上海交通大学：《生物与机器人学》课程教学资源_Rehabilitation and Health Care Robotics

1223 53.Rehabilitation and Health Care Robotics H.F.Machiel Van der Loos,David J.Reinkensmeyer The field of rehabilitation robotics develops robotic 53.2 Physical Therapy and Training Robots....1227 systems that assist persons who have a disability 53.2.1 Grand Challenges and Roadblocks..1227 with necessary activities,or that provide therapy 53.2.2 Movement Therapy after Neurologic Injury ..1228 for persons seeking to improve physical or cog- 53.2.3 Robotic Therapy nitive function.This chapter will discuss both of for the Upper Extremity... ..1229 these domains and provide descriptions of the ma- 53.2.4 Robotic Therapy for Walking..........1231 jor achievements of the field over its short history. Specifically,after providing background informa- 53.3 Aids for People with Disabilities. ..1235 tion on world demographics (Sect.53.1.2)and the 53.3.1 Grand Challenges history (Sect.53.1.3)of the field,Sect.53.2 de- and Enabling Technologies.............1235 scribes physical therapy and training robots,and 53.3.2 Types and Examples Sect.53.3 describes robotic aids for people with of Assistive Rehabilitation Robots...1236 disabilities.Section 53.4 then briefly discusses re- cent advances in smart prostheses and orthoses 53.4 Smart Prostheses and Orthoses..............1240 that are related to rehabilitation robotics.Finally, 53.4.1 Grand Challenges and Roadblocks..1240 53.4.2 Targeted Reinnervation.................1240 Sect.53.5 provides an overview of recent work in 53.4.3 Brain-Machine Interfaces diagnosis and monitoring for rehabilitation as well .1241 53.4.4Advances in Neural Stimulation.....1241 as other health-care issues.At the conclusion of 53.4.5 Embedded Intelligence.................1242 this chapter,the reader will be familiar with the history of rehabilitation robotics and its primary 53.5 Augmentation for Diagnosis accomplishments,and will understand the chal- and Monitoring........ ..1242 lenges the field faces in the future as it seeks to 53.5.1 Introduction:Grand Challenges improve health care and the well-being of persons and Enabling Technologies............1242 with disabilities.In this chapter,we will describe 53.5.2 Smart Clinics with Automated an application of robotics that may in the future Health Care Monitoring and Care....1243 touch many of us in an acutely personal way. 53.5.3 Home-Based Rehabilitation Monitoring Systems..1243 53.10 verview....1223 53.5.4Wearable Monitoring Devices.........1244 53.1.1 Taxonomy 53.6 Safety,Ethics,Access,and Economics.....1244 of Rehabilitation Robotics...........1224 53.1.2 World Demographics..................1224 53.7 Conclusions and Further Readings..........1245 53.1.3 Short History of the Field of Rehabilitation Robotics .1225 References..… .1246 53.1 Overview When we become unable to interact physically with when one of our family members,friends or neighbors our immediate environment as we desire in order to is in this situation,we seek technology-based solutions achieve our personal goals through injury or disease,or to assist us in relearning how to complete our activi-

1223 Rehabilitatio 53. Rehabilitation and Health Care Robotics H.F. Machiel Van der Loos, David J. Reinkensmeyer The field of rehabilitation robotics develops robotic systems that assist persons who have a disability with necessary activities, or that provide therapy for persons seeking to improve physical or cog￾nitive function. This chapter will discuss both of these domains and provide descriptions of the ma￾jor achievements of the field over its short history. Specifically, after providing background informa￾tion on world demographics (Sect. 53.1.2) and the history (Sect. 53.1.3) of the field, Sect. 53.2 de￾scribes physical therapy and training robots, and Sect. 53.3 describes robotic aids for people with disabilities. Section 53.4 then briefly discusses re￾cent advances in smart prostheses and orthoses that are related to rehabilitation robotics. Finally, Sect. 53.5 provides an overview of recent work in diagnosis and monitoring for rehabilitation as well as other health-care issues. At the conclusion of this chapter, the reader will be familiar with the history of rehabilitation robotics and its primary accomplishments, and will understand the chal￾lenges the field faces in the future as it seeks to improve health care and the well-being of persons with disabilities. In this chapter, we will describe an application of robotics that may in the future touch many of us in an acutely personal way. 53.1 Overview.............................................. 1223 53.1.1 Taxonomy of Rehabilitation Robotics ............. 1224 53.1.2 World Demographics..................... 1224 53.1.3 Short History of the Field of Rehabilitation Robotics ............. 1225 53.2 Physical Therapy and Training Robots .... 1227 53.2.1 Grand Challenges and Roadblocks .. 1227 53.2.2 Movement Therapy after Neurologic Injury .................. 1228 53.2.3 Robotic Therapy for the Upper Extremity................. 1229 53.2.4 Robotic Therapy for Walking .......... 1231 53.3 Aids for People with Disabilities............. 1235 53.3.1 Grand Challenges and Enabling Technologies ............ 1235 53.3.2 Types and Examples of Assistive Rehabilitation Robots... 1236 53.4 Smart Prostheses and Orthoses .............. 1240 53.4.1 Grand Challenges and Roadblocks .. 1240 53.4.2 Targeted Reinnervation................. 1240 53.4.3 Brain–Machine Interfaces ............. 1241 53.4.4Advances in Neural Stimulation ..... 1241 53.4.5Embedded Intelligence ................. 1242 53.5 Augmentation for Diagnosis and Monitoring .................................... 1242 53.5.1 Introduction: Grand Challenges and Enabling Technologies ............ 1242 53.5.2 Smart Clinics with Automated Health Care Monitoring and Care.... 1243 53.5.3 Home-Based Rehabilitation Monitoring Systems ...................... 1243 53.5.4Wearable Monitoring Devices......... 1244 53.6 Safety, Ethics, Access, and Economics ..... 1244 53.7 Conclusions and Further Readings.......... 1245 References .................................................. 1246 53.1 Overview When we become unable to interact physically with our immediate environment as we desire in order to achieve our personal goals through injury or disease, or when one of our family members, friends or neighbors is in this situation, we seek technology-based solutions to assist us in relearning how to complete our activi￾Part F 53

1224 Part F Field and Service Robotics ties of daily living(ADLs).or to assist us in actually Assistive robots are generally grouped accord- doing them if we are unable to relearn.While hu- ing to whether they focus on manipulation,mobility, man therapists and attendants can indeed provide the or cognition.Manipulation aids are further classi- types of assistance required,the projected short-term fied into fixed-platform,portable-platform,and mobile demographics of China,Japan,and the Scandinavian autonomous types.Fixed-platform robots perform func- countries show a growing shortage of working-age tions in the kitchen,on the desktop,or by the bed. adults.Age-related disabilities will soon dominate the Portable types are manipulator arms attached to an elec- service sector job market,put many older and disabled tric wheelchair to hold and move objects and to interact people at risk,and increase the need for institutional- with other devices and equipment,as in opening a door. ization when there is no viable home-based solution.Mobile autonomous robots can be controlled by voice or National programs to develop personal robots,robotic other means to carry out manipulation and other errands therapy,smart prostheses,smart beds,smart homes,in the home or workplace.Mobility aids are divided and tele-rehabilitation services have accelerated in the into electric wheelchairs with navigation systems and past ten years and will need to continue apace with the mobile robots that act as smart,motorized walkers,al- ever-increasing ability of health care to allow people to lowing people with mobility impairments to lean on live longer through the repression of disease and im-them to prevent falls and provide stability.The third main provements in surgical and medication interventions. type,cognitive aids,assist people who have dementia, Rehabilitation robotics,although only a 40-year-old dis- autism or other disorders that affect communication and cipline [53.1-3],is projected to grow quickly in the physical well-being. coming decades. The fields of prosthetics and FNS are closely al- 驾 lied with rehabilitation robotics.Prostheses are artificial 53.1.1 Taxonomy of Rehabilitation Robotics hands,arms,legs,and feet that are worn by the user to replace amputated limbs.Prostheses are increasingly The field of rehabilitation robotics is generally divided incorporating robotic features.FNS systems seek to into the categories of therapy and assistance robots.In reanimate the limb movements of weak or paralyzed addition,rehabilitation robotics includes aspects of arti- people by electrically stimulating nerve and muscle. ficial limb(prosthetics)development,functional neural FNS control systems are analogous to robotic control stimulation,(FNS)and technology for the diagnosis and systems,except that the actuators being controlled are monitoring of people during ADLs. human muscles.Another related field is technology for Therapy robots generally have at least two main monitoring and diagnosing health care issues as a person users simultaneously,the person with a disability who performs ADLs. is receiving the therapy and the therapist who sets up The chapter is organized according to this taxonomy. and monitors the interaction with the robot.Types of After providing background information on world de- therapy that have benefited from robotic assistance are mographics (Sect.53.1.2)and the history (Sect.53.1.3) upper-and lower-extremity movement therapy,enabling of the field,Sect.53.2 describes physical therapy and communication for children with autism,and enabling training robots,and Sect.53.3 describes robotic aids exploration (education)for children with cerebral palsy for people with disabilities.Section 53.4 then reviews (CP)or other developmental disabilities.A robot may be recent advances in smart prostheses and orthoses that a good alternative to a physical or occupational therapist are related to rehabilitation robotics.Finally,Sect.53.5 for the actual hands-on intervention for several reasons: provides an overview of recent work in diagnosis and (1)once properly set up,an automated exercise machine monitoring for rehabilitation as well as other health care can consistently apply therapy over long periods of time issues. without tiring;(2)the robot's sensors can measure the work performed by the patient and quantify,to an extent 53.1.2 World Demographics perhaps not yet measurable by clinical scales,any re- covery of function that may have occurred,which may The various areas of rehabilitation robotics focus on be highly motivating for a person to continue with the different user populations,but a common characteris- therapy;and(3)the robot may be able to engage the tic of these populations is that they have a disability. patient in types of therapy exercises that a therapist can- Disability is defined in the Americans with Disabilities not do,such as magnifying movement errors to provoke Act as "a physical or mental impairment that substan- adaptation [53.4,5]. tially limits one or more of the major life activities

