ACKNOWLEDGEMENTS This work was funded by the Quality of Life Engineering Research Center, NSF EEC-0540865.
CONFLICTS OF INTEREST The authors declare no conflicts of interest.
Bambi R Brewer, Rehabilitation Science and Technology, 5044 Forbes Tower, University of Pittsburgh, Pittsburgh, PA 15260, USA. E-mail: email@example.com
Robotic devices for therapy have the potential to enable intensive, fully customized home rehabilitation over extended periods for individuals with stroke and traumatic brain injury, thus empowering them to maximize their functional recovery. For robotic rehabilitation to be most effective, systems must have the capacity to assign performance goals to the user and to increment those goals to encourage performance improvement. Otherwise, individuals may plateau at an artificially low level of function. Frequent goal change is needed to motivate improvements in performance by individuals with brain injury; but because of entrenched habits, these individuals may avoid striving for goals that they perceive as becoming ever more difficult. For this reason, implicit, undetectable goal change (distortion) may be more effective than explicit goal change at optimizing the motor performance of some individuals with brain injury. This paper reviews a body of work that provides a basis for incorporating implicit goal change into a robotic rehabilitation paradigm. This work was conducted with individuals without disability to provide foundational knowledge for using goal change in a robotic environment. In addition, we compare motor performance with goal change to performance with no goal or with a static goal for individuals without brain injury. Our results show that goal change can improve motor performance when participants attend to visual feedback. Building on these preliminary results can lead to more effective robotic paradigms for the rehabilitation of individuals with brain injury, including individuals with cerebral palsy.
Individuals with brain injury who receive intensive therapy have the potential to make substantial functional improvements years after injury.1,2 Because of the high cost of treatment and the limited number of therapists, however, most individuals with motor impairment resulting from brain injury do not receive the treatment needed to maximize function. Robotic therapy has been proposed as a potential solution to this problem. Robots have the ability to deliver the intensive therapy necessary for functional improvement while incorporating this therapy into interesting games to improve compliance.3 While robotic rehabilitation currently takes place primarily in the clinic, a vision of those in this field is to enable daily practice by patients in their own homes.4–6
In conventional rehabilitation, a therapist assesses an individual’s progress and helps him or her set a performance goal for each exercise. This process is important since assigned difficult goals have been shown to improve performance in a variety of tasks for individuals with stroke and traumatic brain injury (TBI).7,8 To optimize motor performance, assigned goals should increase in difficulty over the course of each session because chronic brain injury is frequently associated with learned non-use. ‘Learned non-use’ refers to entrenched habits of limited movement that an individual develops in the period immediately following injury.9,10 These learned limits do not reflect an individual’s actual motor capacity, and individuals must be encouraged to move beyond these habitual limits.2,10 Learned non-use makes it difficult to determine the appropriate goal to assign to a given individual. Incrementing goals in a structured way provides a straightforward method of ensuring that each individual is sufficiently challenged during a rehabilitation session.
Goal change should be a part of robotic rehabilitation, but explicit goals that become more difficult with time may not be the best way to optimize the performance of individuals with chronic brain injury. As a result of repeated failures when attempting motor tasks, these individuals may have developed low self-efficacy11 or a high motive to avoid failure, which may cause them to resist working toward difficult goals and to prefer to practice tasks at an undemanding level of performance.12,13 Depression14,15 and pathological fatigue16 may also cause individuals to be reluctant to participate fully in rehabilitation. Individuals with depression and/or pathological fatigue or who have low self-efficacy and a high motive to avoid failure may be reluctant to strive for goals that they perceive as continually increasing in difficulty.
Implicit goal change is our proposed solution to this problem for individuals with chronic brain injury. By ‘implicit goal change,’ we mean gradually and imperceptibly changing the visual feedback received by the participant such that improved performance is required to reach the same visual goal. For example, gradually increasing the amount of joint extension required to cause a particular change in the visual display might encourage a patient to extend the joint farther without realizing that he or she is doing so. Implicit goal change can be thought of as a type of distortion, and the terms will be used interchangeably in this paper.
While we originally framed our work in the context of stroke and TBI, we believe the idea of implicit goal change may be equally applicable to individuals aging with cerebral palsy (CP). Like individuals who acquire brain injuries later in life, persons with CP exhibit characteristics of learned non-use. However, individuals with CP have never learned to use the affected limb in a typical manner; therefore, for these individuals, the concept of learned non-use has a developmental component as well. Constraint-induced movement therapy has shown promise for individuals with CP,17 which indicates that the massed-practice paradigm typical of robotic rehabilitation may also be useful in improving function. In addition, robotic training could help individuals avoid secondary complications of CP. As with individuals with other types of chronic brain injury, individuals with CP are at high risk for low self-efficacy, depression, and pathological fatigue. For these reasons, individuals with CP may perform better with goal change that is implicit.
