A computational model of human-robot load sharing during robot-assisted arm movement training after stroke

An important goal in robot-assisted movement therapy after neurologic injury is to provide an optimal amount of mechanical assistance to patients as they complete motor tasks. This paper presents a computational model of how humans interact with robotic therapy devices for the task of lifting a load to a desired height. The model predicts that an adaptive robotic therapy device will take over performance of the lifting task if the human motor control system contains a slacking term (i.e. a term that tries to the reduce force output of the arm when error is small) but the robot does not. We present experimental data from people with a chronic stroke as they train with a robotic arm orthosis that confirms this prediction. We also show that incorporating a slacking term into the robot overcomes this problem, increasing load sharing by the patient while still keeping kinematic errors small. These results provide insight into the computational mechanisms of human motor adaptation during rehabilitation therapy, and provide a framework for optimizing robot-assisted therapy.

Feasibility of manual teach-and-replay and continuous impedance shaping for robotic locomotor training following spinal cord injury

Robotic gait training is an emerging technique for retraining walking ability following spinal cord injury (SCI). A key challenge in this training is determining an appropriate stepping trajectory and level of assistance for each patient, since patients have a wide range of sizes and impairment levels. Here, we demonstrate how a lightweight yet powerful robot can record subject-specific, trainer-induced leg trajectories during manually assisted stepping, then immediately replay those trajectories. Replay of the subject-specific trajectories reduced the effort required by the trainer during manual assistance, yet still generated similar patterns of muscle activation for six subjects with a chronic SCI. We also demonstrate how the impedance of the robot can be adjusted on a step-by-step basis with an error-based, learning law. This impedance-shaping algorithm adapted the robot's impedance so that the robot assisted only in the regions of the step trajectory where the subject consistently exhibited errors. The result was that the subjects stepped with greater variability, while still maintaining a physiologic gait pattern. These results are further steps toward tailoring robotic gait training to the needs of individual patients.

Quantification of therapists' manual assistance on the leg during treadmill gait training with partial body-weight support after spinal cord injury

We are studying the forces applied by therapists when assisting on the legs of patients with spinal cord injury (SCI) during body-weight supported treadmill training (BWSTT). This paper presents results from three experiments with three patients and six different therapists. We attached a sensorized orthosis to one of the legs of the patients and measured the forces exerted by the therapists when assisting on that leg through the orthosis. We also measured the kinematics of the shank. The hips and opposite leg were assisted by two other therapists as usual. We found a high repeatability across steps for the forces applied by each therapist. Better-skilled therapists elicited significantly better leg extension in stance phase by exerting significantly larger horizontal forces on the knee during stance. Better-skilled therapists also elicited better toe clearance during initial-to-mid swing, with fewer or no episodes of toe dragging. This better toe clearance was not generally correlated with larger therapist push-up forces at the ankle. We did find, however, that better toe clearance in swing was correlated with better extension of the same leg in stance, suggesting that a normative leg extension in stance contributed to better muscle activity of the same leg in swing. These results provide insight into therapist-assisted gait training, which may be useful for optimizing robotic gait-training devices.

A robot and control algorithm that can synchronously assist in naturalistic motion during body-weight-supported gait training following neurologic injury

Locomotor training using body weight support on a treadmill and manual assistance is a promising rehabilitation technique following neurological injuries, such as spinal cord injury (SCI) and stroke. Previous robots that automate this technique impose constraints on naturalistic walking due to their kinematic structure, and are typically operated in a stiff mode, limiting the ability of the patient or human trainer to influence the stepping pattern. We developed a pneumatic gait training robot that allows for a full range of natural motion of the legs and pelvis during treadmill walking, and provides compliant assistance. However, we observed an unexpected consequence of the device's compliance: unimpaired and SCI individuals invariably began walking out-of-phase with the device. Thus, the robot perturbed rather than assisted stepping. To address this problem, we developed a novel algorithm that synchronizes the device in real-time to the actual motion of the individual by sensing the state error and adjusting the replay timing to reduce this error. This paper describes data from experiments with individuals with SCI that demonstrate the effectiveness of the synchronization algorithm, and the potential of the device for relieving the trainers of strenuous work while maintaining naturalistic stepping.

Tools for understanding and optimizing robotic gait training

This article reviews several tools we have developed to improve the understanding of locomotor training following spinal cord injury (SCI), with a view toward implementing locomotor training with robotic devices. We have developed (1) a small-scale robotic device that allows testing of locomotor training techniques in rodent models, (2) an instrumentation system that measures the forces and motions used by experienced human therapists as they manually assist leg movement during locomotor training, (3) a powerful, lightweight leg robot that allows investigation of motor adaptation during stepping in response to force-field perturbations, and (4) computational models for locomotor training. Results from the initial use of these tools suggest that an optimal gait-training robot will minimize disruptive sensory input, facilitate appropriate sensory input and gait mechanics, and intelligently grade and time its assistance. Currently, we are developing a pneumatic robot designed to meet these specifications as it assists leg and pelvic motion of people with SCI.

Robot-assisted movement training for the stroke-impaired arm: Does it matter what the robot does?

Robot-assisted movement training improves arm movement ability following acute and chronic stroke. Such training involves two interacting processes: the patient trying to move and the robot applying forces to the patient's arm. A fundamental principle of motor learning is that movement practice improves motor function; the role of applied robotic forces in improving motor function is still unclear. This article reviews our work addressing this question. Our pilot study using the Assisted Rehabilitation and Measurement (ARM) Guide, a linear robotic trainer, found that mechanically assisted reaching improved motor recovery similar to unassisted reaching practice. This finding is inconclusive because of the small sample size (n = 19), but suggest that future studies should carefully control the amount of voluntary movement practice delivered to justify the use of robotic forces. We are optimistic that robotic forces will ultimately show additional therapeutic benefits when coupled with movement practice. We justify this optimism here by comparing results from the ARM Guide and the Mirror Image Movement Enabler robotic trainer. This comparison suggests that requiring a patient to generate specific patterns of force before allowing movement is more effective than mechanically completing movements for the patient. We describe the engineering implementation of this "guided-force training" algorithm.