1224 Part F Field and Service Robotics ties of daily living (ADLs), or to assist us in actually doing them if we are unable to relearn. While hu￾man therapists and attendants can indeed provide the types of assistance required, the projected short-term demographics of China, Japan, and the Scandinavian countries show a growing shortage of working-age adults. Age-related disabilities will soon dominate the service sector job market, put many older and disabled people at risk, and increase the need for institutional￾ization when there is no viable home-based solution. National programs to develop personal robots, robotic therapy, smart prostheses, smart beds, smart homes, and tele-rehabilitation services have accelerated in the past ten years and will need to continue apace with the ever-increasing ability of health care to allow people to live longer through the repression of disease and im￾provements in surgical and medication interventions. Rehabilitation robotics, although only a 40-year-old dis￾cipline [53.1–3], is projected to grow quickly in the coming decades. 53.1.1 Taxonomy of Rehabilitation Robotics The field of rehabilitation robotics is generally divided into the categories of therapy and assistance robots. In addition, rehabilitation robotics includes aspects of arti- ficial limb (prosthetics) development, functional neural stimulation, (FNS) and technology for the diagnosis and monitoring of people during ADLs. Therapy robots generally have at least two main users simultaneously, the person with a disability who is receiving the therapy and the therapist who sets up and monitors the interaction with the robot. Types of therapy that have benefited from robotic assistance are upper- and lower-extremity movement therapy, enabling communication for children with autism, and enabling exploration (education) for children with cerebral palsy (CP) or other developmental disabilities. A robot may be a good alternative to a physical or occupational therapist for the actual hands-on intervention for several reasons: (1) once properly set up, an automated exercise machine can consistently apply therapy over long periods of time without tiring; (2) the robot’s sensors can measure the work performed by the patient and quantify, to an extent perhaps not yet measurable by clinical scales, any re￾covery of function that may have occurred, which may be highly motivating for a person to continue with the therapy; and (3) the robot may be able to engage the patient in types of therapy exercises that a therapist can￾not do, such as magnifying movement errors to provoke adaptation [53.4, 5]. Assistive robots are generally grouped accord￾ing to whether they focus on manipulation, mobility, or cognition. Manipulation aids are further classi- fied into fixed-platform, portable-platform, and mobile autonomous types. Fixed-platform robots perform func￾tions in the kitchen, on the desktop, or by the bed. Portable types are manipulator arms attached to an elec￾tric wheelchair to hold and move objects and to interact with other devices and equipment, as in opening a door. Mobile autonomous robots can be controlled by voice or other means to carry out manipulation and other errands in the home or workplace. Mobility aids are divided into electric wheelchairs with navigation systems and mobile robots that act as smart, motorized walkers, al￾lowing people with mobility impairments to lean on them to prevent falls and provide stability. The third main type, cognitive aids, assist people who have dementia, autism or other disorders that affect communication and physical well-being. The fields of prosthetics and FNS are closely al￾lied with rehabilitation robotics. Prostheses are artificial hands, arms, legs, and feet that are worn by the user to replace amputated limbs. Prostheses are increasingly incorporating robotic features. FNS systems seek to reanimate the limb movements of weak or paralyzed people by electrically stimulating nerve and muscle. FNS control systems are analogous to robotic control systems, except that the actuators being controlled are human muscles. Another related field is technology for monitoring and diagnosing health care issues as a person performs ADLs. The chapter is organized according to this taxonomy. After providing background information on world de￾mographics (Sect. 53.1.2) and the history (Sect. 53.1.3) of the field, Sect. 53.2 describes physical therapy and training robots, and Sect. 53.3 describes robotic aids for people with disabilities. Section 53.4 then reviews recent advances in smart prostheses and orthoses that are related to rehabilitation robotics. Finally, Sect. 53.5 provides an overview of recent work in diagnosis and monitoring for rehabilitation as well as other health care issues. 53.1.2 World Demographics The various areas of rehabilitation robotics focus on different user populations, but a common characteris￾tic of these populations is that they have a disability. Disability is defined in the Americans with Disabilities Act as “a physical or mental impairment that substan￾tially limits one or more of the major life activities.” Part F 53.1

Rehabilitation and Health Care Robotics 53.1 Overview 1225 Table 53.1 Prevalence and incidence of disability and aging in selected countries [53.6] Country Number of people Percentage of population Number of elderly Percentage of population with disabilities with disabilities people that is elderly France 5146000 83 12151000 19.6 USA 52591000 20.0 35000000 12.4 Great Britain 4453000 7.3 12200000 29.5 The Netherlands 1432000 9.5 2118808 13.4 Spain 3528220 8.9 6936000 17.6 Japan 5136000 4.3 44982000 35.7 Korea 3195000 7.1 16300000 36.0 In the industrialized countries (e.g.,Japan,US,Canada, (early 1970s)(reviewed in [53.7])were both adaptations and Europe),the incidence of disability varies between of replacement mechanical arms meant as powered or- 8%and 20%,with differences likely due primarily to thoses [53.1].The user drove the Golden Arm with a set varying definitions of disability and reporting conven- of tongue-operated switches,joint-by-joint,an arduous tions (Table 53.1).Age is a risk factor for disability,and means of endpoint control.In the mid 1970s,the De- lower birth rates and life-extending health care are the partment of Veterans Affairs began funding a group at dominant factors contributing to the aging of the pop- the Applied Physics Lab under the guidance of Seamone ulation and a concomitant rise in disability.In China,and Schmeisser to computerize an orthosis mounted on the population control policies of the 1970s have created a workstation to do activities of daily living(ADL)tasks art alack of working-age adults to support theeconomy.The such as feeding a person and turning pages [53.9].For disproportionate incidence of disability in the elderly the first time,a rehabilitation robot had a command-type population makes it clear that developers of rehabili-interface,not just a joint-by-joint motion controller. tation robotics will also be faced with users who,as The 1970s also saw the French Spartacus system be- a demographic group,generally have lower levels of ing developed,guided by the vision of Jean Vertut,for sensory and motor capability,and may have impaired use by people with high-level spinal cord injury as well cognition as well.The urgency of making advances in as children with cerebral palsy [53.10].This system did this field is increasing in line with these demographic not emerge from the P&O field but was developed by changes. the French Atomic Energy Commission(CEA),which used large telemanipulators for nuclear fuel rod han- 53.1.3 Short History of the Field dling.One of these was adapted so that people with of Rehabilitation Robotics movement impairment could control it using a joystick for pick-and-place tasks.A decade later,one of the re- The history of rehabilitation robotics is almost as old as searchers on the Spartacus project,Hok Kwee,began the that of robotics itself,although emanating from very dif-MANUS project,a dedicated effort to develop the first ferent sources.Several books,chapters,and papers have wheelchair-mounted manipulator designed expressly as been written on the history of rehabilitation robotics in a rehabilitation robot,not adapted from a design from more detail than this section [53.1,7,8],and numerous another field. papers in the proceedings of the Institute of Electrical However,in between,several other major programs and Electronics Engineers (IEEE)International Con- were begun.In 1978,at Stanford University,and then ference on Rehabilitation Robotics also provide more with decades-long funding from the US Department of grounding for historical perspective.The chronology Veterans Affairs,Larry Leifer started the vocational as- below pays particular attention to early work and to sistant robot program,culminating in several clinically projects with notable clinical and/or commercial impact.tested versions of the desktop vocational assistant robot Early robotics,starting in the late 1950s,focused (DeVAR)[53.3,11,12],a mobile version,the mobile vo- on large manipulators to replace workers in factories cational assistant robot(MoVAR)[53.13],and finally the for dirty,dangerous,and undesirable tasks.The earliest professional vocational assistant robot(ProVAR),which rehabilitation robots came from the field of prosthet-had the advanced ability for the user to program tasks in ics and orthotics(P&O).The Case Western University an easy-to-use browser-type environment [53.14].This arm (1960s)and the Rancho Los Amigos Golden Arm step was made since,although DeVAR made it briefly

