Socially assistive robotics: Human augmentation versus automation

+ See all authors and affiliations

Science Robotics  15 Mar 2017:
Vol. 2, Issue 4, eaam5410
DOI: 10.1126/scirobotics.aam5410


Intelligent, interactive systems provide assistance by facilitating social interactions rather than by automating physical tasks.

Robotics development has been driven by the desire to automate work. The possible implications of machines doing human work are a topic of ongoing debates. Discussions often center on economic concerns, but another important outcome is the impact on human health. The science on longevity and resilience indicates that the drive to stay physically and cognitively active is necessary for health and wellness. The effects of increasingly sedentary lifestyles are already widespread and well known. The prevalence of mental health disorders, including depression and social isolation, is increasingly being recognized.

From surgery to telemedicine, robots are improving the accuracy, consistency, safety, and accessibility of health care. In the past 15 years, robotics has begun to enter a new domain of health and wellness: A growing body of evidence from behavioral science and neuroscience demonstrates that people experience interactions with physically embodied copresent agents (other people and pets, as well as robots) as more enjoyable and motivating than interactions with screens. In addition, people are more likely to be active, change their behavior, and learn in such contexts. Therefore, robotics (machines that were originally invented to automate work) is helping people, not by doing physical work, but instead by facilitating social interaction: encouraging people to do their own work.

Socially assistive robotics (SAR) focuses on developing intelligent, socially interactive machines that provide assistance through social rather than physical means (1). SAR is complementary to the more established field of rehabilitation robotics, wherein robots provide hands-on assistance, typically through perception and application of force to affected limbs. In contrast, SAR is hands-off, using noncontact methods for monitoring, coaching, and companionship that support convalescence and training. For example, a SAR stroke rehabilitation coach presents movement exercise challenges, often through demonstration; monitors the stroke patient’s movements; and provides feedback on exercise quality and needed improvements, with encouragement and motivation to continue.

SAR systems have been developed for challenging health care contexts—including stroke rehabilitation (2), autism behavior therapy (3), mental health care (4), Alzheimer’s disease and other types of dementia, and healthy elderly care (5, 6)—and are actively being explored for enhancing early childhood education (7) and other training contexts. All of these domains require dedicated, one-on-one, daily, time-extended interaction. Rather than replacing scarce human caregivers, SAR systems fill the gaps where humans are not available and serve to amplify human work.

Human augmentation amplifies and enhances human ability to do work. The field (8) encompasses many technologies: prosthetics, orthotics, and physically assistive devices that replace missing or lost functions; exoskeletons that extend physical abilities; collaborative systems that work alongside people to fill in and complement human abilities; and socially assistive robots that monitor and motivate human work and effort. Telepresence/remote-presence/teleoperation devices enable work at a distance and make it more precise.

In contrast to other areas of human augmentation, SAR requires social rather than physical interaction with the user. Social interaction is complex; evolutionary theories implicate complex social structures as a major driver of human intelligence. SAR designers must determine ways to achieve similar, compatible, socially interactive embodied systems that smoothly integrate the physical, cognitive, and social aspects of the robot. Inconsistency among those components and the user’s expectations are highly salient and constitute a distinct form of system failure: user rejection. The notion of the Uncanny Valley (9) is an example of such failure: A mismatch between the user’s expectations and the system’s behavior in any modality (physical appearance, movement, voice, verbal content, etc.) produces the “creepy” effect. Timing the response to the social partner(s) is another challenge of social interaction; SAR systems must operate on a social time scale, which is both personal and contextual; responding too slowly (due to sensing, computing, or other reasons), too quickly, or too repetitively (“robotically,” and therefore unnaturally for human users) breaks the social dynamic. Timing challenges apply to both physical (movement, gesture, and facial expressions) and nonphysical (speech and nonspeech vocalizations) communication modalities and their temporal coordination. Embodied communication is an open research area: More is conveyed in the subtleties of facial expression, head orientation, eye gaze, body position and orientation, locus of gesture, voice volume, cadence, and affect (emotion) in voice, all so-called back channels, than in spoken words. Observability is yet another challenge: Effective social interaction involves not only understanding the user’s observable behavior (movement, facial expressions, etc.) but also interpreting the user’s mood, emotional state, and intent. Last, personalization and adaptation of interaction are both necessary: Users are different in their specific social behavior, drives, behavior patterns, and individual preferences for novelty and variety versus consistency. SAR systems must detect and adjust to individual differences quickly and accurately if they are to be accepted and to remain effective.

Progress in SAR requires roboticists to work with social scientists, cognitive scientists, developmental scientists, and domain experts in the relevant applications, such as stroke, dementia, autism, early childhood education, and healthy aging, among others. It also depends on bringing together signal processing, machine learning, natural language processing, and other areas of computing for human-machine interaction. As in other areas of health research, there is a need for large data sets that meet human privacy rules and can be shared by the growing SAR research community to move the field forward beyond small, anecdotal user studies. Last, there is still a dearth of interactive, safe, and affordable robot platforms for research into SAR and human-robot interaction more generally; current platforms largely lack the physical features needed for addressing the research challenges outlined above.

The great need for effective, affordable, and personalized care for large and growing populations across the age span is a major as-yet-untapped driver for growth in robotics aimed at human augmentation. It is easy to imagine how such systems could serve to monitor and to assist physical, cognitive, and social development of children; aid in convalescence after brain injury or other trauma; rehabilitate after stroke; mitigate Parkinson’s disease, Alzheimer’s disease, and other neurodegenerative conditions; and aid in aging in place. SAR and other forms of human augmentation are complementary to automation and present a broad range of research challenges with opportunities for substantial impact.

Fig. 1 An example of a socially assistive robot.

Chili, a squash-and-stretch 6 degree-of-freedom robot with a digital face, was used to interact with 5- to 8-year-old first graders to teach and motivate them to make healthy food choices. The study involved 26 children interacting with the robot twice per week for 3 weeks and found that the children learned nutrition concepts and engaged with the robot and that their verbal responses to the robot became richer over time.



View Abstract

Navigate This Article