Locomotor ability in spinal rats is dependent on the amount of activity imposed on the hindlimbs during treadmill training

Studies have shown that treadmill training with body weight support is effective for enhancing locomotor recovery following a complete spinal cord transection (ST) in animals. However, there have been no studies that have investigated the extent that functional recovery in ST animals is dependent on the amount of activity imposed on the hindlimbs during training. In rats transected as neonates (P5), we used a robotic device to impose either a high or a low amount of hindlimb activity during treadmill training starting 23 days after transection. The rats were trained 5 days per week for 4 weeks. One group (n = 13) received 1000 steps/training session and a second group (n = 13) received 100 steps/training session. During training, the robotic device imposed the maximum amount of weight that each rat could bear on the hindlimbs, and counted the number of stepping movements during each session. After 4 weeks of training, the number of steps performed during treadmill testing was not significantly different between the two groups. However, the quality of stepping in the group that received 1000 steps/training session improved over a range of levels of weight bearing on the hindlimbs and at different treadmill speeds. In contrast, little improvement in the quality of stepping was observed in the group that received only 100 steps/training session. These findings indicate that the ability of the lumbar spinal cord to adjust to load- and speed-related sensory stimuli associated with stepping is dependent on the number of repetitions of the same activity that is imposed on the spinal circuits during treadmill training.

Robotic assessment of locomotor recovery in spinal contused rats

The purpose of this study was to investigate the ability of a robotic device, "the rat stepper", to assess intrinsic locomotor recovery following spinal cord contusion injury in adult rats. The device consists of a motorized body weight support mechanism that precisely controls the load to the hindlimbs during stepping, and two small robotic arms that measure and manipulate hindlimb movement. Sixteen rats received a contusion injury to the mid thoracic spinal cord with different severity levels (mild, moderate, severe, and sham). The animals were then evaluated weekly using the rat stepper, beginning one week after injury and continuing for a period of twelve weeks, across a range of body weight support levels. The contused animals demonstrated recovery in a standard locomotor assessment score (the BBB score), with most of the recovery occurring by four weeks post injury. We analyzed fourteen robotic measures of stepping and found that the measures that were most sensitive to intrinsic recovery were step velocity and inter limb coordination. These measures were also significantly correlated with the BBB score. The number of steps taken during testing was not sensitive to intrinsic recovery, nor correlated to the BBB score. These results suggest that step quality, rather than quantity, best reflects recovery after contusion injury in adult, untrained rats. Thus, robotic motion capture of only a few steps can provide a sensitive, valid measure of locomotor recovery after contusion.

Adaptive assistance for guided force training in chronic stroke

This work describes a novel form of robotic therapy for the upper extremity in chronic stroke. Based on previous results, we hypothesized that a training task that encourages subjects to consciously guide endpoint forces generated by the hemiparetic arm will result in significant gains in functional ability of the arm, superior to more conventional methods of therapy. In addition, since stroke survivors present with varying degrees of arm movement ability, we developed an adaptive algorithm that tailors the amount of assistance provided in completing the guided force training task. The algorithm adapts a coefficient for velocity-dependent assistance based on measured movement speed, on a trial-to-trial basis. The training algorithm has been implemented with a simple linear robotic device called the ARM Guide. One participant completed a two month training program with the adaptive algorithm, resulting in significant improvements in the performance of functional tasks.

Motor Adaptation as a Greedy Optimization of Error and Effort

Motor adaptation to a novel dynamic environment is primarily thought of as a process in which the nervous system learns to anticipate the environmental forces to eliminate kinematic error. Here we show that motor adaptation can more generally be modeled as a process in which the motor system greedily minimizes a cost function that is the weighted sum of kinematic error and effort. The learning dynamics predicted by this minimization process are a linear, auto-regressive equation with only one state, which has been identified previously as providing a good fit to data from force-field-type experiments. Thus we provide a new theoretical result that shows how these previously identified learning dynamics can be viewed as arising from an optimization of error and effort. We also show that the coefficients of the learning dynamics must fall within a specific range for the optimization model to be valid and verify with experimental data from walking in a force field that they indeed fall in this range. Finally, we attempted to falsify the model by performing experiments in two conditions (repeated exposure to a force field, exposure to force fields of different strengths) for which the single-state, auto-regressive equation might be expected to not fit the data well. We found however that the equation adequately captured the pattern of errors and thus conclude that motor adaptation to a force field can be approximated as an optimization of effort and error for a range of experimental conditions.

Human-robot cooperative movement training: learning a novel sensory motor transformation during walking with robotic assistance-as-needed

Background

A prevailing paradigm of physical rehabilitation following neurologic injury is to "assist-as-needed" in completing desired movements. Several research groups are attempting to automate this principle with robotic movement training devices and patient cooperative algorithms that encourage voluntary participation. These attempts are currently not based on computational models of motor learning.

Methods

Here we assume that motor recovery from a neurologic injury can be modelled as a process of learning a novel sensory motor transformation, which allows us to study a simplified experimental protocol amenable to mathematical description. Specifically, we use a robotic force field paradigm to impose a virtual impairment on the left leg of unimpaired subjects walking on a treadmill. We then derive an "assist-as-needed" robotic training algorithm to help subjects overcome the virtual impairment and walk normally. The problem is posed as an optimization of performance error and robotic assistance. The optimal robotic movement trainer becomes an error-based controller with a forgetting factor that bounds kinematic errors while systematically reducing its assistance when those errors are small. As humans have a natural range of movement variability, we introduce an error weighting function that causes the robotic trainer to disregard this variability.