Rehabilitation and Health Care Robotics 53.1 Overview 1225 Table 53.1 Prevalence and incidence of disability and aging in selected countries [53.6] Country Number of people Percentage of population Number of elderly Percentage of population with disabilities with disabilities people that is elderly France 5 146 000 8.3 12 151 000 19.6 USA 52 591 000 20.0 35 000 000 12.4 Great Britain 4 453 000 7.3 12 200 000 29.5 The Netherlands 1 432 000 9.5 2 118 808 13.4 Spain 3 528 220 8.9 6 936 000 17.6 Japan 5 136 000 4.3 44 982 000 35.7 Korea 3 195 000 7.1 16 300 000 36.0 In the industrialized countries (e.g., Japan, US, Canada, and Europe), the incidence of disability varies between 8% and 20%, with differences likely due primarily to varying definitions of disability and reporting conven￾tions (Table 53.1). Age is a risk factor for disability, and lower birth rates and life-extending health care are the dominant factors contributing to the aging of the pop￾ulation and a concomitant rise in disability. In China, the population control policies of the 1970s have created a lack of working-age adults to support the economy. The disproportionate incidence of disability in the elderly population makes it clear that developers of rehabili￾tation robotics will also be faced with users who, as a demographic group, generally have lower levels of sensory and motor capability, and may have impaired cognition as well. The urgency of making advances in this field is increasing in line with these demographic changes. 53.1.3 Short History of the Field of Rehabilitation Robotics The history of rehabilitation robotics is almost as old as that of robotics itself, although emanating from very dif￾ferent sources. Several books, chapters, and papers have been written on the history of rehabilitation robotics in more detail than this section [53.1, 7, 8], and numerous papers in the proceedings of the Institute of Electrical and Electronics Engineers (IEEE) International Con￾ference on Rehabilitation Robotics also provide more grounding for historical perspective. The chronology below pays particular attention to early work and to projects with notable clinical and/or commercial impact. Early robotics, starting in the late 1950s, focused on large manipulators to replace workers in factories for dirty, dangerous, and undesirable tasks. The earliest rehabilitation robots came from the field of prosthet￾ics and orthotics (P&O). The Case Western University arm (1960s) and the Rancho Los Amigos Golden Arm (early 1970s) (reviewed in [53.7]) were both adaptations of replacement mechanical arms meant as powered or￾thoses [53.1]. The user drove the Golden Arm with a set of tongue-operated switches, joint-by-joint, an arduous means of endpoint control. In the mid 1970s, the De￾partment of Veterans Affairs began funding a group at the Applied Physics Lab under the guidance of Seamone and Schmeisser to computerize an orthosis mounted on a workstation to do activities of daily living (ADL) tasks such as feeding a person and turning pages [53.9]. For the first time, a rehabilitation robot had a command-type interface, not just a joint-by-joint motion controller. The 1970s also saw the French Spartacus system be￾ing developed, guided by the vision of Jean Vertut, for use by people with high-level spinal cord injury as well as children with cerebral palsy [53.10]. This system did not emerge from the P&O field but was developed by the French Atomic Energy Commission (CEA), which used large telemanipulators for nuclear fuel rod han￾dling. One of these was adapted so that people with movement impairment could control it using a joystick for pick-and-place tasks. A decade later, one of the re￾searchers on the Spartacus project, Hok Kwee, began the MANUS project, a dedicated effort to develop the first wheelchair-mounted manipulator designed expressly as a rehabilitation robot, not adapted from a design from another field. However, in between, several other major programs were begun. In 1978, at Stanford University, and then with decades-long funding from the US Department of Veterans Affairs, Larry Leifer started the vocational as￾sistant robot program, culminating in several clinically tested versions of the desktop vocational assistant robot (DeVAR) [53.3,11,12], a mobile version, the mobile vo￾cational assistant robot (MoVAR) [53.13], and finally the professional vocational assistant robot (ProVAR), which had the advanced ability for the user to program tasks in an easy-to-use browser-type environment [53.14]. This step was made since, although DeVAR made it briefly Part F 53.1

1226 Part F Field and Service Robotics onto the market in the early 1990s,multisite user test- Rich Mahoney,moved to ARC and was instrumental ing revealed it was still too costly for the functionality in extending the company's repertoire to the RAPTOR it had:ProVAR development ensued,then continued by wheelchair-mounted arm [53.18]. Machiel Van der Loos.All these versions were based In Europe,the most significant mobile manipula- on the Puma-260 industrial manipulator to achieve ro- tor project was the MANUS project [53.19]mentioned bust,safe operation.Research shifted in 2006 to the earlier.With much of the work done under the di- Veterans Affairs(VA)in Syracuse,NY,to integrate sens- rection of Hok Kwee at the Rehabilitation Research ing and autonomous features and explore new,more and Development Center (iRV)in the Netherlands,the cost-effective manipulator options. project culminated in a robot specifically designed for In the mid 1980s.from observations on the wheelchair mounting,with control through a joystick unsuitability of existing industrial,educational,and and feedback by a small display on the arm itself.This orthosis-derived manipulators for rehabilitation appli- project has led to numerous follow-on research projects, cations,Tim Jones at Universal Machine Intelligence and,most significantly,to the commercialization of the (later Oxford Intelligent Machines,OxIM)in the UK system by Exact Dynamics BV,in the Netherlands.It is began an intensive effort to provide the rehabilitation currently offered free on physician prescription by the robotics community with its first workhorse system spe- Dutch government to qualified people with a disability cially designed from the ground up for human service such as cerebral palsy or tetraplegia from a spinal cord tasks.Over ten years,a series of systems,starting with injury. the RTX model,were used in numerous research labs and Autonomous navigation systems on electric clinics around the world.The most extensive effort to wheelchairs also began in the 1980s,benefiting initially use the OxIM arm was in France,and a suite of research from the development by Polaroid Corporation of range projects,funded by the French government and the Eu- finders for its cameras using ultrasonic sensors.They ropean Research Commission,starting as the robot for were inexpensive,and small enough,at 30 mm in diam- assisting the integration of the disabled(RAID),then eter,that dozens of them could be placed around the as MASTER [53.15],developed and clinically tested periphery of a wheelchair to aid medium-range navi- workstation-based assistive systems based on the RTX gation 10-500cm).In the 1990s and early 2000s, and subsequent OxIM arms.When OxIM ceased build- with the advent of vision-based servoing and laser range ing its arms,the French company Afma Robotics [53.16] scanners,algorithms for faster,smarter,less-error-prone took over efforts to commercialize the MASTER system, navigation and obstacle avoidance dominated research which it continues to do today (2007). advances in this sector.In Korea,for example,Zenn Bien The UK was also the site of the first commercially at the Korea Advanced Institute for Science and Tech- available feeding robot,Handy-I,an inexpensive and nology(KAIST)Human Welfare Robotics Center began well-received device first designed by Mike Topping developing the KAIST rehabilitation engineering system and then commercialized by Rehabilitation Robotics. (KARES)line of wheelchair-based navigation systems Ltd.in the 1990s [53.17].Primarily aimed to enabled in the late 1990s [53.20]and the NavChair project at the people with cerebral palsy to achieve a measure of in- University of Michigan was the start of a development dependence in feeding themselves,task environments line that led to the commercialized Hephaestus system later also included face washing and the application of at the University of Pittsburgh [53.21,22]. cosmetics,areas of high demand identified by its users. Therapy robots had a later start than assistive robots, The history of mobile manipulator applications be- with early exercise devices such as the BioDex [53.23] gan in the 1980s with adaptations of educational and a first step in programmable,force-controlled,albeit industrial robots,and achieved a boost with the funding single-axis devices,in the mid 1980s.The first multi-axis of the US National Institute on Disability and Rehabilita- concept was published by Khalili and Zomlefer [53.24], tion Research(NIDRR)for a Rehabilitation Engineering and the first tested system by Robert Erlandson at Research Center on Rehabilitation Robotics (RERC)Wayne State University emerged in the mid 1980s at the Alfred I.duPont Hospital in Delaware from as well [53.25].The RTX manipulator had a touch- 1993-1997.With its ability to fund dozens of research sensitive pad as an end-effector,presenting targets at projects in parallel,it also formed a partnership with different locations for patients with upper-extremity a local company,Applied Resources,Corp.(ARC), weakness (e.g.,following a stroke)to hit after the which developed and marketed several rehabilitation screen gave a visual signal.Software logged response technology products.One of the RERC researchers, times,thereby providing a score that was tallied and

