad posture, please tilted it back", "you are doing it right" and " Yourtorso is bowed, you can improve your posture, please correct it".For cervical posture (see Fig 3), the patient reached the threshold

CHICAGO’18, March 2018, Chicago, IL USA N. Céspedes et al.

angle 8 occasions, with evident inclinations at minutes 10 and 17.Similarly, for thoracic posture (see Fig 4) the patient reached thethreshold angle 4 times at minutes 4, 6.3, 17.4 and 23.Heart rate and Borg scale were also measured. Fig 5 represents theheart rate data during the session, including the last 10 minutescorresponding to the cool-down phase. The exercises performedduring Lokomat sessions weren’t of high intensity, thus in Fig 5a normal heart rate was observed, keeping in mind intensity ofroutines with Lokomat.

Figure 5: Heart rate during 30 min of Lokomat session

The borg scale is required during the complete session (Fig 6), theperception of the patient was low and moderate effort, at the 20thminute the hardness perceived level increase due the therapy con-ditions related to the Lokomat’s treadmill speed. Finally as shown

Figure 6: Borg Scale during 30 min of Lokomat session

in Fig 7, different events occurred during the therapy, includingBorg scale with a request frequency of 3 minutes, NAO motivationinterventions and postural feedback.

5 DISCUSSIONThe figures in the results section show heart rate, IMU’s and joysticksensors were appropriated to measure physiological and cognitive

Figure 7: Main events during 30 min of Lokomat session

aspects during the complete therapy. Usability of the system wascomfortable for the patient and did not disrupt the physical inter-vention or the therapist.The feedback given by NAO allowed the interaction with the pa-tient. During this preliminary session, a positive impact was ob-served in aspects such as attention, in cases when the robot asksthe user to improve the posture and motivation provided by verbalphrases. As shown in Fig 7 the social robot supported differentevents during the 30 minutes of therapy. As mentioned before thereare correctional, motivational and advice feedbacks, where the cer-vical posture correction and the motivational feedback were themost frequent events. An initial observation was that NAO accom-panied the therapy in cases when a poor posture occurred and itwas not perceived by the therapist, being an important tool to re-duce the number of tasks performed by the therapist. For instance,in different moments of the exercise where the spinal postureswere corrected by the robot and kinematic gait parameters werecorrected at the same time by the therapist.This leads towards an interesting research questions about the useand the benefits of social assistive robots in physical rehabilitationwith Lokomat. Currently the design of the interface is being up-dated to include a strong relational robot, through memory anddynamic body gestures, and time-extending studies with Lokomatphysical rehabilitation patients.

6 CONCLUSIONSIn the present work, an interface was designed, it combines relevantcognitive and physiological parameters with the aim of assessinguser’s performance in PR with Lokomat, using a sensory systemcomposed by a heart rate monitor, inertial measurements units anda joystick obtaining the heart function, cervical posture, thoracicposture and effort perception. These variables can be acquired andpresented in real-time to the patients and the medical team, and the

Robot for Coaching during Gait Training with Lokomat: Preliminary Experiment with a Multiple Sclerosis PatientCHICAGO’18, March 2018, Chicago, IL USA

architecture allows data processing, on-line feedback, visualization,and data management. The functionality and usability of the systemfor this therapy was appropriated, showing reliable measurements.Moreover, the robot gave different feedback corresponding to thevariables and motivate the patient with randomly verbal phrases,allowing the interaction with the patient. This study shows the po-tential of SAR in physical rehabilitation with Lokomat for coachingin terms of supporting the patient and accompanying the thera-pist’s task. Regarding the observations made in the preliminarypilot study, a well-received behavior and a positive impact by thepatient and the therapist were seen during the session.In the future, it is important to test the social robot impact with along term study following a greater number of Lokomat patients al-lowing a complete statistical analysis, finding relevant informationabout the measurable parameters in different sessions that includea social robot agent. Also, improvements to the Human-Robot inter-face will be conducted, by adding other relevant clinical variablesand a more relational NAO through visual feedback of the posturalbehavior.

ACKNOLEDGMENTThis work was supported by the Industry Academia PartnershipProgramme Colombia UK (IAPP/1516/137 / 2016-2018): Human-Robot Interaction Strategies for Rehabilitation based on SociallyAssistive Robotics.

REFERENCES[1] 2014. OMS | Lesiones medulares. WHO (2014). http://www.who.int/mediacentre/

factsheets/fs384/es/[2] 2015. OMS | Accidente cerebrovascular. WHO (2015). http://www.who.int/topics/

cerebrovascular[3] Juul Achten and Asker E Jeukendrup. 2003. Heart rate monitoring: applications

and limitations. Sports medicine (Auckland, N.Z.) 33, 7 (2003), 517–38. http://www.ncbi.nlm.nih.gov/pubmed/12762827

[4] Rocco Salvatore Calabrò, Simone Reitano, Antonio Leo, Rosaria De Luca, Cor-rado Melegari, and Placido Bramanti. 2014. Can robot-assisted movementtraining (Lokomat) improve functional recovery and psychological well-beingin chronic stroke? Promising findings from a case study. Functional neurol-ogy 29, 2 (2014), 139–41. http://www.ncbi.nlm.nih.gov/pubmed/25306125http://www.pubmedcentral.nih.gov/articlerender.fcgi?artid=PMC4198163

[5] G Colombo, M Joerg, R Schreier, and V Dietz. [n. d.]. Treadmill training ofparaplegic patients using a robotic orthosis. Journal of rehabilitation researchand development 37, 6 ([n. d.]), 693–700. http://www.ncbi.nlm.nih.gov/pubmed/11321005