Results

We experimentally validated the controller with ten unimpaired subjects by demonstrating how it helped the subjects learn the novel sensory motor transformation necessary to counteract the virtual impairment, while also preventing them from experiencing large kinematic errors. The addition of the error weighting function allowed the robot assistance to fade to zero even though the subjects' movements were variable. We also show that in order to assist-as-needed, the robot must relax its assistance at a rate faster than that of the learning human.

Conclusion

The assist-as-needed algorithm proposed here can limit error during the learning of a dynamic motor task. The algorithm encourages learning by decreasing its assistance as a function of the ongoing progression of movement error. This type of algorithm is well suited for helping people learn dynamic tasks for which large kinematic errors are dangerous or discouraging, and thus may prove useful for robot-assisted movement training of walking or reaching following neurologic injury.

Accelerating motor adaptation by influencing neural computations

When people learn to reach or step in a novel dynamic environment, they initially exhibit a large trajectory error, which they gradually reduce with practice. The error evolution is well modeled by a process in which the motor command on the next movement is adjusted in proportion to the previous movement's trajectory error. We hypothesized that we could accelerate motor adaptation by transiently increasing trajectory error. We tested this hypothesis by quantifying adaptation to a viscous force field applied during the swing phase of stepping in two conditions. In the first condition, we applied then removed the field for 75 steps each, for four iterations. Subjects adapted to each field exposure with a mean time constant of 3.4 steps. In the second condition, we repeated this experiment, but increased the strength of the field for only the first step in each field exposure. We predicted the field strength increase needed by solving a finite difference equation that described the error evolution. Adaptation was significantly faster when the field was transiently amplified (mean time constant = 2 trials). These results demonstrate that it is possible to increase the rate of adaptation to a novel dynamic environment based on knowledge of the computational mechanisms that underlie adaptation.

Robot-assisted hindlimb extension increases the probability of swing initiation during treadmill walking by spinal cord contused rats

Training and evaluation of locomotion in animals with spinal cord injury will likely be improved with the development of techniques that increase stepping activity. We hypothesized that robot-assisted extension of the hindlimbs of spinal cord injured rats during stance would increase the probability that the swing phase of gait would be initiated. Thirty-three adult, Sprague-Dawley rats received a contusion injury to the mid-thoracic spinal cord. The animals' hindlimbs were pulled into extension using small robotic arms to pull at the ankle, as the rat stepped on either a reciprocating, robotic paw platform or a conventional treadmill belt. The animals demonstrated an increase in the probability of swing initiation with spontaneous recovery during the first 4 weeks following injury (p < 0.0001). The probability of swing initiation was found to be greater with the use of robot-assisted extension versus no extension force at the ankle (75+/-16.9% versus 38.9+/-16.6%, p < 0.001). Swing initiation occurred most frequently at a particular distance of hindlimb extension (50 mm caudal from the neutral position at stance), but was broadly tuned about this distance. These results indicate that a greater amount of stepping activity can be elicited by robot-assisted extension, thereby providing possible benefits to evaluation and training of gait following SCI.

Motor adaptation to a small force field superimposed on a large background force

The human motor system adapts to novel force field perturbations during reaching by forming an internal model of the external dynamics and by modulating arm impedance. We studied whether it uses similar strategies when the perturbation is superimposed on a much larger background force. Consistent with the Weber-Fechner law for force perception, subjects had greater difficulty consciously perceiving the force field perturbation when it was superimposed on the large background force. However, they still adapted to the perturbation, decreasing trajectory distortion with repeated reaching and demonstrating kinematic after effects when the perturbation was unexpectedly removed. They also adapted by increasing their arm impedance when the background force was not present, but did not vary the arm impedance when the background force was present. The identified parameters of a previously proposed mathematical model of motor adaptation changed significantly with the presence of the background force. These results indicate that the motor system maintains its sensitivity for internal model formation even when there are large background forces that mask perception. Further, the motor system modulates arm impedance differently in response to the same perturbation depending on the background force onto which that perturbation is superimposed. Finally, these results suggest that computational models of motor adaptation will likely need to include force-dependent parameters to accurately predict errors.

Automating arm movement training following severe stroke: functional exercises with quantitative feedback in a gravity-reduced environment

An important goal in rehabilitation engineering is to develop technology that allows individuals with severe motor impairment to practice arm movement without continuous supervision from a rehabilitation therapist. This paper describes the development of such a system, called Therapy WREX or ("T-WREX"). The system consists of an orthosis that assists in arm movement across a large workspace, a grip sensor that detects hand grip pressure, and software that simulates functional activities. The arm orthosis is an instrumented, adult-sized version of the Wilmington Robotic Exoskeleton (WREX), which is a five degrees-of-freedom mechanism that passively counterbalances the weight of the arm using elastic bands. After providing a detailed design description of T-WREX, this paper describes two pilot studies of the system's capabilities. The first study demonstrated that individuals with chronic stroke whose arm function is compromised in a normal gravity environment can perform reaching and drawing movements while using T-WREX. The second study demonstrated that exercising the affected arm of five people with chronic stroke with T-WREX over an eight week period improved unassisted movement ability (mean change in Fugl-Meyer score was 5 points +/- 2 SD; mean change in range of motion of reaching was 10%, p < 0.001). These results demonstrate the feasibility of automating upper-extremity rehabilitation therapy for people with severe stroke using passive gravity assistance, a grip sensor, and simple virtual reality software.