1226 Part F Field and Service Robotics onto the market in the early 1990s, multisite user test￾ing revealed it was still too costly for the functionality it had: ProVAR development ensued, then continued by Machiel Van der Loos. All these versions were based on the Puma-260 industrial manipulator to achieve ro￾bust, safe operation. Research shifted in 2006 to the Veterans Affairs (VA) in Syracuse, NY, to integrate sens￾ing and autonomous features and explore new, more cost-effective manipulator options. In the mid 1980s, from observations on the unsuitability of existing industrial, educational, and orthosis-derived manipulators for rehabilitation appli￾cations, Tim Jones at Universal Machine Intelligence (later Oxford Intelligent Machines, OxIM) in the UK began an intensive effort to provide the rehabilitation robotics community with its first workhorse system spe￾cially designed from the ground up for human service tasks. Over ten years, a series of systems, starting with the RTX model, were used in numerous research labs and clinics around the world. The most extensive effort to use the OxIM arm was in France, and a suite of research projects, funded by the French government and the Eu￾ropean Research Commission, starting as the robot for assisting the integration of the disabled (RAID), then as MASTER [53.15], developed and clinically tested workstation-based assistive systems based on the RTX and subsequent OxIM arms. When OxIM ceased build￾ing its arms, the French company Afma Robotics [53.16] took over efforts to commercialize the MASTER system, which it continues to do today (2007). The UK was also the site of the first commercially available feeding robot, Handy-I, an inexpensive and well-received device first designed by Mike Topping and then commercialized by Rehabilitation Robotics, Ltd. in the 1990s [53.17]. Primarily aimed to enabled people with cerebral palsy to achieve a measure of in￾dependence in feeding themselves, task environments later also included face washing and the application of cosmetics, areas of high demand identified by its users. The history of mobile manipulator applications be￾gan in the 1980s with adaptations of educational and industrial robots, and achieved a boost with the funding of the US National Institute on Disability and Rehabilita￾tion Research (NIDRR) for a Rehabilitation Engineering Research Center on Rehabilitation Robotics (RERC) at the Alfred I. duPont Hospital in Delaware from 1993–1997. With its ability to fund dozens of research projects in parallel, it also formed a partnership with a local company, Applied Resources, Corp. (ARC), which developed and marketed several rehabilitation technology products. One of the RERC researchers, Rich Mahoney, moved to ARC and was instrumental in extending the company’s repertoire to the RAPTOR wheelchair-mounted arm [53.18]. In Europe, the most significant mobile manipula￾tor project was the MANUS project [53.19] mentioned earlier. With much of the work done under the di￾rection of Hok Kwee at the Rehabilitation Research and Development Center (iRV) in the Netherlands, the project culminated in a robot specifically designed for wheelchair mounting, with control through a joystick and feedback by a small display on the arm itself. This project has led to numerous follow-on research projects, and, most significantly, to the commercialization of the system by Exact Dynamics BV, in the Netherlands. It is currently offered free on physician prescription by the Dutch government to qualified people with a disability such as cerebral palsy or tetraplegia from a spinal cord injury. Autonomous navigation systems on electric wheelchairs also began in the 1980s, benefiting initially from the development by Polaroid Corporation of range finders for its cameras using ultrasonic sensors. They were inexpensive, and small enough, at 30 mm in diam￾eter, that dozens of them could be placed around the periphery of a wheelchair to aid medium-range navi￾gation (≈ 10–500 cm). In the 1990s and early 2000s, with the advent of vision-based servoing and laser range scanners, algorithms for faster, smarter, less-error-prone navigation and obstacle avoidance dominated research advances in this sector. In Korea, for example, Zenn Bien at the Korea Advanced Institute for Science and Tech￾nology (KAIST) Human Welfare Robotics Center began developing the KAIST rehabilitation engineering system (KARES) line of wheelchair-based navigation systems in the late 1990s [53.20] and the NavChair project at the University of Michigan was the start of a development line that led to the commercialized Hephaestus system at the University of Pittsburgh [53.21, 22]. Therapy robots had a later start than assistive robots, with early exercise devices such as the BioDex [53.23] a first step in programmable, force-controlled, albeit single-axis devices, in the mid 1980s. The first multi-axis concept was published by Khalili and Zomlefer [53.24], and the first tested system by Robert Erlandson at Wayne State University emerged in the mid 1980s as well [53.25]. The RTX manipulator had a touch￾sensitive pad as an end-effector, presenting targets at different locations for patients with upper-extremity weakness (e.g., following a stroke) to hit after the screen gave a visual signal. Software logged response times, thereby providing a score that was tallied and Part F 53.1

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1227 compared to previous sessions.Later robots used ad- several demonstration systems were developed.In the vanced force-based control,which required significantly early 2000s,Corinna Latham of Anthrotronix,Inc.com- more computer power.The early 1990s saw the start mercialized small robot systems to enable children with of the MIT-MANUS Project with Neville Hogan and physical disabilities to play games with simple inter- Igo Krebs,followed a few years later by the Palo Alto faces.Later.small mobile robots were used in clinics VA mirror image movement enabler(MIME)project by Kerstin Dautenhahn's group [53.26]with children and its derivative,Driver's simulation environment for who have autism;since robots have such simple inter- arm therapy (SEAT),with Charles Burgar,Machiel Van faces,communication with them does not appear not der Loos,and Peter Lum,as well as the Rehabilita-be as challenging as with other humans.The early tion Institute of Chicago ARM project with Zev Rymer 2000s also saw the advent of pet robots,such as the and David Reinkensmeyer.Each had a different phi- Paro seal robot developed by Shibata et al.[53.27],as losophy on upper-extremity stroke therapy and each companions for both children and the elderly who are was able to demonstrate clinical effectiveness in a dif- confined to clinics and have limited real companion- ferent way.All three programs,now a decade later,ship. have made significant technical advances and are still The applications for robotics continue to increase active. in number as advances in materials.control software, Cognitive robotics had a start in the early 1980s to higher robustness and the diminishing size of sensors aid children with communication disorders and physi-and actuators allow designers to attempt new ways of cal disorders to achieve a measure of control of their using mechatronics technology to further the well-being physical space.Using mostly educational manipulators, of people with disabilities. 53.2 Physical Therapy and Training Robots 53.2.1 Grand Challenges and Roadblocks nize beginning in the late 1980s,neuro-rehabilitation is a logical target for automation because of its labor- The human neuromuscular system exhibits use-intensive,mechanical nature,and because the amount of dependent plasticity,which is to say that use alters the recovery is linked with the amount of repetition.Robots properties of neurons and muscles,including the pattern could deliver at least the repetitive parts of movement of their connectivity,and thus their function [53.28-30]. therapy at lower cost than human therapists,allowing The process of neuro-rehabilitation seeks to exploit this patients to receive more therapy. use-dependent plasticity in order to help people re- The grand challenge for automating movement ther- learn how to move following neuromuscular injuries or apy is determining how to optimize use-dependent diseases.Neuro-rehabilitation is typically provided by plasticity.That is,researchers in this field must de- skilled therapists,including physical,occupational and termine what the robot should do in cooperation with speech therapists.This process is time-consuming,in-the patient's own movement attempts in order to maxi- volving daily,intensive movement practice over many mally improve movement ability.Meeting this challenge weeks.It is also labor-intensive,requiring hands-on involves solving two key problems:determining appro- assistance from therapists.For some tasks,such as teach- priate movement tasks(what movements should patients ing a person with poor balance and weak legs to walk, practise and what feedback should they receive about this hands-on assistance requires that the therapist have their performance),and determining an appropriate pat- substantial strength and agility. tern of mechanical input to the patient during these Because neuro-rehabilitation is time-and labor-movement tasks(what forces should the robot apply to intensive,in recent years health care payers have put the patient's limb to provoke plasticity).The prescription limits on the amount of therapy that they will pay for,in of movement tasks and mechanical input fundamen- an effort to contain spiraling health care costs.Ironically,tally constrains the mechanical and control design of at the same time,there has been increasing scientific ev- the robotic therapy device. idence that more therapy can in some cases increase There are two main roadblocks to achieving the movement recovery via use-dependent plasticity.As grand challenge.The first is a scientific roadblock: robotics and rehabilitation researchers began to recog- neither the optimal movement tasks nor the optimal

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1227 compared to previous sessions. Later robots used ad￾vanced force-based control, which required significantly more computer power. The early 1990s saw the start of the MIT-MANUS Project with Neville Hogan and Igo Krebs, followed a few years later by the Palo Alto VA mirror image movement enabler (MIME) project and its derivative, Driver’s simulation environment for arm therapy (SEAT), with Charles Burgar, Machiel Van der Loos, and Peter Lum, as well as the Rehabilita￾tion Institute of Chicago ARM project with Zev Rymer and David Reinkensmeyer. Each had a different phi￾losophy on upper-extremity stroke therapy and each was able to demonstrate clinical effectiveness in a dif￾ferent way. All three programs, now a decade later, have made significant technical advances and are still active. Cognitive robotics had a start in the early 1980s to aid children with communication disorders and physi￾cal disorders to achieve a measure of control of their physical space. Using mostly educational manipulators, several demonstration systems were developed. In the early 2000s, Corinna Latham of Anthrotronix, Inc. com￾mercialized small robot systems to enable children with physical disabilities to play games with simple inter￾faces. Later, small mobile robots were used in clinics by Kerstin Dautenhahn’s group [53.26] with children who have autism; since robots have such simple inter￾faces, communication with them does not appear not be as challenging as with other humans. The early 2000s also saw the advent of pet robots, such as the Paro seal robot developed by Shibata et al. [53.27], as companions for both children and the elderly who are confined to clinics and have limited real companion￾ship. The applications for robotics continue to increase in number as advances in materials, control software, higher robustness and the diminishing size of sensors and actuators allow designers to attempt new ways of using mechatronics technology to further the well-being of people with disabilities. 53.2 Physical Therapy and Training Robots 53.2.1 Grand Challenges and Roadblocks The human neuromuscular system exhibits use￾dependent plasticity, which is to say that use alters the properties of neurons and muscles, including the pattern of their connectivity, and thus their function [53.28–30]. The process of neuro-rehabilitation seeks to exploit this use-dependent plasticity in order to help people re￾learn how to move following neuromuscular injuries or diseases. Neuro-rehabilitation is typically provided by skilled therapists, including physical, occupational and speech therapists. This process is time-consuming, in￾volving daily, intensive movement practice over many weeks. It is also labor-intensive, requiring hands-on assistance from therapists. For some tasks, such as teach￾ing a person with poor balance and weak legs to walk, this hands-on assistance requires that the therapist have substantial strength and agility. Because neuro-rehabilitation is time- and labor￾intensive, in recent years health care payers have put limits on the amount of therapy that they will pay for, in an effort to contain spiraling health care costs. Ironically, at the same time, there has been increasing scientific ev￾idence that more therapy can in some cases increase movement recovery via use-dependent plasticity. As robotics and rehabilitation researchers began to recog￾nize beginning in the late 1980s, neuro-rehabilitation is a logical target for automation because of its labor￾intensive, mechanical nature, and because the amount of recovery is linked with the amount of repetition. Robots could deliver at least the repetitive parts of movement therapy at lower cost than human therapists, allowing patients to receive more therapy. The grand challenge for automating movement ther￾apy is determining how to optimize use-dependent plasticity. That is, researchers in this field must de￾termine what the robot should do in cooperation with the patient’s own movement attempts in order to maxi￾mally improve movement ability. Meeting this challenge involves solving two key problems: determining appro￾priate movement tasks (what movements should patients practise and what feedback should they receive about their performance), and determining an appropriate pat￾tern of mechanical input to the patient during these movement tasks (what forces should the robot apply to the patient’s limb to provoke plasticity). The prescription of movement tasks and mechanical input fundamen￾tally constrains the mechanical and control design of the robotic therapy device. There are two main roadblocks to achieving the grand challenge. The first is a scientific roadblock: neither the optimal movement tasks nor the optimal Part F 53.2