[6] Heather E. Douglas, Magdalena Z. Raban, Scott R. Walter, and Johanna I. West-brook. 2017. Improving our understanding of multi-tasking in healthcare: Draw-ing together the cognitive psychology and healthcare literature. Applied Er-gonomics 59 (2017), 45–55. https://doi.org/10.1016/j.apergo.2016.08.021

[7] Ana Esclarín-Ruz, Monica Alcobendas-Maestro, Rosa Casado-Lopez, GuillermoPerez-Mateos, Miguel Angel Florido-Sanchez, Esteban Gonzalez-Valdizan, andJose Luis R. Martin. 2014. A Comparison of Robotic Walking Therapy andConventional Walking Therapy in Individuals With Upper Versus Lower MotorNeuron Lesions: A Randomized Controlled Trial. Archives of Physical Medicineand Rehabilitation 95, 6 (jun 2014), 1023–1031. https://doi.org/10.1016/j.apmr.2013.12.017

[8] Juan Fasola and Maja J. Matarić. 2012. Using socially assistive human-robotinteraction to motivate physical exercise for older adults. Proc. IEEE 100, 8 (2012),2512–2526. https://doi.org/10.1109/JPROC.2012.2200539

[9] Zeev Meiner Isabella Schwartz, Ana Sajin, Iris Fisher, Martin Neeb, Mara Sochina,Michal Katz-Leurer. 2009. The Effectiveness of Locomotor Therapy Using Robotic-Assisted Gait Training in Subacute Stroke Patients: A Randomized ControlledTrial. PM&R 1, 6 (jun 2009), 516–523. https://doi.org/10.1016/J.PMRJ.2009.03.009

[10] Kirsten Jack, Sionnadh Mairi McLean, Jennifer Klaber Moffett, and Eric Gardiner.2010. Barriers to treatment adherence in physiotherapy outpatient clinics: Asystematic review. Manual Therapy 15, 3 (2010), 220–228. https://doi.org/10.1016/j.math.2009.12.004

[11] Hee-Tae Jung, Jennifer Baird, Yu-Kyong Choe, and Roderic A. Grupen. 2011.Upper-limb exercises for stroke patients through the direct engagement of anembodied agent. Proceedings of the 6th international conference on Human-robotinteraction - HRI ’11 298, 0704 (2011), 157. https://doi.org/10.1145/1957656.1957709

[12] Cory D. Kidd and Cynthia Breazeal. 2008. Robots at home: Understandinglong-term human-robot interaction. 2008 IEEE/RSJ International Conference onIntelligent Robots and Systems, IROS (2008), 3230–3235. https://doi.org/10.1109/IROS.2008.4651113

[13] Hyunho Kim, Sang-Hoon Shin, Jeong-Kyun Kim, Young-Jae Park, Hwan-SupOh, and Young-Bae Park. 2013. Cervical Coupling Motion Characteristics inHealthy People Using a Wireless Inertial Measurement Unit. Evidence-BasedComplementary and Alternative Medicine 2013 (2013), 1–8. https://doi.org/10.1155/2013/570428

[14] Maja J Matarić, Jon Eriksson, David J Feil-Seifer, and Carolee J Winstein. 2007. So-cially assistive robotics for post-stroke rehabilitation. Journal of NeuroEngineeringand Rehabilitation 4, 1 (2007), 5. https://doi.org/10.1186/1743-0003-4-5

[15] Jean Yves SalleMaxene Compagnant, Jean Cristophe Daviet, StéphaneMandigout,Justine Lcroix, Nicolas Vuillerme. 2017. Reliability of the rating of perceivedexertion (Borg Scale) in post-stroke during 2 tasks of daily life. Annals of Physicaland RehabilitationMedicine 60 (sep 2017), e1–e2. https://doi.org/10.1016/J.REHAB.2017.07.017

[16] Marcela Munera, Alexandra Marroquin, Laura Jimenez, Juan S. Lara, CatalinaGomez, Sandra Rodriguez, Luis E. Rodriguez, and Carlos A. Cifuentes. 2017.Lokomat therapy in Colombia: Current state and cognitive aspects. 2017 (jul2017), 394–399. https://doi.org/10.1109/ICORR.2017.8009279

[17] World Health Organization. 2011. full-text. 4, Rehabil (2011).[18] Susan B. O’Sullivan, Thomas J. Schmitz, and George D. Fulk. [n. d.]. Physical

rehabilitation. 1505 pages.[19] David López Recio. [n. d.]. The NAO models for the elderly. ([n. d.]).[20] Susan Ryerson, Nancy N. Byl, David A. Brown, Rita A. Wong, and Joseph M.

Hidler. 2008. Altered Trunk Position Sense and Its Relation to Balance Functionsin People Post-Stroke. Journal of Neurologic Physical Therapy 32, 1 (mar 2008),14–20. https://doi.org/10.1097/NPT.0b013e3181660f0c

[21] Guangyu Liu Shahid Hussain, Sheng QuanXie. 2011. Robot assisted treadmilltraining: Mechanisms and training strategies. Medical Engineering & Physics 33,5 (jun 2011), 527–533. https://doi.org/10.1016/J.MEDENGPHY.2010.12.010

[22] L. Wallard, G. Dietrich, Y. Kerlirzin, and J. Bredin. 2015. Effects of robotic gaitrehabilitation on biomechanical parameters in the chronic hemiplegic patients.Neurophysiologie Clinique 45, 3 (sep 2015), 215–219. https://doi.org/10.1016/j.neucli.2015.03.002