In this paper, we will review our previous work that has developed a foundation for the use of implicit goal change in a robotic rehabilitation paradigm. This work was conducted with individuals without brain injury. We will then describe the preliminary results of a recent experiment comparing motor performance with goal change to performance with a static goal; this preliminary work was also conducted with individuals without brain injury. Finally, we will discuss the implications of our work for individuals with CP.
Review of previous work
Lower bound on distortion
Before incorporating implicit goal change into experiments or a rehabilitation paradigm, we measured the smallest amounts of force and position change that could be reliably perceived by individuals without neurological dysfunction. These quantities, which are known as the ‘just- noticeable differences’ (JNDs) for force and position, provide a lower bound on the amount of implicit goal change that cannot be perceived by a user in the robotic environment. 18
The experimental environment used in this experiment included a Premium 1.5 model PHANTOM robot from Sensable Technologies (Woburn, MA, USA; Fig. 1). On each trial, the participant used the robot to sample two forces and decided whether they were the same or different. The participants' answers were used to calculate their JND.19,20 The position JND was measured in a similar manner. Both experiments were conducted with both young and elderly participants without disability. In addition, two young individuals with TBI and one elderly individual with stroke participated in both experiments.
The mean (SD) force JND measured for young participants without disability was 19.7% (1.85), which was significantly smaller than the JND of 31.0% (3.99) that was measured for elderly participants without disability. The mean position JND for young participants without disability was 3.99mm (0.434). This was not significantly different from the value of 6.32mm (1.38) measured for elderly participants without disability, though there was a trend in that direction. The JNDs for participants with TBI or stroke were approximately three times as large as those measured for the corresponding control groups.21 Our values for the JNDs are larger than those measured by previous researchers20,22–27 because we varied background dimensions that were constant in other experiments28,29 (i.e. distance as well as force varied from trial to trial). We did this to increase the amount of distortion that would be imperceptible in our robotic environment.
The JNDs for young participants without neurological dysfunction provide a lower bound on the amount of implicit goal change that is imperceptible for force and position. For example, if a visual display of force is changed so that the user must exert 15% more force to reach the same visual goal, the participant will be unable to detect this change because it is below the force JND of 19.7%. In addition, it may be possible to increase the amount of change that is imperceptible if the visual goals are incorporated into a game that distracts the user’s attention. Another factor that could decrease the perceptibility of change is introducing change gradually; the JND is usually measured by allowing the participant to go back and forth between the stimuli, whereas our technique introduces a delay between different levels of distortion. The force JND increases with age, which echoes the results of studies on the relationship between age and two-point discrimination.27,30 Our preliminary results suggest that force and position perception are diminished in individuals with chronic brain injury. Thus, large amounts of distortion or implicit goal change may be imperceptible for older individuals and individuals with brain injury.
Use of distortion to manipulate force production and movement distance
After completing the JND studies, we sought to show that gradual distortions beyond the JND could be used to affect motor performance in a robotic environment. We first considered the effects of distortion on force production.30 This experiment took place in the experimental environment described above. On each trial of the experiment, the participant was asked to push against the PHANTOM robot to produce a particular level of force, from 1 to 5. Throughout the experiment, the participants tried to be as consistent as possible in the force they produced for each level. We implemented distortion by gradually changing the range of force that was mapped to a visual-feedback bar shown on the computer screen. Background dimensions were varied to ensure that the JNDs measured in the previous experiment were valid for this design. No-feedback trials were interspersed throughout the experiment; on these trials, the participant was asked to produce the force level without seeing the visual feedback bar.
We conducted this experiment with young and elderly participants without disability, two participants with TBI, and two participants with stroke. In addition, two control experiments were conducted with young participants without disability. In the first control experiment, participants experienced no distortion; in the second, participants experienced distortion but were informed that the visual display might be distorted. Key results are summarized in Figure 2.
This set of experiments showed that visual feedback distortion can be used to increase force production within a single experimental session. Throughout this experiment, participants attempted to be consistent in the force they produced for each level. Despite this effort, upward trends in force with distortion were seen for both young and elderly participants, as well as for a few participants with brain injury. This upward trend was observed even when participants were informed that the visual display might be distorted. The upward trend was also seen on the no-feedback trials for young participants without disability.