Learning to perform a new movement with robotic assistance: comparison of haptic guidance and visual demonstration

BACKGROUND

Mechanical guidance with a robotic device is a candidate technique for teaching people desired movement patterns during motor rehabilitation, surgery, and sports training, but it is unclear how effective this approach is as compared to visual demonstration alone. Further, little is known about motor learning and retention involved with either robot-mediated mechanical guidance or visual demonstration alone.

METHODS

Healthy subjects (n = 20) attempted to reproduce a novel three-dimensional path after practicing it with mechanical guidance from a robot. Subjects viewed their arm as the robot guided it, so this "haptic guidance" training condition provided both somatosensory and visual input. Learning was compared to reproducing the movement following only visual observation of the robot moving along the path, with the hand in the lap (the "visual demonstration" training condition). Retention was assessed periodically by instructing the subjects to reproduce the path without robotic demonstration.

RESULTS

Subjects improved in ability to reproduce the path following practice in the haptic guidance or visual demonstration training conditions, as evidenced by a 30-40% decrease in spatial error across 126 movement attempts in each condition. Performance gains were not significantly different between the two techniques, but there was a nearly significant trend for the visual demonstration condition to be better than the haptic guidance condition (p = 0.09). The 95% confidence interval of the mean difference between the techniques was at most 25% of the absolute error in the last cycle. When asked to reproduce the path repeatedly following either training condition, the subjects' performance degraded significantly over the course of a few trials. The tracing errors were not random, but instead were consistent with a systematic evolution toward another path, as if being drawn to an "attractor path".

CONCLUSION

These results indicate that both forms of robotic demonstration can improve short-term performance of a novel desired path. The availability of both haptic and visual input during the haptic guidance condition did not significantly improve performance compared to visual input alone in the visual demonstration condition. Further, the motor system is inclined to repeat its previous mistakes following just a few movements without robotic demonstration, but these systematic errors can be reduced with periodic training.

Robotic gait analysis of bipedal treadmill stepping by spinal contused rats: characterization of intrinsic recovery and comparison with BBB

There is a critical need to develop objective, quantitative techniques to assess motor function after spinal cord injury. Here, we assess the ability of a recently developed robotic device (the "rat stepper") to characterize locomotor impairment following contusion injury in rats. In particular, we analyzed how the kinematic features of hindlimb movement during bipedal, weight-supported treadmill stepping change following contusion, and whether these changes correlate with the recovery of open field locomotion. Female, Sprague-Dawley rats (n=29, 8 weeks of age) received mid thoracic contusion injuries of differing severities (11 mild, nine moderate, nine severe, and four sham). In a first experiment, 16 of the animals were evaluated weekly for 12 weeks using the robotic stepping device. In a second experiment, 17 of the animals were evaluated every other day for 4 weeks. The contused animals recovered open field locomotion based on the Basso, Beattie, and Bresnahan Scale (BBB) analysis, with most of the recovery occurring by 4 weeks post-injury. Analysis of 14 robotic measures of stepping revealed that several measures improved significantly during the same 4 weeks: swing velocity, step height, step length, hindlimb coordination, and the ability to support body weight. These measures were also significantly correlated with the BBB score. The number of steps taken during testing was not directly related to intrinsic recovery or correlated to the BBB score. These results suggest that it is the quality of weight-supported steps, rather than the quantity, that best reflects locomotor recovery after contusion injury, and that the quality of these steps is determined by the integrity of extensor, flexor, and bilateral coordination pathways. Thus, by measuring only a few weight-supported steps with motion capture, a sensitive, valid measure of locomotor recovery following contusion injury can be obtained across a broad range of impairment levels.

Robot-assisted reaching exercise promotes arm movement recovery in chronic hemiparetic stroke: a randomized controlled pilot study

Background and purpose

Providing active assistance to complete desired arm movements is a common technique in upper extremity rehabilitation after stroke. Such active assistance may improve recovery by affecting somatosensory input, motor planning, spasticity or soft tissue properties, but it is labor intensive and has not been validated in controlled trials. The purpose of this study was to investigate the effects of robotically administered active-assistive exercise and compare those with free reaching voluntary exercise in improving arm movement ability after chronic stroke.

Methods

Nineteen individuals at least one year post-stroke were randomized into one of two groups. One group performed 24 sessions of active-assistive reaching exercise with a simple robotic device, while a second group performed a task-matched amount of unassisted reaching. The main outcome measures were range and speed of supported arm movement, range, straightness and smoothness of unsupported reaching, and the Rancho Los Amigos Functional Test of Upper Extremity Function.

Results and discussion

There were significant improvements with training for range of motion and velocity of supported reaching, straightness of unsupported reaching, and functional movement ability. These improvements were not significantly different between the two training groups. The group that performed unassisted reaching exercise improved the smoothness of their reaching movements more than the robot-assisted group.

Conclusion

Improvements with both forms of exercise confirmed that repeated, task-related voluntary activation of the damaged motor system is a key stimulus to motor recovery following chronic stroke. Robotically assisting in reaching successfully improved arm movement ability, although it did not provide any detectable, additional value beyond the movement practice that occurred concurrently with it. The inability to detect any additional value of robot-assisted reaching may have been due to this pilot study's limited sample size, the specific diagnoses of the participants, or the inclusion of only individuals with chronic stroke.

