This review explores recent trends in the development and evaluation of assistive robotic arms, both prosthetic and externally mounted. Evaluations have been organized according to the CATOR taxonomy of assistive device outcomes, which takes into consideration device effectiveness, social significance, and impact on subjective well-being. Questions that have informed the review include: (1) Are robotic arms being comprehensively evaluated along axes of the CATOR taxonomy? (2) Are definitions of effectiveness in accordance with the priorities of users? (3) What gaps in robotic arm evaluation exist, and how might these best be addressed? (4) What further advances can be expected in the next 15 years? Results highlight the need for increased standardization of evaluation methods, increased emphasis on the social significance (i.e., social cost) of devices, and increased emphasis on device impact on quality of life. Several open areas for future research, in terms of both device evaluation and device development, are also discussed.
The last 15 years have seen the development of several robotic arms to support activity and occupation of individuals with limb loss or motor disabilities. New, lightweight robotic arms, like the SensorHand SPEED and i-LIMB, can provide good function and cosmesis for individuals with upper limb amputation or deficiency. Heavier robotic arms, even industrial arms, may also support object manipulation by people with motor disabilities. Robotic arms that may be too heavy to be worn on the body may be mounted to desktops, wheelchairs, or overhead workstations. In these configurations, robotic arms can support manipulation required for office work, to eat, or to operate devices like a TV remote control. Many arms specifically designed for use by people with disabilities are, in fact, now commercially available, including the ARM and Raptor manipulators. Commercially available industrial robotic arms are also seeing use in the context of assistive applications, including the Puma robotic arm.
Utility of robotic arms can be judged along many dimensions. A robotic arm may be useful in that it demonstrably enhances the operator’s ability to interact with the physical environment or participate in occupation, for example. Alternatively, it may reduce cost associated with providing assistive care or visits to a hospital or clinic. Finally, a robotic arm may enhance feelings of personal well-being or confidence in activities that require the device as well as those that do not. These dimensions of device “utility” have recently been codified into the CATOR taxonomy of assistive device outcomes, which organizes outcomes according to device effectiveness, social significance, and impact on subjective well-being. Device effectiveness reflects measures of impact on user functioning and activity participation as well as impact on the accessibility of a users’ environment. Social significance includes measures of impact on society more generally; these may reflect the labor of caregivers, use of the health care system, or housing cost. Finally, subjective well-being measures focus on individuals’ satisfaction with a device, and with their lives, more generally, after device adoption.
The purpose of this review is to explore recent trends in the development and evaluation of assistive robotic arms, both prosthetic and externally mounted. Evaluations of the devices have been organized along the axes of the CATOR taxonomy. Questions that have informed the review include:
Are robotic arms being comprehensively evaluated along axes of the CATOR taxonomy?
Are definitions of effectiveness in accordance with the priorities of users?
What gaps in robotic arm evaluation exist, and how might these gaps be best addressed?
What further advances can be expected in the next 15 years?
To perform the review, a literature search was performed in the following research databases: IEEE Explore, PubMed, Cinahl, PsycInfoNet, Ovid, and Embase. Keywords used to guide the searches included combinations of “assistive,” “disability,” “user trial,” “robotic arm,” “robotic manipulator,” “upper extremity,” and “prosthesis.” Several articles were iteratively added to the review based on references contained in previously selected articles. Articles were selected based on the following criteria:
Published in the last 15 years (ie, 1994 or later)
Include evaluation of specific assistive robotic arms, either prosthetic or externally mounted
Include evaluation based on the experience of potential adult users
Five hundred and seven articles were retrieved as a result of the search. Articles that related to orthotic devices were excluded, as were articles with a therapy focus, review articles, and articles in which users were exclusively children. The remaining 51 articles included evaluations of commercial devices based on user trials and surveys, evaluations of devices still in the research and development phase, and evaluations of novel interfaces to existing commercial devices.
Externally mounted manipulators
Robotic manipulators that are not attached directly to an individual’s arm can loosely be organized into 3 categories: those that connect to a wheelchair ; desktop-mounted arms embedded in activity-specific workstations ; or arms that move about the room on their own mobile base. The same devices, moreover, can be organized according to the range of their functionality. Some devices are designed for a single purpose, like self-feeding, whereas others are capable of performing a variety of tasks within a particular activity context, like the office. Finally, general purpose robotic arms exist, and hold the potential to perform a wide range of functional manipulations in many contexts, including factory environments and homes.