We hypothesize that participants used the visual feedback bar to set a visual goal for each level and tried to keep this goal constant across the experiment, even when the actual force corresponding to this visual goal was increasing. The distortion affects performance even on the no-feedback trials because participants adjust the kinesthetic goal to match the changed visual goal. Thus, vision appears to dominate kinesthesis in this robotic environment. This dominance most likely occurs because participants determine that vision provides more precise feedback for this task; similar cases of visual dominance are common.27,31 However, the total amount of distortion in this experiment was quite large, approximately two JNDs. This total distortion was reached via a series of small steps, each step less than the JND for force. This result indicates that even large distortions may be effective in rehabilitation if they are reached via small, imperceptible steps.
This series of experiments focused on force production, but we also used distortion to increase the movement distance of young participants without neurological disease or injury. For this reason, we expect that our results concerning force production can be generalized to movement distance.
Use of distortion in a difficult coordination task
In the work described above, participants were free to set the target force or distance specified by a particular numerical magnitude. The distortion then changed the force or distance corresponding to the participant’s self-chosen visual goal for each level. After completing this set of experiments, we examined a different distortion paradigm using error enhancement.32 In this paradigm, we asked participants to work at a demanding task with an objective performance criterion – a clear definition of ‘perfect’ performance. As a participants learned the task, we presented feedback about their error in performance relative to this criterion. Distortion was implemented as an artificial inflation of this error. We tested this paradigm in the context of a coordinated pinch task because coordination of multiple fingers is important for many activities of daily living. The experimental environment included two 1.0 Premium PHANTOM robots, one coupled to the participant’s index finger and one to the thumb. We conducted this experiment with young participants without brain injury.
Error-enhancing distortion was applied first to one finger, then to the other, and finally to both simultaneously. When the distortion was applied to a single finger, the user focused attention on that finger, and the performance of the other finger deteriorated. Although distortion affected ongoing performance while the individual finger error was distorted, it did not improve performance when both finger errors were distorted together. This result appears to contrast with those of our previous work. This can be understood, however, if we consider that because this task involved a definition of ‘perfect’ performance, each participant was effectively assigned the difficult goal of reaching this objective standard. This goal encouraged participants to exert more effort in the target task, and error-enhancing distortion applied to both fingers may have been ineffective because participants were not able to further increase the effort exerted. Note as well that participants in this experiment were without brain injury and were presumably not withholding reserve effort. Difficult assigned goals may not optimize performance of individuals with brain injury, who tend to withhold reserve effort for the reasons outlined above.
Distortion in a rehabilitation paradigm
After conducting the experiments described above, we designed a preliminary rehabilitation protocol with visual goal change.33 The experimental environment included two 1.0 Premium PHANTOM robots, one coupled to the index finger and one to the thumb. Each participant practiced pinching and extending the index finger and thumb to select a letter in a game of hangman. The range of motion required to play the game was based on measurements made during a short calibration program. We implemented goal change by gradually and imperceptibly increasing the range of motion required to select letters. Each step was designed to be less than the JND for distance, and participants did not notice that this change was occurring. We conducted initial tests of this protocol with one individual with chronic TBI and two post-stroke patients.
Each of the participants was guided by the goal change during sessions in which it was present. For each participant, this was indicated by a significant upward trend with word number (i.e. time-point within the session) in the maximum extension distance used to select a letter in the word. This trend was not observed during sessions with no goal change. In addition, each participant showed a clinically important improvement on a functional test administered by an independent occupational therapist. Practice in our robotic rehabilitation environment transferred to improvements on functional pinching tasks.
The maximum finger span required to play the hangman game was initially set to 80% of the maximum finger span measured during calibration at the beginning of each session. On the basis of consultation with an occupational therapist, this point was expected to be a comfortable maximum for the participant. However, each participant followed the goal change to much larger finger spans. In addition, the participant’s maximum finger span in the absence of visual feedback was measured after each word in the game of hangman, and each participant’s performance in the absence of visual feedback was larger than the starting goal of the hangman game. These results suggest that a participants’ performance during calibration may not represent the extent of their ability and that performance during calibration should be used with caution when determining therapy goals. Assigned goals that advance with time and visual distortion are ways to address this problem of under-performance during calibration.
Effect of implicit and explicit goal change on motor performance
The results of the experiments described above indicate that distortion can be used to influence performance in a robotic environment, either by drawing the individual’s attention to a particular component of the task or by encouraging an increase in the force or range of motion used a task.