1228 Part F Field and Service Robotics mechanical inputs are known.The scientific basis for 53.2.2 Movement Therapy neuro-rehabilitation remains ill-defined,with competing after Neurologic Injury schools of thought.The number of large,randomized, controlled trials that have rigorously compared different At present,much of the activity in physical therapy and therapy techniques is still small,in part because these tri- training robots has been focused on retraining movement als are expensive and difficult to control well.Therefore, ability for individuals who have had a stroke or spinal the first problem that a robotics engineer will encounter cord injury(SCI).The main reasons for this emphasis are when setting out to build a robotic therapy device is that that there are a relatively large number of patients with there is still substantial uncertainty as to what exactly these conditions.the rehabilitation costs associated with the device should do. them are high,and because these patients can sometimes This uncertainty corresponds to an opportunity to use experience large gains with intensive rehabilitation be- robotic therapy devices as scientific tools themselves. cause of use-dependent plasticity.Some systems have Robotic therapy devices have the potential to help iden- also been targeted at assisting in cognitive rehabilitation tify what exactly provokes plasticity during movement of persons with neurologic injury,as reviewed below. rehabilitation,because they can provide well-controlled A stroke refers to an obstruction or breakage of patterns of therapy.They can also simultaneously mea-a blood vessel supplying oxygen and nutrients to the sure the results of that therapy.Better control over brain.Approximately 800000 people suffer a stroke therapy delivery and improved quantitative assessment each year in the USA alone,and about 80%of these peo- of patient improvement are two desirable features for ple experience acute movement deficits [53.32].There clinical trials that have often been lacking in the past.are over 3000000 survivors of stroke in the USA, 驾 Recent work with robotic movement training devices with over half of these individuals experiencing per- is leading,for example,to the characterization of com-sistent,disabling,movement impairments.The number 53.2 putations that underlie motor adaptation,and then to of people who have experienced and survived a stroke is strategies for enhancing adaptation based on optimiza- expected to increase substantially in the USA and other tion approaches [53.5,31]. industrialized countries in the next two decades.because The second roadblock is a technological one:robotic age is a risk factor for stroke and the mean age of peo- therapy devices often have as their goal to assist in ther- ple in industrialized countries is rapidly increasing due apy of many body degrees of freedom (e.g.,the arms and to the baby boom of the 1950s. torso for reaching,or the pelvis and legs for walking). Common motor impairments that result from stroke The devices also require a wide dynamic bandwidth such are hemiparesis,which refers to weakness on one side of that they can,for example,impose a desired movement the body:abnormal tone,which refers to an increase in on a patient who is paralyzed,but also fade-to-nothing the felt resistance to passive movement a limb;impaired as the patient recovers.Furthermore,making the devices coordination,which can manifest itself as an appar- light enough to be wearable is desirable,so that the pa- ent loss in control degrees of freedom and decreased tient can participate in rehabilitation in a natural setting smoothness of movement;and impaired somatosensa- (for example,by walking over ground or working at tion,which refers to a decreased ability to sense the a counter in a kitchen),or even throughout the course movement of body parts.Secondary impairments in- of normal activities of daily living.The development clude muscle atrophy and disuse-related shortening and of high-degree-of-freedom,wearable,high-bandwidth stiffening of soft tissue,resulting in decreased passive robotic exoskeletons is an unsolved problem in robotics. range of motion of joints.Often the ability to open the No device at present comes close to matching the flex- hand,and to a slightly lesser extent close the hand,is ibility of a human therapist,in terms of assisting in dramatically decreased. moving different body degrees of freedom in a vari- The number of people who experience a SCI in the ety of settings (e.g.,walking,reaching,grasping,neck USA each year is relatively smaller-about 15 000,with movement),or the intelligence of a human therapist,in about 200000 people alive who have survived a SCI terms of providing different forms of mechanical input -but the consequences can be even more costly than (e.g.,stretching,assisting,resisting,perturbing)based stroke [53.32].The most common causes of SCI are au- on a real-time assessment of the patient's response.tomobile accidents and falls.These accidents crush the Meeting the grand challenge of robotic therapy there- spinal column and contuse the spinal cord,damaging or fore will require substantial,interrelated advances in destroying neurons within the spinal cord.The resulting both clinical neuroscience and robot engineering. pattern of movement impairment depends strongly on

1228 Part F Field and Service Robotics mechanical inputs are known. The scientific basis for neuro-rehabilitation remains ill-defined, with competing schools of thought. The number of large, randomized, controlled trials that have rigorously compared different therapy techniques is still small, in part because these tri￾als are expensive and difficult to control well. Therefore, the first problem that a robotics engineer will encounter when setting out to build a robotic therapy device is that there is still substantial uncertainty as to what exactly the device should do. This uncertainty corresponds to an opportunity to use robotic therapy devices as scientific tools themselves. Robotic therapy devices have the potential to help iden￾tify what exactly provokes plasticity during movement rehabilitation, because they can provide well-controlled patterns of therapy. They can also simultaneously mea￾sure the results of that therapy. Better control over therapy delivery and improved quantitative assessment of patient improvement are two desirable features for clinical trials that have often been lacking in the past. Recent work with robotic movement training devices is leading, for example, to the characterization of com￾putations that underlie motor adaptation, and then to strategies for enhancing adaptation based on optimiza￾tion approaches [53.5, 31]. The second roadblock is a technological one: robotic therapy devices often have as their goal to assist in ther￾apy of many body degrees of freedom (e.g., the arms and torso for reaching, or the pelvis and legs for walking). The devices also require a wide dynamic bandwidth such that they can, for example, impose a desired movement on a patient who is paralyzed, but also fade-to-nothing as the patient recovers. Furthermore, making the devices light enough to be wearable is desirable, so that the pa￾tient can participate in rehabilitation in a natural setting (for example, by walking over ground or working at a counter in a kitchen), or even throughout the course of normal activities of daily living. The development of high-degree-of-freedom, wearable, high-bandwidth robotic exoskeletons is an unsolved problem in robotics. No device at present comes close to matching the flex￾ibility of a human therapist, in terms of assisting in moving different body degrees of freedom in a vari￾ety of settings (e.g., walking, reaching, grasping, neck movement), or the intelligence of a human therapist, in terms of providing different forms of mechanical input (e.g., stretching, assisting, resisting, perturbing) based on a real-time assessment of the patient’s response. Meeting the grand challenge of robotic therapy there￾fore will require substantial, interrelated advances in both clinical neuroscience and robot engineering. 53.2.2 Movement Therapy after Neurologic Injury At present, much of the activity in physical therapy and training robots has been focused on retraining movement ability for individuals who have had a stroke or spinal cord injury (SCI). The main reasons for this emphasis are that there are a relatively large number of patients with these conditions, the rehabilitation costs associated with them are high, and because these patients can sometimes experience large gains with intensive rehabilitation be￾cause of use-dependent plasticity. Some systems have also been targeted at assisting in cognitive rehabilitation of persons with neurologic injury, as reviewed below. A stroke refers to an obstruction or breakage of a blood vessel supplying oxygen and nutrients to the brain. Approximately 800 000 people suffer a stroke each year in the USA alone, and about 80% of these peo￾ple experience acute movement deficits [53.32]. There are over 3 000 000 survivors of stroke in the USA, with over half of these individuals experiencing per￾sistent, disabling, movement impairments. The number of people who have experienced and survived a stroke is expected to increase substantially in the USA and other industrialized countries in the next two decades, because age is a risk factor for stroke and the mean age of peo￾ple in industrialized countries is rapidly increasing due to the baby boom of the 1950s. Common motor impairments that result from stroke are hemiparesis, which refers to weakness on one side of the body; abnormal tone, which refers to an increase in the felt resistance to passive movement a limb; impaired coordination, which can manifest itself as an appar￾ent loss in control degrees of freedom and decreased smoothness of movement; and impaired somatosensa￾tion, which refers to a decreased ability to sense the movement of body parts. Secondary impairments in￾clude muscle atrophy and disuse-related shortening and stiffening of soft tissue, resulting in decreased passive range of motion of joints. Often the ability to open the hand, and to a slightly lesser extent close the hand, is dramatically decreased. The number of people who experience a SCI in the USA each year is relatively smaller – about 15 000, with about 200 000 people alive who have survived a SCI – but the consequences can be even more costly than stroke [53.32]. The most common causes of SCI are au￾tomobile accidents and falls. These accidents crush the spinal column and contuse the spinal cord, damaging or destroying neurons within the spinal cord. The resulting pattern of movement impairment depends strongly on Part F 53.2

