Hands-off Assistive Robotics for Post-Stroke Arm RehabilitationJon Eriksson1, Maja J. Matari´c2and Carolee J. Winstein31,2Robotics Research LaboratoriesDepartment of Computer ScienceUniversity of Southern California941 West 37th PlaceLos Angeles, CA 90089-0781jonerik | [email protected] Behavior and Neurorehabilitation LabDept. of Biokinesiology and Physical TherapyUniversity of Southern California1540 E. Alcazar St., CHP-155Los Angeles, CA [email protected]—This paper describes an autonomous assistivemobile robot that aids stroke patient rehabilitation by providingmonitoring, encouragement, and reminders. The robot navi-gates autonomously, monitors the patient’s arm activity, andhelps the patient remember to follow a rehabilitation program.Our experiments show that patients post-stroke are positiveabout this approach and that increasingly active and animatedrobot behavior is positively received by stroke survivors.I. INTRODUCTIONThis project aimed to develop an autonomous assistivemobile robot that could aid persons post-stroke with armdisabilities during their rehabilitation. By encouraging thepatient to use the disabled arm, and by monitoring progress,the robot would provide support and motivation when regularphysical therapy is not suitable or available. Thus, the robotmay increase the rate and amount of recovery as well asthe general well-being of stroke survivors. This work ispart of our ongoing effort toward the development of novelhuman-robot interaction techniques for non-contact (hands-off) mobile robots, which is being incrementally evaluatedand improved through experiments with patients.Stroke is a major cause of neurological disability. Mostof those affected are left with some movement disability.Through concerted use and training of the affected limbduring the critical post-stroke period, such disability canbe significantly reduced [1]. The rate and amount of re-covery greatly depends on the amount of focused training.Evidence shows that the intensity and frequency of focusedtherapy can improve functional outcomes [2]. However, sincesuch rehabilitation normally requires supervision of trainedprofessionals, lack of resources limits the amount of timeavailable for supervised rehabilitation. As a result, the qualityof life of stroke patients is permanently reduced, and medicalcosts and lost productivity continue to be incurred.In this work, we propose the use of an assistive mobilerobot (Figure 1) as a complement to conventional physicaltherapy and rehabilitation, as a means of addressing theabove challenges of intensive rehabilitation. The approachinvolves the development of a safe, user-friendly, and afford-able mobile robot that is capable of following the patientin the home or hospital environment. The robot moni-tors the patient’s use of the stroke-affected limb, providesencouragement, guidance, and reminders. It also logs thepatient’s movement of the affected limb and keeps track ofrehabilitation progress for reporting to the physical therapist.The patient’s arm movement is registered with a light-weight inertial measurement unit (IMU) worn on the forearmmuch like a wristwatch. The robot behaves in response tothe sensed movements of the affected limb. For instance,it provides gentle reminders to the patient if the affectedarm has not been active for some period, and praise andencouragement if it has. The robot is also able to reportperformance data in analytical form to the rehabilitation staff,who can then use it to fine-tune the robot-assisted therapy.Our approach to robot-assisted hands-off rehabilitation isbased on an already validated post-stroke therapy method,which constrains the patient’s use of the healthy limb in orderto encourage the use and thus recovery of function in thestroke-affected one [3]. This process of exercising towardregaining lost movement is cumbersome and can be veryfrustrating for the patient. The robot’s task is to, in part,provide monitoring and reminding, but also in part to serveas a friendly companion during a difficult time. While therobot tracks and follows the patient by default, it leaves whenasked to; it is designed to be welcomed by the patient, notto be annoying, ignored, and shut off. Our work so far hasbeen in validating patients’ acceptance of a robot with a fewsimple behaviors; our ongoing work continues to improvethe robot’s capabilities of engaging the patient.II. RELATED WORKBecause of the growing elderly population and thus thepredicted rise in stroke [4], much effort has been madeduring the past decade to improve post-stroke rehabilitation.A robot-assisted arm therapy workstation has been developedfor training upper limbs and evaluating limb performance[5]. A number of similar devices employing hands-on reha-bilitation robotics technology have been developed [6] andinvestigated [7], [8]. In general, such mechanical devicesare worn by the patient during exercises, and use built-insensors to measure generated limb forces and movement.Results suggest that using such mechanical assistive guidesseems to be sufficient for enhancing movement recoverybut not that it is necessary. It appears that the purposeful,Fig. 1. An illustration of the interaction among the physical therapist,patient, and robot in the context of robot-assisted stroke rehabilitation. Therobot does not attempt to take the role of the trained professional; insteadit helps the patient by providing time-extended and individualized attention.meaningful functional task practice is the primary stimulusfor arm movement recovery.The above devices all depend on physical contact be-tween the patient and the robot, raising issues of safety,affordability, and feisiblity for in-home use. A non-contactapproach of the type we propose raises fewer of thoseconcerns. Furthermore, by focusing on regular activitiesand tasks performed in daily life rather than constrainedrepetitive exercises, the patients can be more easily motivatedto perform time-extended rehabilitation activities, potentiallyresulting in improved overall recovery outcomes.Research shows that pet ownership increases health andemotional well-being of elderly [9]. Moreover, it has beendemonstrated that people often attribute animal or humancharacteristics to even simple robot pets [10]. These inher-ently human tendencies provide a grounding for the hands-off assistive robot rehabilitation methods we describe, whichare aimed at having a positive effect on