Many externally mounted manipulation aids are now commercially available; their cost is governed, at least in part, by their functionality. Commercially available tools include the ARM (Assistive Robotic Manipulator, Exact Dynamics; formerly the MANUS) ( Fig. 1 ) and Raptor robotic arms, the MASTER/EPI-RAID system (sold as the AFMaster, Afma Robots, France ), the DeVAR system (sold by The Tolfa Corporation, Palo Alto, CA ) and various desktop-mounted feeders (examples include HANDY, Rehab Robotics, UK ; Neater Eater, Neater Solutions Ltd, UK ; My Spoon, SECOM Co. Ltd, Tokyo ; and Winsford Feeder, Winsford Products Inc, Pennington, NJ ) ( Fig. 2 ). As is perhaps to be expected, the more functionality provided by a particular system, the higher its cost. Feeders start at roughly $2500. The HANDY costs $6300 yet supports a range of functional activities in addition to eating, like shaving, painting, and applying make-up. The Raptor and ARM arms, which are intended for general-purpose manipulation, retail at $12,000 and $35,000, respectively. Finally, workstation robots that include a general-purpose robotic arm tend to be most costly. In 2004, the AFMaster retailed for $50,000. Costs of several devices, drawn from Hillman, are indicated in Table 1 .
|Robotic arm name||ARM (MANUS)||HANDY||IRVIS||ProVAR||Winsford feeder|
|Arm profile||6 DOF manipulator |
1 DOF grip
|5 DOF manipulator |
1 DOF “scoop”
|5 DOF manipulator |
|6 DOF manipulator |
1 DOF grip
|2 DOF manipulator |
1 DOF “scoop”
|Category||Wheelchair mount||Desktop mount (feeder)||Desktop mount (vocational)||Desktop mount (vocational)||Desktop mount (feeder)|
|Cost||$35,000||$6300||ProVAR is not sold commercially; DeVAR, which was extended by ProVAR, sold for $50000–100000||$2500|
|Robotic arm name||BEESON FEEDER||MOVAID||EL-E||MASTER/EPI-RAID||HELPING HAND/RAPTOR|
|Arm profile||2 DOF manipulator |
1 DOF “scoop”
|8 DOF manipulator |
1 DOF grip
|5 DOF manipulator |
1 DOF grip
|6 DOF manipulator |
1 DOF grip
|5 DOF manipulator |
1 DOF grip
|Category||Desktop mount (feeder)||Autonomous mobile||Autonomous mobile||Desktop mounted (vocational)||Wheelchair mounted|
|Cost||No longer sold||$50000||$12000|
Widespread access to commercially available robotic manipulators is governed partly by recommendations of health care service providers and partly by reimbursement policies. For costly devices, reimbursement potential exists only for limited groups of devices and users. The ARM robotic arm, for example, has received reimbursement support from the government of the Netherlands; between 2001 and 2004 this was provided through an exploratory government grant and since 2004 it has been provided formally by the country’s health care system. There are now more than 150 ARM arms owned by individual users. Comparable reimbursement support from United States insurance providers, however, does not exist at present. Evidence that might support reimbursement in the United States for similarly priced systems has been mixed; in 1994, for example, evaluators from the United States Veterans Affairs found the DeVAR desktop workstation insufficiently robust and overly task limited to be considered medically prescribable.
In the United States, use of robotic manipulation aids has additionally been limited by a lack of standardized prescription protocols among service providers. A 1996 study of 12 United States care facilities, for example, found occupational therapists received little or no training in the use of robotic feeders, yet they felt the devices to be too expensive, difficult to use, unattractive, unreliable, inconvenient, or time consuming to set up and install. In the surveyed facilities, 920 individuals were found to be completely unable to feed themselves, yet only 3 were using a powered feeding device. These aids had been acquired as a result of factors like device availability and the motivation of individuals and their therapists. None of the centers had a standardized procedure by which clients’ needs or appropriateness for robotic feeders was evaluated.