Spinal cord-transected mice learn to step in response to quipazine treatment and robotic training

In the present study, concurrent treatment with robotic step training and a serotonin agonist, quipazine, generated significant recovery of locomotor function in complete spinal cord-transected mice (T7-T9) that otherwise could not step. The extent of recovery achieved when these treatments were combined exceeded that obtained when either treatment was applied independently. We quantitatively analyzed the stepping characteristics of spinal mice after alternatively administering no training, manual training, robotic training, quipazine treatment, or a combination of robotic training with quipazine treatment, to examine the mechanisms by which training and quipazine treatment promote functional recovery. Using fast Fourier transform and principal components analysis, significant improvements in the step rhythm, step shape consistency, and number of weight-bearing steps were observed in robotically trained compared with manually trained or nontrained mice. In contrast, manual training had no effect on stepping performance, yielding no improvement compared with nontrained mice. Daily bolus quipazine treatment acutely improved the step shape consistency and number of steps executed by both robotically trained and nontrained mice, but these improvements did not persist after quipazine was withdrawn. At the dosage used (0.5 mg/kg body weight), quipazine appeared to facilitate, rather than directly generate, stepping, by enabling the spinal cord neural circuitry to process specific patterns of sensory information associated with weight-bearing stepping. Via this mechanism, quipazine treatment enhanced kinematically appropriate robotic training. When administered intermittently during an extended period of robotic training, quipazine revealed training-induced stepping improvements that were masked in the absence of the pharmacological treatment.

A robotic device for studying rodent locomotion after spinal cord injury

We have developed a robotic device (the "rat stepper") for evaluating and training locomotor function of spinal cord injured rodents. This paper provides a detailed description of the device design and a characterization of its robotic performance capabilities.

Impaired motor control in patients with benign focal epilepsy of childhood

Upper extremity motor function was quantitatively assessed in 6 children (age 7-11 years) treated with antiepileptic drugs for benign focal epilepsies of childhood and compared with that of 30 age-matched normal children. Both motor performance and adaptation to perturbing mechanical constraints imposed by a robotic device were significantly impaired in children with benign focal epilepsies of childhood. Our findings thus question whether certain "benign" epilepsies are truly benign and whether pharmacologic treatment might contribute to motor impairment.

Effect of muscle fatigue on internal model formation and retention during reaching with the arm

The motor system adapts to novel dynamic environments by forming internal models that predict the muscle forces needed to move skillfully. The goal of this study was to determine how muscle fatigue affects internal model formation during arm movement and whether an internal model acquired while fatigued could be recalled accurately after rest. Twelve subjects adapted to a viscous force field applied by a lightweight robot as they reached to a target. They then reached while being resisted by elastic bands until they could no longer touch the target. This protocol reduced the strength of the muscles used to resist the force field by ∼20%. The bands were removed, and subjects adapted again to the viscous force field. Their adaptive ability, quantified by the amount and time constant of adaptation, was not significantly impaired following fatigue. The subjects then rested, recovering ∼70% of their lost force-generation ability. When they reached in the force field again, their prediction of the force field strength was different than in a nonfatigued state. This alteration was consistent with the use of a higher level of effort than normally used to counteract the force field. These results suggest that recovery from fatigue can affect recall of an internal model, even when the fatigue did not substantially affect the motor system’s ability to form the model. Recovery from fatigue apparently affects recall because the motor system represents internal models as a mapping between effort and movement and relies on practice to recalibrate this mapping.

Control of a pneumatic orthosis for upper extremity stroke rehabilitation

A key challenge in rehabilitation robotics is the development of a lightweight, large force, high degrees-of-freedom device that can assist in functional rehabilitation of the arm. Pneumatic actuators can potentially help meet this challenge because of their high power-to-weight ratio. They are currently not widely used for rehabilitation robotics because they are difficult to control. This paper describes the control development of a pneumatically actuated, upper extremity orthosis for rehabilitation after stroke. To provide the sensing needed for good pneumatic control, position and velocity of the robot are estimated by a unique implementation of a Kalman filter using MEMS accelerometers. To compensate for the nonlinear behavior of the pneumatic servovalves, force control is achieved using a new method for air flow mapping using experimentally measured data in a least-squares regression. To help patients move with an inherently compliant robot, a high level controller that assists only as needed in reaching exercises is developed. This high level controller differs from traditional trajectory-based, position controllers, allowing free voluntary movements toward a target while resisting movements away from the target. When the target cannot be reached voluntarily, the controller slowly builds up force, pushing the arm toward the target. As each target position is reached, the controller builds an internal model of the subject's capability, learning the forces necessary to complete movements. Preliminary testing performed on a non-disabled subject demonstrated the ability of the orthosis to complete reaching movements with graded assistance and to adapt to the effort level of the subject. Thus, the orthosis is a promising tool for upper extremity rehabilitation after stroke.

Dynamic motion planning for the design of robotic gait rehabilitation

In this paper we examine a method to control the stepping motion of a paralyzed person suspended over a treadmill using a robot attached to the pelvis. A leg swing motion is created by moving the pelvis without contact with the legs. The problem is formulated as an optimal control problem for an underactuated articulated chain. The optimal control problem is converted into a discrete parameter optimization and an efficient gradient-based algorithm is used to solve it. Motion capture data from an unimpaired human subject is compared to the simulation results from the dynamic motion optimization. Our results suggest that it is feasible to drive repetitive stepping on a treadmill by a paralyzed person by assisting in torso movement alone. The optimized, pelvic motion strategies are comparable to "hip-hiking" gait strategies used by people with lower limb prostheses or hemiparesis. The resulting motions can be found at the web site http://ww.eng.uci.edu/-chwang/project/stepper/stepper.html.