These experiments led us to reflect on the interaction of assigned goals with distortion. For example, in the preliminary rehabilitation protocol, the effects of assigning a goal to the individual are difficult to separate from the effects of the change in that goal. In addition, few differences between implicit and explicit goal change could be distinguished. To address these issues, we designed a new experimental paradigm. Our preliminary results with this paradigm are presented in the paragraphs below.
In this experiment, we used a PHANTOM Premium 1.5 HF haptic robot. Our outcome variable was the distance that the participant moved the robot endpoint within the specified period. As the participants moved the endpoint of the robot, they experienced a velocity-dependent, force-resisting movement (viscosity). This force made the task more difficult, expanding the range of possible distance outcomes for the task and enabling us to more precisely determine the effects of goals and goal change on motor performance. Each participant completed the target task while receiving various types of visual feedback, as outlined below.
Each participant completed 200 trials over approximately 45 minutes. The experiment began with a block of 80 trials to allow the participant to learn the target task. During this block of trials, participant the instructed to move the endpoint of the robot as far as possible from right to left; performance was shown on a visual-feedback bar, the length of which was mapped to the distance moved. After this initial block of trials, the participant completed a block of trials in each of four visual feedback conditions: no visual feedback (NoF); visual feedback with a static performance goal (NoC: visual goal); visual feedback with a performance goal that was explicitly changed (ExC: visual goal with explicit change); and visual feedback with a performance goal that was implicitly changed (ImpC). The order of these four blocks was randomized (visual group) or counterbalanced (kinesthetic group), and a break was given after each block.
In the NoF condition (no visual feedback), the only feedback that the participant received on the computer screen was a prompt to complete the target task.
In the NoC condition (visual feedback with a constant assigned goal), the participant was shown a white bar constituting a visual goal. As the participant moved the robot endpoint, this bar was shaded, in correspondence with the distance moved. The visual goal was held constant at 110% of the participant’s mean distance moved on trials 75–80 (the last five practice trials). We called this the static visual goal. A distance scale in millimeters was shown on the screen throughout the trial, and the distance moved was displayed after completion of each trial.
The visual display in the ExC condition was similar to that in the NoC condition, except that the visual goal was gradually increased. This change occurred over four steps, each increment corresponding to 6% of the baseline level established in the NoC condition. Increases in the visual goal were shaded purple to ensure that the user was aware of them. A participant completed five trials for each of the five visual goals (including baseline). The relationship between the distance moved by the participant and the visual distance shown on the screen remained unchanged as the goal increased.
In the ImpC condition, the user was shown a static visual goal consisting of a white bar on the screen, as in condition NoC. This goal remained constant, while the mapping between arm movement and the visual feedback bar was changed implicitly over trials. The first five trials were identical to the baseline trials in conditions NoC and ExC. After this point, the distance corresponding to the visual goal was implicitly increased in steps of 6%, without changing the visible goal itself. This means that the displayed distance corresponding to a user movement was re-scaled to the screen by a factor of 1/1.06, or 94.3% of its previous value, at each step. Thus, the user had to move farther to reach the visual goal, which was implicitly changed.
Preliminary results for this experiment were collected with 14 adults (over 18 years of age) without neurological disease or injury. Each participant performed the experiment with his or her dominant hand. Participants were divided into two groups: the vision group and the kinesthesis group. Each group had seven members. The two groups differed in the instructions they were given; in addition, the time limit for each trial was lower for the kinesthesis group. The vision group was told to focus on the visual feedback shown on the computer screen, and that the goal was to fill the white rectangle. These participants were told that when the white rectangle was incremented with a purple section (ExC condition), the goal was to fill the entire white-and-purple area. Members of the kinesthesis group were informed about the presence of the white-and-purple rectangle but were instructed that their primary goal was to move the robot arm as far as possible on each trial. Thus the two groups executed the same protocol, but the vision group attended to a visual goal, while the kinesthesis group attended to a kinesthetic goal. The experiment was approved by the institutional review board of the University of Pittsburgh, and each participant provided informed consent.
The 25 trials of each condition after baseline were divided into five blocks consisting of five trials each. These blocks corresponded to the five steps of goal change for the ExC and ImpC conditions. The mean maximum distance moved was computed for each block. To determine whether performance with visual goals differed from performance in their absence, we used a t-test to compare the last block of each condition to the mean baseline performance measured during the final five practice trials. This resulted in eight tests (four per group), so a Bonferroni correction was made for multiple testing; each p-value was accepted as significant if it was less than 0.05/8.