Unlike prosthetic arms, control of externally mounted robotic arms is rarely driven by myoelectric signals (see Kyberd and Chappell for an exception). Instead, users must supply control signals to the robots using a variety of possible input devices, which may include voice input, head-mounted switches, track balls, joysticks, mice, keyboards, or touchscreen devices. The robotic device El-E additionally allows users to input control commands using a laser pointer, which may be mounted to the ear or controlled with the hand.
The choice of input device depends on users’ comfort and physical capacity. In the study by Tijsma and colleagues, prototypical development of a wide range of potential input devices, like large lapboard mounted push buttons and individually molded hand-held grips, are described in detail. User trials with El-E found input device preference among users with amyotrophic lateral sclerosis (ALS) to be partially explained by levels of functional arm impairment recorded on the Revised ALS Functional Rating Scale (ALSFRS-R). Those with higher “handwriting” scores on the ALSFRS-R were more likely to prefer controlling a laser pointer input with the hand; those with lower scores were more likely to prefer a pointer mounted to the ear.
Inputs to robotic manipulators may also be constrained by cognitive factors. The standard interface to the ARM, for example, requires an individual to activate a switch to select which degree of freedom (DOF) he or she would like to manipulate; after selection, this DOF may be manipulated independently of the others. The ARM may alternatively be controlled in Cartesian coordinates, using a similar “switching” form of control. Using either switching method, performing a complex task like grasping and lifting a glass of water requires many DOF switches and an ability to plan the order of these switches, which typically results in lengthy task times. In the study by Mokhtari and colleagues, for example, a user’s effort to grasp a remote control was found to take more than 9 minutes with the ARM. Task performance may be further complicated by the fact that, like most prosthetic arms, externally mounted robotic arms typically do not provide force-feedback to the user. Tijsma and colleagues designed several training tasks in an effort to demonstrate to users the vertical forces they were applying to objects with the robot. The ability to translate what was learned during these training tasks to novel tasks, however, depended on the cognitive abilities of the user.
To reduce physical and cognitive demands associated with device control, many systems employ varying degrees of abstraction in their command structure. El-E, for example, takes commands in the form of a gesture to an object on the floor, and this gesture can be made using a laser pointer or touch screen device. Robot navigation to the object and grasp are then entirely automated. Even for devices that must currently be manually controlled, some degree of abstraction may be preferred by users. Eftring and colleagues found that roughly 50% of task time with the ARM, on average, related not to moving the robot’s end effector but to the execution of grasping. Users indicated they would prefer the device with an automated grasp, were it to exist.
ProVAR, MASTER/EPI-RAID, and Movaid all feature task-level degrees of abstraction in the control of a robotic arm. Using MASTER/EPI-RAID, for example, a user may grasp a book from the bookshelf by selecting prerecorded locations of books from a pull-down menu; alternatively, he or she may select a book by title. Movaid similarly allows users to control a moving robot to execute tasks, like operating a microwave, with abstract and task-level commands. ProVAR not only allows users to invoke prerecorded activities, but to visually draft new task “plans” by manipulating a virtual arm via click and drag operations or keyboard inputs of joint angles. Actions on the virtual arm can be refined until they meet a user’s needs, then stored and played on the physical arm at any point in the future. Visualizing 3-dimensional kinematics of the virtual arm on a 2-dimensional screen has, however, proven to be difficult for some users.
Trends in Manipulator Evaluation
Users and environments explored
Studies that have evaluated external assistive manipulators have generally taken place in clinical environments, yet many evaluations now focus on use in homes. Comparatively few evaluations have taken place in vocational environments. Exceptions can be found, although results from these studies relate to the experience of a relatively small number of users (less than 5). Both MASTER/EPI-RAID and ProVAR, however, have been fairly extensively evaluated in clinics and hold potential to be evaluated in working offices. DeVAR, the device that later evolved into ProVAR, was in fact deployed at the Pacific Gas and Electric Company during its evaluation.
Users typically include severely disabled individuals with spinal cord injuries, spina bifida, ALS, cerebral palsy, and traumatic brain injuries, as well as others. For the most part, user studies with externally mounted arms have been performed on small populations, typically less than 15 in number. Notable exceptions relate user studies with MASTER/EPI-RAID, the ARM, Movaid, and HANDY robots.