Robot-enhanced motor learning: accelerating internal model formation during locomotion by transient dynamic amplification

When adapting to novel dynamic environments the nervous system learns to anticipate the imposed forces by forming an internal model of the environmental dynamics in a process driven by movement error reduction. Here, we tested the hypothesis that motor learning could be accelerated by transiently amplifying the environmental dynamics. A novel dynamic environment was created during treadmill stepping by applying a perpendicular viscous force field to the leg through a robotic device. The environmental dynamics were amplified by an amount determined by a computational learning model fit on a per-subject basis. On average, subjects significantly reduced the time required to predict the applied force field by approximately 26% when the field was transiently amplified. However, this reduction was not as great as that predicted by the model, likely due to nonstationarities in the learning parameters. We conclude that motor learning of a novel dynamic environment can be accelerated by exploiting the error-based learning mechanism of internal model formation, but that nonlinearities in adaptive response may limit the feasible acceleration. These results support an approach to movement training devices that amplify rather than reduce movement errors, and provide a computational framework for both implementing the approach and understanding its limitations.

Hindlimb loading determines stepping quantity and quality following spinal cord transection

We compared the bipedal hindlimb stepping ability of untrained and trained (step-trained 6 min/day) spinal rats (mid-thoracic spinal cord transection at post-natal day 5) at different levels of body weight support on a treadmill over a 40-day period, starting at 69 days of age. A robotic device provided precise levels of body weight support and recorded hindlimb movement. We assessed stepping ability using: (1) step quantity determined from the measured hindlimb movement, (2) ordinal scales of paw placement, weight-bearing, and limb flexion, and (3) the lowest level of body weight support at which stepping was maintained. Stepping quantity and quality depended strongly on the level of support provided. Stepping ability improved with time, but only at the higher levels of weight-bearing, and independently of training. Increasing limb loading by gradually decreasing body weight support altered the spatiotemporal properties of the steps, resulting in an increase in step length and stance duration and a decrease in swing and step cycle duration. The rats progressively improved their ability to support more load before collapsing from a maximum of about 42 g ( approximately 25% of body weight) at Day 1 to 73 g ( approximately 35% of body weight) at Day 40. We conclude that the level of hindlimb loading provided to a spinally transected rat strongly influences the quantity and quality of stepping. Furthermore, the relationship between stepping ability and loading conditions changes with time after spinal cord transection and is unaltered by small amounts of step training. Finally, load-bearing failure point can be a quantitative measure of locomotor recovery following spinal cord injury, especially for severely impaired animals that cannot step unassisted.

Robotics, motor learning, and neurologic recovery

Robotic devices are helping shed light on human motor control in health and injury. By using robots to apply novel force fields to the arm, investigators are gaining insight into how the nervous system models its external dynamic environment. The nervous system builds internal models gradually by experience and uses them in combination with impedance and feedback control strategies. Internal models are robust to environmental and neural noise, generalized across space, implemented in multiple brain regions, and developed in childhood. Robots are also being used to assist in repetitive movement practice following neurologic injury, providing insight into movement recovery. Robots can haptically assess sensorimotor performance, administer training, quantify amount of training, and improve motor recovery. In addition to providing insight into motor control, robotic paradigms may eventually enhance motor learning and rehabilitation beyond the levels possible with conventional training techniques.

Modeling Reaching Impairment After Stroke Using a Population Vector Model of Movement Control That Incorporates Neural Firing-Rate Variability

The directional control of reaching after stroke was simulated by including cell death and firing-rate noise in a population vector model of movement control. In this model, cortical activity was assumed to cause the hand to move in the direction of a population vector, defined by a summation of responses from neurons with cosine directional tuning. Two types of directional error were analyzed: the between-target variability, defined as the standard deviation of the directional error across a wide range of target directions, and the within-target variability, defined as the standard deviation of the directional error for many reaches to a single target.

Both between and within-target variability increased with increasing cell death. The increase in between-target variability arose because cell death caused a nonuniform distribution of preferred directions. The increase in within-target variability arose because the magnitude of the population vector decreased more quickly than its standard deviation for increasing cell death, provided appropriate levels of firing-rate noise were present. Comparisons to reaching data from 29 stroke subjects revealed similar increases in between and within-target variability as clinical impairment severity increased. Relationships between simulated cell death and impairment severity were derived using the between and within-target variability results. For both relationships, impairment severity increased similarly with decreasing percentage of surviving cells, consistent with results from previous imaging studies. These results demonstrate that a population vector model of movement control that incorporates cosine tuning, linear summation of unitary responses, firing-rate noise, and random cell death can account for some features of impaired arm movement after stroke.

Neuromotor noise limits motor performance, but not motor adaptation, in children

Children do not typically appear to move with the same skill and dexterity as adults, although they can still improve their motor performance in specific tasks with practice. One possible explanation is that their motor performance is limited by an inherently higher level of movement variability, but that their motor adaptive ability is robust to this variability. To test this hypothesis, we examined motor adaptation of 43 children (ages 6-17) and 12 adults as they reached while holding the tip of a lightweight robot. The robot applied either a predictable, velocity-dependent field (the "mean field") or a similar field that incorporated stochastic variation (the "noise field"), thereby further enhancing the variability of the subjects' movements. We found that children exhibited greater initial trial-to-trial variability in their unperturbed movements but were still able to adapt comparably to adults in both the mean and noise fields. Furthermore, the youngest children (ages 6-8) were able to reduce their variability with practice to levels comparable to the remaining children groups although not as low as adults. These results indicate that children as young as age 6 possess adult-like neural systems for motor adaptation and internal model formation that allow them to adapt to novel dynamic environments as well as adults on average despite increased neuromotor or environmental noise. Performance after adaptation is still more variable than adults, however, indicating that movement inconsistency, not motor adaptation inability, ultimately limits motor performance by children and may thus account for their appearance of incoordination and more frequent motor accidents (e.g., spilling, tripping). The results of this study also suggest that movement variability in young children may arise from two sources--a relatively constant, intrinsic source related to fundamental physiological constraints of the developing motor system and a more rapidly modifiable source that is modulated depending on the current motor context.