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1229 the vertebrae at which the spinal cord is injured,since MIT-MANUS nerves branch out to the head,arms,legs,and bladder The first robotic therapy device to undergo extensive and bowel at different vertebrae.About 50%of spinal clinical testing,and now to achieve some commercial cord injuries are incomplete,meaning that some sensa- success,is the MIT-MANUS,sold as the InMotion2 tion or motor function is preserved below the level of the by Interactive Motion,Inc.[53.33].MIT-MANUS is injury.Spinal cord injuries are commonly bilateral and a planar two-joint arm that makes use of the selective thus are often more functionally devastating in compar-compliant assembly robot arm (SCARA)configura- ison to strokes,which at least leave a person with one tion,allowing two large,mechanically grounded motors side of their body that is relatively normal(which we to drive a lightweight linkage.The patient sits across will refer to as the less impaired side).Individuals ex- from the device,with the weaker hand attached to the perience especially severe disability if the lesion is high end-effecter,and the arm supported on a table with a low- enough to involve the arms as well as the legs. friction support.By virtue of the use of the SCARA configuration,the MIT-MANUS is perhaps the simplest 53.2.3 Robotic Therapy possible mechanical design that allows planar move- for the Upper Extremity ments while also allowing a large range of forces to be applied to the arm without requiring force feedback The following sections describe the best-known clini- control. cally tested upper-limb therapy robot systems that have MIT-MANUS assists the patient in moving the arm been developed since the 1980s(Fig.53.1). across the tabletop as the patient plays simple video Part F 53.2 Fig.53.1a-e Arm-therapy robotic systems that have undergone extensive clinical testing;(a)MIT-MANUS,developed by Hogan,Krebs,and colleagues at the Massachusetts Institute of Technology (USA);(b)MIME,developed at the Department of Veterans Affairs in Palo Alto in collaboration with Stanford University (USA);(c)GENTLE/s developed in the EU,(d)ARM-Guide,developed at the Rehabilitation Institute of Chicago and the University of California,Irvine (USA),and (e)Bi-Manu-Track,developed by Reha-Stim(Germany)

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1229 the vertebrae at which the spinal cord is injured, since nerves branch out to the head, arms, legs, and bladder and bowel at different vertebrae. About 50% of spinal cord injuries are incomplete, meaning that some sensa￾tion or motor function is preserved below the level of the injury. Spinal cord injuries are commonly bilateral and thus are often more functionally devastating in compar￾ison to strokes, which at least leave a person with one side of their body that is relatively normal (which we will refer to as the less impaired side). Individuals ex￾perience especially severe disability if the lesion is high enough to involve the arms as well as the legs. 53.2.3 Robotic Therapy for the Upper Extremity The following sections describe the best-known clini￾cally tested upper-limb therapy robot systems that have been developed since the 1980s (Fig. 53.1). a) b) c) d) e) Fig. 53.1a–e Arm-therapy robotic systems that have undergone extensive clinical testing; (a) MIT-MANUS, developed by Hogan, Krebs, and colleagues at the Massachusetts Institute of Technology (USA); (b) MIME, developed at the Department of Veterans Affairs in Palo Alto in collaboration with Stanford University (USA); (c) GENTLE/s developed in the EU, (d) ARM-Guide, developed at the Rehabilitation Institute of Chicago and the University of California, Irvine (USA), and (e) Bi-Manu-Track, developed by Reha-Stim (Germany) MIT-MANUS The first robotic therapy device to undergo extensive clinical testing, and now to achieve some commercial success, is the MIT-MANUS, sold as the InMotion2 by Interactive Motion, Inc. [53.33]. MIT-MANUS is a planar two-joint arm that makes use of the selective compliant assembly robot arm (SCARA) configura￾tion, allowing two large, mechanically grounded motors to drive a lightweight linkage. The patient sits across from the device, with the weaker hand attached to the end-effecter, and the arm supported on a table with a low￾friction support. By virtue of the use of the SCARA configuration, the MIT-MANUS is perhaps the simplest possible mechanical design that allows planar move￾ments while also allowing a large range of forces to be applied to the arm without requiring force feedback control. MIT-MANUS assists the patient in moving the arm across the tabletop as the patient plays simple video Part F 53.2

1230 Part F Field and Service Robotics games,such as moving a cursor into a target that finger and wrist muscles [53.41].Again,significant ben- changes locations on a computer screen.Assistance is efits were found for both therapies,and those benefits achieved using a position controller with an adjustable were specific to the movements practised,but the ben- impedance.Additional modules have been developed for efits were not significantly different between therapies. the device for allowing vertical motion [53.34],wrist We note that the lack of a significant difference in these motion [53.35],and hand grasp [53.36].Software has studies may simply be due to the limited number of been developed for providing graded resistance as well patients who participated in these studies (i.e.,inade- as assistance to movement [53.37],and for varying the quate study power),rather than a close similarity of the firmness and timing of assistance based on real-time effectiveness of the therapies. measurements of the patient's performance on the video games [53.38]. MIME MIT-MANUS has undergone extensive clinical test- Several other systems have undergone clinical testing. ing in several studies,summarized as follows.The first The mirror image movement enhancer(MIME)system clinical test of the device compared the motor recovery uses a Puma-560 robot arm to assist in movement of of acute stroke patients who received an additional dose the patient's arm [53.42].The device is attached to the of robot therapy on top of their conventional therapy,to hand through a customized splint and a connector that that of a control group,who received conventional ther- is designed to break away if interaction forces become apy and a brief,sham exposure to the robot [53.39].The too large.Compared to MIT-MANUS,the device al- robot group patients received the additional robotic ther-lows more naturalistic motion of the arm because of apy for an hour each day,five days per week,for several its six degrees of freedom(DOFs),but must rely on 驾 weeks.The robot group recovered more arm move- force feedback so that the patient can drive the robot ment ability than the control group according to clinical arm.Four control modes were developed for MIME. 53.2 scales,without any increase in adverse effects such as In the passive mode,the patient relaxes and the robot shoulder pain.The improvements might subjectively be moves the arm through a desired pattern.In the active characterized as small but somewhat meaningful to the assist mode,the patient initiates a reach toward a tar- patient.The improvements were sustained at three-year get,indicated by physical cones on a table top,which follow-up. then triggers a smooth movement of the robot toward This first study with MIT-MANUS demonstrated the target.In the active constrained mode,the device that acute stroke patients who received more therapy acts as a sort of virtual ratchet,allowing movement to- recover better,and that this extra therapy can be deliv- ward the target,but preventing the patient from moving ered by a robotic device.It did not answer the question away from the target.Finally,in mirror-image mode, as to whether the robotic features of the robotic device the motion of the patient's less impaired arm is meas- were necessary.In other words,it may have been that pa- ured with a digitizing linkage,and the impaired arm is tients would have also improved their movement ability controlled to follow along in a mirror-symmetric path. if they had practised additional movements with MIT- The initial clinical test of MIME found that chronic MANUS with the motors off(thus making it equivalent stroke patients who received therapy with the device to a computer mouse),simply by virtue of the increased improved their movement ability about as much as pa- dose of movement practice stimulating use-dependent tients who received conventional tabletop exercises with plasticity.Thus,while this study indicated the promise an occupational therapist [53.42].The robot group even of robots for rehabilitation therapy,it did not close the surpassed the gains from human-delivered therapy for gap of knowledge as to how external mechanical forces the outcome measures of reaching range of motion and provoke use-dependent plasticity. strength at key joints of the arm.A follow-on study Subsequent studies with MIT-MANUS confirmed is being undertaken to elucidate which of the control that robotic therapy can also benefit chronic stroke pa-modes or what combination of MIME exercises caused tients [53.40].The device has been used to compare the gains [53.43]. two different types of therapy-assisting movement ver- sus resisting movement-in chronic stroke subjects,but ARM Guide with inconclusive results:both types of therapy pro- The question of the effect of robot forces on move- duced benefits [53.37].The device has also been used to ment recovery was also left unresolved by a study with compare assistive robot therapy with another technolog- another device,the ARM guide.The ARM guide is ical approach to rehabilitation-electrical stimulation of a trombone-like device that can be oriented then locked