Use of the ARM robot was evaluated in the context of user trials by the Netherlands. Gelderblom and colleagues describe 2 sets of year-long experiments. One compared 13 users of the ARM to a disability-matched control group of 21 nonusers, and the other introduced the ARM to 10 novel users with disabilities. Comparisons in the latter study were made within subjects, before and after device acquisition. Results focused on the degree to which individuals could perform activities independently and whether the device replaced or minimized required care. Users without the ARM were found to receive an average of 3.7 hours of caregiver assistance daily, whereas those with the device required 2.8. In addition, provision of an ARM to novel users reduced dependence on a caregiver by 0.7 to 1.8 hours per day. Reduced dependence on a caregiver was estimated to translate into a saving, per client, of between 7000 and 18,000 euros annually.
In a similarly large study, use of the MASTER/EPI-RAID was evaluated by 91 users with tetraplegia in France. MASTER/EPI-RAID was installed at various locations for a year, including clinics and individual homes. In these locations, users were asked to evaluate the device’s support of their leisure, vocational, and domestic activities. Results indicated several leisure activities, like turning pages of a magazine and setting up a television, to be both effectively implemented by the device and of high priority to users. Among domestic tasks, using the refrigerator and serving guests were similarly viewed as both extremely important and well implemented.
Evaluations on this scale and of this time duration have resulted from various European funding initiatives including SPARTACUS in France, TIDE in the European Union, and funding from the Netherlands government. No comparably large or long-term evaluations of nonprosthetic assistive robotic devices were found to have taken place in North America in the last 15 years. The largest North American evaluations have focused on the ProVAR workstation and off-the-shelf feeding devices ; in each of these studies, 12 users provided data regarding device effectiveness and usability. Durations of both studies were less than 1 week.
Popular evaluation metrics
Many evaluations of externally mounted assistive arms focus on device effectiveness. Task success rates and times to complete a task, for example, are measures of device effectiveness, either quantitatively or qualitatively. To evaluate functional impact on users, standardized occupational therapy assessments were used to evaluate the ASEL fixed workstation arm.
Evaluations of the social significance of robotic manipulators, by comparison, are relatively infrequently performed. Training time and the level of caregiver prompting required to effectively use a device are measured in several studies ; results from these studies, however, are mixed. Gelderblom and colleagues found that required activities of daily living (ADL) assistance dropped when the ARM was introduced to a home by as much as 1.8 hours in a day. By contrast, queried caregivers felt the HANDY was time consuming to set up and clean, and did not reduce time assisting with meals. In the study by Hermann and colleagues, however, 92% of clients and their assistants indicated use of a feeding device was an improvement over the way meals had previously taken place. Device setup and cleanup time in this study were found to be, on average, less than 4 minutes, whereas meals lasted 30 to 40 minutes. Of all surveyed studies, only Gelderblom and colleagues evaluated the monetary costs associated with device use in terms of reduced caregiver assistance. Cost savings for individual care resulting from device use were estimated to be able to pay for the device within 3 years.
Many of the studies included some measure of subjective well-being attributable to device use, although the precise measures used vary substantially. Several studies include, in their evaluations, some measure as to users’ satisfaction with devices, more often than not a Likert scale. Standardized device evaluation measures, like the Quebec User Evaluation of Satisfaction with Assistive Technology (QUEST 2.0), were used in the evaluation of the ARM.
Opportunities for Future Research
Several robotic manipulators are on the market and are reasonably affordable (several feeders cost less than $5000) and at least one general-purpose robotic arm retails for less than $15,000. Along with increasingly moderate price tags, there is evidence to suggest that devices like robotic feeders and general-purpose arms reduce the amount of care required by an individual and increase his or her independence. Nevertheless, robotic manipulation devices have yet to see widespread use in North America.
Comprehensive evaluation and effective dissemination of information to health care practitioners may promote increased use. Cost savings associated with the ARM device, for example, have been demonstrated in the European Union, but have yet to be translated into a North American context. Device evaluations that span longer time periods can also provide information needed to gauge costs associated with device maintenance and repair.
Consumer perception of factors that effect long-term use of robotic arms have been identified and prioritized; these may prove to be valid predictors of device acceptance and reduce the rate of device abandonment. For robotic arms, the 5 most important factors are device effectiveness, operability, dependability, affordability, and flexibility.