Hemiparetic stroke impairs anticipatory control of arm movement

Internal models are sensory motor mappings used by the nervous system to anticipate the force requirements of movement tasks. The ability to use internal models likely underlies the development of skillful control of the arm throughout life. It is currently unknown to what extent individuals with hemiparetic stroke can form and implement such internal models. To examine this issue, we measured whether such individuals could learn to anticipate forces applied to their arms by a lightweight robotic device as they practiced reaching to a target. Thirteen subjects with post-stroke hemiparesis were tested. Forces were applied to the arm, which curved the hand path in either the medial or lateral direction, as the subjects reached repeatedly towards a target located in front of them at their workspace boundary. The subjects exhibited a decreased ability to adapt to the perturbing forces with their hemiparetic arms. That is, they did not straighten their reaching path as well, compared to their ipsilesional arms, and they exhibited smaller aftereffects when the perturbing force was unexpectedly removed. The ability to adapt to the force improved significantly with decreasing impairment severity, as quantified using both clinical scales and quantitative strength measurements. Some subjects with strength reductions as severe as 60% were able to adapt to the fields, generating significant aftereffects. We conclude that hemiparetic stroke impairs the ability to implement internal models used for anticipatory control of arm movement, although even some severely weakened subjects retain at least a partial ability to form and use internal models. Finding ways to fully restore this adaptive ability, or to make use of what adaptive ability remains during rehabilitation, is an important goal for improving functional motor recovery.

Use of robotics in assessing the adaptive capacity of the rat lumbar spinal cord

We have developed a robotic device that can record the trajectory of the hindlimb movements in rats. The robotic device can also impose programmed forces on the limbs during stepping. In the present paper we describe experiments using this robotic device, i.e. the rat stepper, to determine whether step training improves the locomotor capacity of adult rats that received complete spinal cord transections as neonates. We also determined to what extent the locomotor patterns can be maintained when the step cycle is physically perturbed by the robotic device. The results of the present study demonstrate that a robotic device can be used effectively to quantify the improvements in the locomotor capacity of spinal transected rats that occurs over a period of step training. The present results also demonstrate that when an external force is imposed to disrupt the step cycle, the spinal cord has the neural control elements necessary to normalize the kinematics over a number of steps, in the face of the disrupted forces.

The rat lumbosacral spinal cord adapts to robotic loading applied during stance

Load-related afferent information modifies the magnitude and timing of hindlimb muscle activity during stepping in decerebrate animals and spinal cord-injured humans and animals, suggesting that the spinal cord mediates load-related locomotor responses. In this study, we found that stepping on a treadmill by adult rats that received complete, midthoracic spinal cord transections as neonates could be altered by loading the hindlimbs using a pair of small robotic arms. The robotic arms applied a downward force to the lower shanks of the hindlimbs during the stance phase and measured the position of the lower shank during stepping. No external force was applied during the swing phase of the step. When applied bilaterally, this stance force field perturbed the hindlimb trajectories so that the ankle position was shifted downward during stance. In response to this perturbation, both the stance and step cycle durations decreased. During swing, the hindlimb initially accelerated toward the normal, unperturbed swing trajectory and then tracked the normal trajectory. Bilateral loading increased the magnitude of the medial gastrocnemius electromyographic (EMG) burst during stance and increased the amplitude of the semitendinosus and rectus femoris EMG bursts. When the force field was applied unilaterally, stance duration decreased in the loaded hindlimb, while swing duration was decreased in the contralateral hindlimb, thereby preserving interlimb coordination. These results demonstrate the feasibility of using robotic devices to mechanically modulate afferent input to the injured spinal cord during weight-supported locomotion. In addition, these results indicate that the lumbosacral spinal cord responds to load-related input applied to the lower shank during stance by modifying step timing and muscle activation patterns, while preserving normal swing kinematics and interlimb coordination.

Using robotics to teach the spinal cord to walk

We have developed a robotic device (e.g. the rat stepper) that can be used to impose programmed forces on the hindlimbs of rats during stepping. In the present paper we describe initial experiments using this robotic device to determine the feasibility of robotically assisted locomotor training in complete spinally transected adult rats. The present results show that using the robots to increase the amount of load during stance by applying a downward force on the ankle improved lift during swing. The trajectory pattern during swing was also improved when the robot arms were programmed to move the ankle in a path that approximated the normal swing trajectory. These results suggest that critical elements for successful training of hindlimb stepping in spinal cord injured rats can be implemented rigorously and evaluated using the rat stepper.

Web-based telerehabilitation for the upper extremity after stroke

Stroke is a leading cause of disability in the United States and yet little technology is currently available for individuals with stroke to practice and monitor rehabilitation therapy on their own. This paper provides a detailed design description of a telerehabilitation system for arm and hand therapy following stroke. The system consists of a Web-based library of status tests, therapy games, and progress charts, and can be used with a variety of input devices, including a low-cost force-feedback joystick capable of assisting or resisting in movement. Data from home-based usage by a chronic stroke subject are presented that demonstrate the feasibility of using the system to direct a therapy program, mechanically assist in movement, and track improvements in movement ability.

Alterations in reaching after stroke and their relation to movement direction and impairment severity

OBJECTIVES

To examine the effects of stroke severity and target location on reaching (1) to identify regions in space that are difficult to reach, (2) to determine whether specific alterations in reaching are associated with particular clinical impairment levels, and (3) to characterize relationships between reaching alterations.

DESIGN

Participants reached toward a screen of 75 targets spanning an approximate range from +/-90 degrees side to side and from waist to head.

SETTING

Rehabilitation research center.

PARTICIPANTS

Sixteen chronic stroke patients with a wide range in residual arm function and 4 control subjects.