1230 Part F Field and Service Robotics games, such as moving a cursor into a target that changes locations on a computer screen. Assistance is achieved using a position controller with an adjustable impedance. Additional modules have been developed for the device for allowing vertical motion [53.34], wrist motion [53.35], and hand grasp [53.36]. Software has been developed for providing graded resistance as well as assistance to movement [53.37], and for varying the firmness and timing of assistance based on real-time measurements of the patient’s performance on the video games [53.38]. MIT-MANUS has undergone extensive clinical test￾ing in several studies, summarized as follows. The first clinical test of the device compared the motor recovery of acute stroke patients who received an additional dose of robot therapy on top of their conventional therapy, to that of a control group, who received conventional ther￾apy and a brief, sham exposure to the robot [53.39]. The robot group patients received the additional robotic ther￾apy for an hour each day, five days per week, for several weeks. The robot group recovered more arm move￾ment ability than the control group according to clinical scales, without any increase in adverse effects such as shoulder pain. The improvements might subjectively be characterized as small but somewhat meaningful to the patient. The improvements were sustained at three-year follow-up. This first study with MIT-MANUS demonstrated that acute stroke patients who received more therapy recover better, and that this extra therapy can be deliv￾ered by a robotic device. It did not answer the question as to whether the robotic features of the robotic device were necessary. In other words, it may have been that pa￾tients would have also improved their movement ability if they had practised additional movements with MIT￾MANUS with the motors off (thus making it equivalent to a computer mouse), simply by virtue of the increased dose of movement practice stimulating use-dependent plasticity. Thus, while this study indicated the promise of robots for rehabilitation therapy, it did not close the gap of knowledge as to how external mechanical forces provoke use-dependent plasticity. Subsequent studies with MIT-MANUS confirmed that robotic therapy can also benefit chronic stroke pa￾tients [53.40]. The device has been used to compare two different types of therapy – assisting movement ver￾sus resisting movement – in chronic stroke subjects, but with inconclusive results: both types of therapy pro￾duced benefits [53.37]. The device has also been used to compare assistive robot therapy with another technolog￾ical approach to rehabilitation – electrical stimulation of finger and wrist muscles [53.41]. Again, significant ben￾efits were found for both therapies, and those benefits were specific to the movements practised, but the ben￾efits were not significantly different between therapies. We note that the lack of a significant difference in these studies may simply be due to the limited number of patients who participated in these studies (i. e., inade￾quate study power), rather than a close similarity of the effectiveness of the therapies. MIME Several other systems have undergone clinical testing. The mirror image movement enhancer (MIME) system uses a Puma-560 robot arm to assist in movement of the patient’s arm [53.42]. The device is attached to the hand through a customized splint and a connector that is designed to break away if interaction forces become too large. Compared to MIT-MANUS, the device al￾lows more naturalistic motion of the arm because of its six degrees of freedom (DOFs), but must rely on force feedback so that the patient can drive the robot arm. Four control modes were developed for MIME. In the passive mode, the patient relaxes and the robot moves the arm through a desired pattern. In the active assist mode, the patient initiates a reach toward a tar￾get, indicated by physical cones on a table top, which then triggers a smooth movement of the robot toward the target. In the active constrained mode, the device acts as a sort of virtual ratchet, allowing movement to￾ward the target, but preventing the patient from moving away from the target. Finally, in mirror-image mode, the motion of the patient’s less impaired arm is meas￾ured with a digitizing linkage, and the impaired arm is controlled to follow along in a mirror-symmetric path. The initial clinical test of MIME found that chronic stroke patients who received therapy with the device improved their movement ability about as much as pa￾tients who received conventional tabletop exercises with an occupational therapist [53.42]. The robot group even surpassed the gains from human-delivered therapy for the outcome measures of reaching range of motion and strength at key joints of the arm. A follow-on study is being undertaken to elucidate which of the control modes or what combination of MIME exercises caused the gains [53.43]. ARM Guide The question of the effect of robot forces on move￾ment recovery was also left unresolved by a study with another device, the ARM guide. The ARM guide is a trombone-like device that can be oriented then locked Part F 53.2

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1231 in different directions and assist in reaching in a straight cylinders to help extend or flex the fingers,and has been line.Chronic stroke patients who received assistance shown to improve hand movement ability of chronic during reaching with the robot improved their move- stroke subjects [53.50].Simple force-feedback con- ment ability [53.44].However,they improved about as trolled devices,including a one-DOF wrist manipulator much as a control group that simply practised a matched and a two-DOF elbow-shoulder manipulator,were also number of reaches without assistance from the robot. recently shown to improve movement ability of chronic This suggests that movement effort by the patient is stroke subjects who exercised with the devices [53.51]. a key factor for recovery,although the small sample size A passive exoskeleton,the T-WREX arm orthosis,pro- of this study limited its ability to resolve the size of the vides support to the arm against gravity using elastic difference between guided and unguided therapy. bands,while still allowing a large range of motion of the arm [53.52].By incorporating a simple hand-grasp Bi-Manu-Track sensor,this device allows substantially weakened pa- Perhaps the most striking clinical results generated so tients to practise simple virtual reality exercises that far have come from one of the simplest devices built.simulate functional tasks such as shopping and cooking. Similar to a design proposed previously [53.45],the Chronic stroke patients who practised exercising with Bi-Manu-Track uses two motors,one for each hand,to this non-robotic device recovered significant amounts allow bimanual wrist-flexion extension [53.46].The de-of movement ability,comparable with the Fugl-Meyer vice can also assist in forearm pronation/supination if it gains seen with MIT-MANUS and MIME.NeReBot is tilted downward and the handles are changed.In an is a three-DOF wire-based robot that can slowly move extensive clinical test of the device,22 subacute patients a stroke patient's arm in spatial paths.Acute stroke (i.e.,4-6 weeks after stroke)practised 800 movements patients who received additional movement therapy art with the device for 20 min per day,five days per week beyond their conventional rehabilitation therapy with for six weeks [53.461.For half of the movements,the de-NeReBot recovered significantly more movement ability vice drove both arms,and for the other half,the patient's than patients who received just conventional rehabilita- stronger arm drove the motion of the more-impaired arm.tion therapy [53.53].RehaRob uses an industrial robot A control group received a matched duration of electri- arm to mobilize patients'arms along arbitrary trajecto- cal stimulation(ES)of their wrist extensor muscles,with ries following stroke [53.54]. the stimulation triggered by voluntary activation of their muscles when possible,as measured by electromyo- Other Systems Currently under Development graphy (EMG).The number of movements performed Several other robotic therapy devices are currently under with EMG-triggered ES was 60-80 per session.The development.For example,at the high end of cost and robot-trained group improved by 15 points more on the complexity are the ARM-In [53.55]and Pneu-WREX Fugl-Meyer scale,a standard clinical scale of movement systems [53.56],which are exoskeletons that accommo- ability with a range from 0 to 66 points in upper extrem- date nearly naturalistic movement of the arm while still ity function.It assigns a score of 0(cannot complete),I achieving a wide range of force control.A system that (completes partially),or 2(completes normally)for 33 couples a immersive virtual-reality display with a haptic test movements,such as lifting the arm without flexing robot arm is described in [53.571.A wearable exoskele- the elbow.For comparison,reported gains in Fugl-Meyer ton driven by pneumatic muscles is described in [53.58]. score after therapy with the MIT-MANUS and MIME At the lower end of cost/complexity are devices that use devices ranged from 0-5 points [53.47]. force feedback joysticks and steering wheels with a view toward implementation in the home [53.59-62].Exam- Other Devices to Undergo Clinical Testing ples of recent,novel robotic devices for the hand are Other devices to undergo clinical testing are as fol-given in [53.63-65]:these devices typically follow an lows.The GENTLE/s system uses a commercial robot,active assist therapy paradigm in that they are designed the HapticMaster,to assist in patient movement as the to help open and close the hand.One robotic therapy patient plays video games.The HapticMaster allows system for the hand incorporates the idea of using visual four degrees of freedom of movement and achieves feedback distortion to enhance motivation of patients a high bandwidth of force control using force feedback. during movement therapy [53.66].Using robotic force Chronic stroke patients who exercised with GENTLE/s fields to amplify the kinematic errors of stroke patients improved their movement ability [53.48,49].The Rut- during reaching may provoke novel forms of adaptation gers hand robotic device uses low-friction pneumatic of those patterns [53.4,67]

