We found that participants with stroke substantially reduced their force production during a typical robot-assisted therapy tracking task, when presented with a secondary visual distractor. This effect was more pronounced when the arm used for tracking was hemiparetic. Introduction of sound feedback of tracking error allowed participants to perform the distractor task while maintaining their effort at the tracking task. We first discuss the implications of these results for robot-assisted therapy, and then discuss sound feedback with respect to robotic therapy device design.
Distraction, attention demands, and robot-assisted therapy
An unintended consequence of robot-assisted therapy is that the patient may sometimes reduce his or her efforts toward trying to move, as has been documented for arm  and gait training . Ironically, this reduction of effort is facilitated at least in part by the robot itself: robotic assistance preserves the desired kinematics of motion, reducing the errors that might normally keep effort levels high. Such a reduction in effort may reduce the effectiveness of training. For example, one recent study found that training with a gait robot without any feedback of effort, a training approach which had previously been documented to reduce the energy consumption of individuals affected by stroke during walking  compared to therapist-assisted gait training, was about half as effective as conventional gait training without robotic assistance to the legs, at least for chronic stroke subjects who were ambulatory at the study onset. Another recent study compared passive range of motion exercise of the upper extremity to EMG-triggered FES, which required effort from the patient, and found that the passive exercise was substantially less effective . Comparisons of active and passive motor learning in non-impaired subjects are consistent with this finding [34–37]. If patient effort is important for promoting motor recovery, then identifying the factors that reduce effort, and designing ways to counteract these factors is important. In the present study, we found that introduction of a simple visual distracter task substantially reduced the effort of participants with chronic stroke during a standard robot-assisted therapy tracking task.
A similar reduction was not found for age-matched participants with stroke who used their non-paretic arm to reach, nor for participants without impairment. We hypothesize, first, that stroke survivors required increased attention to move their paretic arms; i.e. they have reduced automaticity for arm movement. Then, the propensity for slacking is likely tied to this increased attention requirement. These results are consistent with the finding that a secondary cognitive task reduces gait speed after stroke , although in that study, unlike the current one, the reduction seemed more associated with aging than the stroke per se. An interesting follow-up experiment would be to measure whether non-impaired participants slack when they make high-effort movements, to determine if the increased attention demand is related to weakness due to the stroke or the stroke itself. Attentional demand has previously been found to affect maximum force production in non-impaired subjects .
In this study we examined how effort changed with distraction, because we hypothesize that effort is linked to clinical outcomes. Other studies have found that short-term motor learning itself degrades in the presence of a distracter, with the degradation worse in the beginning of learning or when subjects have a motor deficit [20, 36, 40–44]. The present study confirms that even a simple visual task acts as an interfering influence on movement control of task after stroke, leading us to hypothesize that short term learning also would be affected by a visual distracter. This research thus suggests that it is important to remove even simple distractors from the training environment during robot-assisted movement training of people with stroke. Failure to control for distracting influences may at a minimum increase variability of results, and at worse diminish clinical benefits of robotic therapy. Another important direction for design of robot therapy is to reduce the assistance as much as possible. For example, if users of the devices experience obvious kinematic consequences when they are distracted, they may be less inclined to become distracted. In the optimization framework for modeling slacking we developed previously , the effects of a distractor as observed here could be accounted for by a reduction in the internal weight assigned to the effort component of the cost to minimized. In this framework, the cost function that the motor system minimizes would thus be affected by the attention demands placed on the motor system.
Sound feedback and robot-assisted therapy
Remarkably, we found that introduction of a simple form of auditory feedback eliminated the slacking that arose from performing the secondary distracter task. Participants not only continued to perform the distracter task with a similar success rate, but increased their effort back toward their baseline levels with the aid of auditory feedback. A likely explanation is that introduction of the visual distracter task overloaded the visual-motor channel; provision of feedback through the auditory system allowed better parallel processing. Rather than acting as a confounding influence or another distracter, the sound feedback enhanced the visuo-motor control because it provided similar information .
An important implication of this finding is that increased attention should be paid to incorporating effective forms of auditory feedback during robot-assisted movement training. Our impression is that auditory feedback is underutilized in most robotic therapy systems, playing a role as background music or signifying only task completion, although there are attempts to use auditory feedback in a more sophisticated way (e.g. [22, 46–48]. In one study, when people with chronic stroke practiced reaching with sound feedback that informed them about the deviation of their hand from the ideal path, they significantly reduced their position error after training . A control group that did the same exercise without feedback did not improve its performance. In another study, a virtual reality training system that incorporated sound feedback of reach position and speed helped subjects with traumatic brain injury improve their reaching ability . Another study found that lower extremity training of individuals with chronic hemiparesis using a robotic device coupled with Virtual Reality (including visual and audio feedback) improved walking ability in the laboratory and the community better than robot training alone .
These studies suggest that incorporation of augmented feedback can improve not only performance but also long-term motor learning after stroke. In the present study, we only demonstrated that auditory feedback improves short-term performance, measured by force output and tracking error. Future studies are needed to determine how providing auditory feedback of error can best improve learning of arm movement after stroke. We hypothesize that auditory feedback can serve to keep the subjects effort level elevated, as demonstrated here, which should improve use-dependent plasticity by reducing passivity. However, there is a possibility that subjects could come to rely on the auditory feedback to drive their performance, reducing transfer to real-life arm movements in which auditory feedback is not available. Thus, in testing the long-term effect of auditory feedback, in may be important to fade the feedback, or to provide it only intermittently, in order to reduce any possible growing dependence on it. Further, challenging the patient by intermittently providing a distracting environment with and without the aid of auditory feedback to overcome that distraction may be an appropriate way to allow people to learn to move well in the presence of distractors.
Another recent study found that the effect of sound feedback during reaching after chronic stroke depended on the hemisphere that was damaged by the stroke . In this study, participants heard a buzzing sound similar to the sound of a fly, with the volume of the buzz increasing with proximity to a reach target, and in some cases, the spatial balance of stereo sound was also altered by the orientation of the hand with respect to the target. Such sound feedback improved abnormal curvature in participants with right hemisphere damage (i.e. participants who were left hemiparetic, like the ones in our study), and degraded curvature, peak velocity, and smoothness in participants with left hemisphere damage . Robertson suggested that this result might be explained by either a difference in processing of auditory information, possibly due to receptive aphasia associated with left hemisphere damage, or to the fact that each hemisphere has a different role in movement control.
In the current study, we used a small sample of people with left hemiparesis for convenience: the robot was setup for left-handed use, and switching it was cumbersome. This choice may have been fortuitous, as the Robertson study suggests that people with left hemiparesis benefit more from sound feed-back. Further investigation is needed to understand if the sound feedback provided during a distraction task could be helpful also for right-hemiparetic subjects. Another factor affecting generalizability of the current results is that the participants recruited presented a narrow range of impairments at the affected upper extremity (Fugl-Meyer score range 15-32). In addition, the study excluded individuals presenting severe impairments at the affected upper extremity, which represent up to 30% of stroke survivors . Future studies should look also at the impact of auditory feedback on a broader spectrum of level of impairment after stroke. Finally, upcoming research should also examine how auditory feedback can best be crafted to improve learning and motor recovery.