For each group, we performed a two-way repeated-measures analysis of variance (ANOVA) with the within-participants factors of block (1–5) and condition (NoF, NoC, ExC, ImpC). For main effects and interactions where the assumption of sphericity was not met, we report the p value corresponding to the Huynh-Feldt correction. This correction was chosen over the more conservative Greenhouse-Geisser option due to the preliminary nature of this study. A significant interaction was followed up using linear trend analysis to test for an increase in maximum distance with block for each condition for each group (four tests). A Bonferroni correction was made for multiple testing; each p was accepted as significant if it was <0.05/4.
The results for the vision group are shown in Figure 3, and the results for the kinesthesis group are in Figure 4. For comparisons between baseline performance and the last block in each of the four visual feedback conditions, we found no significant difference for the kinesthesis group. For the vision group, we found significant differences between baseline and the last block of NoC (t=−4.14, p=0.006), ExC (t=−5.10, p=0.002), and ImpC (t=−5.33, p=0.002). There was no significant difference between baseline performance and the last block of NoF (t=−1.30, p=0.24).
Results of the ANOVA analyses indicated that for the kinesthesis group, there was no main effect of block (F[1.64, 9.82]=0.47, p=0.60) or condition (F[3, 18]=2.51, p=0.092), and there was no significant interaction of block and condition (F[9.47, 56.82], p=0.90). For the vision group, on the other hand, there was a significant main effect for block (F[1.67, 10.00], p<0.001), but not for condition (F[1.51, 9.08]=4.31, p=0.056). There was a significant interaction of block and condition (F[6.18, 37.09], p=0.014). Follow-up tests were completed only for the vision group. The linear trend in the maximum distance as a function of trial block was not significant for the NoF or NoC conditions (p=0.13 and 0.11 respectively), but was significant for the ExC condition (p=0.003) and the ImpC condition (p<0.001).
Our previous work indicates that in our robotic environment, vision dominates kinesthesis. When given both visual and kinesthetic cues, individuals tend to focus primarily upon the visual feedback. Here, the rectangle on the screen provides a precise visual representation of distance moved. In addition, sight of the hand is available and may contribute to position estimation.34 In contrast, kinesthesis is known to contribute relatively weak cues to hand position.35,36 Visual dominance would be even more pronounced in a rehabilitation paradigm that uses visual feedback to transform repetitive practice into an engaging game. In this situation, the visual feedback would present the user with the tasks and goals of the game and would thus be the focus of the user’s attention.
These data were primarily collected with individuals without brain injury in order to acquire foundational knowledge about the effects of goal change in a robotic environment. In the future, this basic knowledge can be applied to optimize rehabilitation for individuals with brain injury. A large body of evidence for learned non-use indicates that the habitual performance of individuals with a brain injury may not be an accurate representation of their actual ability. We observed this phenomenon in our preliminary work with individuals with stroke and TBI. Learned non-use can affect performance in a robotic environment, and this must be considered when designing protocols for robotic rehabilitation. An individual can be encouraged to improve his or her performance through the use of assigned goals. Because the limits of an individual’s ability are not always known, goals that are incremented with time may be a good way in which to encourage individuals to improve their performance. This work shows that incremental goal change can affect motor performance in a robotic environment when an individual attends primarily to a visual goal. In particular, individuals perform better when the goal changes than they do in the presence of a static goal, even a static goal designed to be challenging. This phenomenon does not occur when the individual attends primarily to a kinesthetic goal. However, as discussed above, individuals are most likely to focus on the visual goal in a robotic rehabilitation environment.
In the preliminary results described here, implicit and explicit goal change had similar results. This may be because this experiment included only participants without known neurological impairment. For individuals with brain injury, implicit goal change may help them overcome any reluctance to move beyond habitual limits. This work provides a foundation with which to compare future results with individuals with stroke, TBI, and CP. While implicit goal change may be useful for each of these groups, there also likely to be differences in their response to goal change in a robotic environment, and the investigation of these differences would be valuable.
The concepts discussed here were conceived for application to rehabilitation of individuals with stroke and TBI. Extending the range of potential users to include individuals with CP requires special considerations. Individuals with stroke and traumatic brain injury most often have normal movement patterns before the injury. Because CP affects individuals from birth or shortly thereafter, these individuals may never develop ‘typical’ movement patterns. Thus, rehabilitation for stroke and TBI is often viewed as relearning previous patterns of movement, while individuals with CP must often try to learn entirely new patterns of movement. This may mean that goal change must occur in smaller increments for individuals with CP. Implicit goal change, or distortion, may be necessary to address reluctance to move beyond habitual limits. Future work is needed to investigate these and other issues in order to incorporate appropriate and effective goal change into robotic rehabilitation for CP.