Rehabilitation and Health Care Robotics 53.2 Physical Therapy and Training Robots 1231 in different directions and assist in reaching in a straight line. Chronic stroke patients who received assistance during reaching with the robot improved their move￾ment ability [53.44]. However, they improved about as much as a control group that simply practised a matched number of reaches without assistance from the robot. This suggests that movement effort by the patient is a key factor for recovery, although the small sample size of this study limited its ability to resolve the size of the difference between guided and unguided therapy. Bi-Manu-Track Perhaps the most striking clinical results generated so far have come from one of the simplest devices built. Similar to a design proposed previously [53.45], the Bi-Manu-Track uses two motors, one for each hand, to allow bimanual wrist-flexion extension [53.46]. The de￾vice can also assist in forearm pronation/supination if it is tilted downward and the handles are changed. In an extensive clinical test of the device, 22 subacute patients (i. e., 4–6 weeks after stroke) practised 800 movements with the device for 20 min per day, five days per week for six weeks [53.46]. For half of the movements, the de￾vice drove both arms, and for the other half, the patient’s stronger arm drove the motion of the more-impaired arm. A control group received a matched duration of electri￾cal stimulation (ES) of their wrist extensor muscles, with the stimulation triggered by voluntary activation of their muscles when possible, as measured by electromyo￾graphy (EMG). The number of movements performed with EMG-triggered ES was 60–80 per session. The robot-trained group improved by 15 points more on the Fugl-Meyer scale, a standard clinical scale of movement ability with a range from 0 to 66 points in upper extrem￾ity function. It assigns a score of 0 (cannot complete), 1 (completes partially), or 2 (completes normally) for 33 test movements, such as lifting the arm without flexing the elbow. For comparison, reported gains in Fugl-Meyer score after therapy with the MIT-MANUS and MIME devices ranged from 0–5 points [53.47]. Other Devices to Undergo Clinical Testing Other devices to undergo clinical testing are as fol￾lows. The GENTLE/s system uses a commercial robot, the HapticMaster, to assist in patient movement as the patient plays video games. The HapticMaster allows four degrees of freedom of movement and achieves a high bandwidth of force control using force feedback. Chronic stroke patients who exercised with GENTLE/s improved their movement ability [53.48, 49]. The Rut￾gers hand robotic device uses low-friction pneumatic cylinders to help extend or flex the fingers, and has been shown to improve hand movement ability of chronic stroke subjects [53.50]. Simple force-feedback con￾trolled devices, including a one-DOF wrist manipulator and a two-DOF elbow–shoulder manipulator, were also recently shown to improve movement ability of chronic stroke subjects who exercised with the devices [53.51]. A passive exoskeleton, the T-WREX arm orthosis, pro￾vides support to the arm against gravity using elastic bands, while still allowing a large range of motion of the arm [53.52]. By incorporating a simple hand-grasp sensor, this device allows substantially weakened pa￾tients to practise simple virtual reality exercises that simulate functional tasks such as shopping and cooking. Chronic stroke patients who practised exercising with this non-robotic device recovered significant amounts of movement ability, comparable with the Fugl-Meyer gains seen with MIT-MANUS and MIME. NeReBot is a three-DOF wire-based robot that can slowly move a stroke patient’s arm in spatial paths. Acute stroke patients who received additional movement therapy beyond their conventional rehabilitation therapy with NeReBot recovered significantly more movement ability than patients who received just conventional rehabilita￾tion therapy [53.53]. RehaRob uses an industrial robot arm to mobilize patients’ arms along arbitrary trajecto￾ries following stroke [53.54]. Other Systems Currently under Development Several other robotic therapy devices are currently under development. For example, at the high end of cost and complexity are the ARM-In [53.55] and Pneu-WREX systems [53.56], which are exoskeletons that accommo￾date nearly naturalistic movement of the arm while still achieving a wide range of force control. A system that couples a immersive virtual-reality display with a haptic robot arm is described in [53.57]. A wearable exoskele￾ton driven by pneumatic muscles is described in [53.58]. At the lower end of cost/complexity are devices that use force feedback joysticks and steering wheels with a view toward implementation in the home [53.59–62]. Exam￾ples of recent, novel robotic devices for the hand are given in [53.63–65]: these devices typically follow an active assist therapy paradigm in that they are designed to help open and close the hand. One robotic therapy system for the hand incorporates the idea of using visual feedback distortion to enhance motivation of patients during movement therapy [53.66]. Using robotic force fields to amplify the kinematic errors of stroke patients during reaching may provoke novel forms of adaptation of those patterns [53.4, 67]. Part F 53.2

1232 Part F Field and Service Robotics 53.2.4 Robotic Therapy for Walking automation.The efforts of roboticists have been espe- cially focused on BWSTT rather than overground gait Background training because BWSTT is done on a stationary setup Scientific evidence that gait training improves recovery in a well-defined manner and thus can be more easily of mobility after neurologic injury started to accumu- automated than overground gait training.Randomized, late in the 1980s through studies with cats.Cats with controlled clinical trials have shown that BWSTT is SCI can be trained to step with their hind limbs on comparable in effectiveness to conventional physical a treadmill with partial support of the body weight therapy for various gait-impairing diseases [53.75-80]. and assistance of leg movements [53.68,69].Follow-These trials support the efforts towards automation of ing the animal studies,various laboratories developed BWSTT,as the working conditions of physical thera- a rehabilitation approach in which the patient steps on pists will improve if the robots do much of the physical a treadmill with the body weight partially supported by work,which in the case of BWSTT actually leads to an overhead harness and assistance from up to three occasional back injuries to therapists.Usually,only therapists [53.70-73].Depending on the patient's im-one therapist is needed in robot-assisted training,for pairment level,from one to three therapists are needed the tasks of helping the patient into and out of the for body-weight supported treadmill training(BWSTT),robot and monitoring the therapy.In the case of SCI with one therapist assisting in stabilizing and moving patients,a small randomized,controlled trial [53.76] the pelvis,while two additional therapists sit next to reported that robotic-assisted BWSTT with a first- the treadmill and assist the patient's legs in swing and generation robot required significantly less labor than stance.This type of training is based on the principle both conventional overground training and therapist- Part FI of generating normative,locomotor-like sensory input assisted BWSTT,with no significant difference found that promotes functional reorganization and recovery in effectiveness. 53.2 of the injured neural circuitry [53.74].In the 1990s, several independent studies indicated that BWSTT im- Gait-Training Robots in Current Clinical Use proves stepping in people with SCI or hemiplegia after Three gait-training robot systems are already in use stroke[53.70-721. for therapy in several clinics worldwide:the gait Gait training is particularly labor-intensive and trainer GT-I[53.81],the LokomatR [53.82],and the strenuous for therapists,so it is an important target for AutoAmbulatorTM [53.83](Fig.53.2). b) Fig.53.2a-c Gait-training robotic systems currently in use in clinics;(a)the gait trainer GT-I,developed by Hesse's group and commercialized by Reha-Stim (Germany):(b)the Lokomat,developed by Colombo and colleagues and commercialized by Hocoma AG(Switzerland),and(c)AutoAmbulatorTM,developed by the HealthSouth Corporation (USA)

1232 Part F Field and Service Robotics 53.2.4 Robotic Therapy for Walking Background Scientific evidence that gait training improves recovery of mobility after neurologic injury started to accumu￾late in the 1980s through studies with cats. Cats with SCI can be trained to step with their hind limbs on a treadmill with partial support of the body weight and assistance of leg movements [53.68, 69]. Follow￾ing the animal studies, various laboratories developed a rehabilitation approach in which the patient steps on a treadmill with the body weight partially supported by an overhead harness and assistance from up to three therapists [53.70–73]. Depending on the patient’s im￾pairment level, from one to three therapists are needed for body-weight supported treadmill training (BWSTT), with one therapist assisting in stabilizing and moving the pelvis, while two additional therapists sit next to the treadmill and assist the patient’s legs in swing and stance. This type of training is based on the principle of generating normative, locomotor-like sensory input that promotes functional reorganization and recovery of the injured neural circuitry [53.74]. In the 1990s, several independent studies indicated that BWSTT im￾proves stepping in people with SCI or hemiplegia after stroke [53.70–72]. Gait training is particularly labor-intensive and strenuous for therapists, so it is an important target for a) b) c) Fig. 53.2a–c Gait-training robotic systems currently in use in clinics; (a) the gait trainer GT-I, developed by Hesse’s group and commercialized by Reha-Stim (Germany); (b) the Lokomatr , developed by Colombo and colleagues and commercialized by Hocoma AG (Switzerland), and (c) AutoAmbulatorTM, developed by the HealthSouth Corporation (USA) automation. The efforts of roboticists have been espe￾cially focused on BWSTT rather than overground gait training because BWSTT is done on a stationary setup in a well-defined manner and thus can be more easily automated than overground gait training. Randomized, controlled clinical trials have shown that BWSTT is comparable in effectiveness to conventional physical therapy for various gait-impairing diseases [53.75–80]. These trials support the efforts towards automation of BWSTT, as the working conditions of physical thera￾pists will improve if the robots do much of the physical work, which in the case of BWSTT actually leads to occasional back injuries to therapists. Usually, only one therapist is needed in robot-assisted training, for the tasks of helping the patient into and out of the robot and monitoring the therapy. In the case of SCI patients, a small randomized, controlled trial [53.76] reported that robotic-assisted BWSTT with a first￾generation robot required significantly less labor than both conventional overground training and therapist￾assisted BWSTT, with no significant difference found in effectiveness. Gait-Training Robots in Current Clinical Use Three gait-training robot systems are already in use for therapy in several clinics worldwide: the gait trainer GT-I [53.81], the Lokomatr [53.82], and the AutoAmbulatorTM [53.83] (Fig. 53.2). Part F 53.2