INTERVENTIONS

Not applicable.

MAIN OUTCOME MEASURES

Chedoke-McMaster Stroke Arm Assessment, distance, velocity, smoothness, straightness, and direction of the hand path during each reach. Hand position trajectories were recorded with an electromagnetic sensor.

RESULTS

Reaches performed with the impaired arms showed significant degradation in all performance measures. Although only modestly dependent on the target location, these features correlated strongly with impairment level, as well as with each other. Reaching distance showed the strongest correlations with the other parameters.

CONCLUSIONS

Stroke alters a broad array of features of reaching, yet largely the same degree of movement control is preserved across a range of target locations. The only consistently problematic task is to reach far out from the torso, independent of the movement direction. Thus, active range of motion (AROM), rather than control over a specific subset of movement directions, is a logical focus for therapy. In addition, measuring AROM is a simple clinical measure that yields much information.

Directional control of reaching is preserved following mild/moderate stroke and stochastically constrained following severe stroke

Recent evidence suggests that brain injury can impair the ability to independently activate shoulder and elbow muscles. We hypothesized that if muscle activation patterns are constrained, then brain-injured subjects should not be able to accurately grade initial hand movement direction during reaching toward a broad range of target directions. To test this hypothesis, we measured hand trajectories during reaching in three-space by 16 hemiparetic stroke subjects to an array of 75 targets distributed throughout the workspace. Contrary to our hypothesis, we found that the ability to grade movement direction was largely preserved following mild and moderate stroke. However, the most severely impaired subjects exhibited a degradation of directional control consistent with a loss of independent muscle control. Initial and final hand movement directions for these subjects were grouped roughly in two opposing directions, in a plane parallel with the coronal plane of the body, rather than distributed across the normal range. Selection between the two movement directions appeared partially random, in that subjects initiated over 50% of movements in the direction generally opposite the intended target, for targets to one side of the body. These results suggest that individuals with severe stroke are constrained to use only two gross, stereotypical muscle coactivation patterns for reaching control, and that selection between these patterns is stochastically influenced as the actual direction of motion is not strictly predictable given the desired direction.

Retraining the injured spinal cord

The present review presents a series of concepts that may be useful in developing rehabilitative strategies to enhance recovery of posture and locomotion following spinal cord injury. First, the loss of supraspinal input results in a marked change in the functional efficacy of the remaining synapses and neurons of intraspinal and peripheral afferent (dorsal root ganglion) origin. Second, following a complete transection the lumbrosacral spinal cord can recover greater levels of motor performance if it has been exposed to the afferent and intraspinal activation patterns that are associated with standing and stepping. Third, the spinal cord can more readily reacquire the ability to stand and step following spinal cord transection with repetitive exposure to standing and stepping. Fourth, robotic assistive devices can be used to guide the kinematics of the limbs and thus expose the spinal cord to the new normal activity patterns associated with a particular motor task following spinal cord injury. In addition, such robotic assistive devices can provide immediate quantification of the limb kinematics. Fifth, the behavioural and physiological effects of spinal cord transection are reflected in adaptations in most, if not all, neurotransmitter systems in the lumbosacral spinal cord. Evidence is presented that both the GABAergic and glycinergic inhibitory systems are up-regulated following complete spinal cord transection and that step training results in some aspects of these transmitter systems being down-regulated towards control levels. These concepts and observations demonstrate that (a) the spinal cord can interpret complex afferent information and generate the appropriate motor task; and (b) motor ability can be defined to a large degree by training.

Impedance control and internal model formation when reaching in a randomly varying dynamical environment

We investigated the effects of trial-to-trial, random variation in environmental forces on the motor adaptation of human subjects during reaching. Novel sequences of dynamic environments were applied to subjects' hands by a robot. Subjects reached first in a "mean field" having a constant gain relating force and velocity, then in a "noise field," having a gain that varied randomly between reaches according to a normal distribution with a mean identical to that of the mean field. The unpredictable nature of the noise field did not degrade adaptation as quantified by final kinematic error and rate of adaptation. To achieve this performance, the nervous system used a dual strategy. It increased the impedance of the arm as evidenced by a significant reduction in aftereffect size following removal of the noise field. Simultaneously, it formed an internal model of the mean of the random environment, as evidenced by a minimization of trajectory error on trials for which the noise field gain was close to the mean field gain. We conclude that the human motor system is capable of predicting and compensating for the dynamics of an environment that varies substantially and randomly from trial to trial, while simultaneously increasing the arm's impedance to minimize the consequence of errors in the prediction.

Understanding and treating arm movement impairment after chronic brain injury: progress with the ARM guide

Significant potential exists for enhancing physical rehabilitation following neurologic injury through the use of robotic and mechatronic devices (or "rehabilitators"). We review the development of a rehabilitator (the "ARM Guide") to diagnose and treat arm movement impairment following stroke and other brain injuries. As a diagnostic tool, the ARM Guide provides a basis for evaluation of several key motor impairments, including abnormal tone, incoordination, and weakness. As a therapeutic tool, the device provides a means to implement and evaluate active assist therapy for the arm. Initial results with three stroke subjects demonstrate that such therapy can produce quantifiable benefits in the chronic hemiparetic arm. Directions for future research regarding the efficacy and practicality of rehabilitators are discussed.

Persistence of motor adaptation during constrained, multi-joint, arm movements

We studied the stability of changes in motor performance associated with adaptation to a novel dynamic environment during goal-directed movements of the dominant arm. Eleven normal, human subjects made targeted reaching movements in the horizontal plane while holding the handle of a two-joint robotic manipulator. This robot was programmed to generate a novel viscous force field that perturbed the limb perpendicular to the desired direction of movement. Following adaptation to this force field, we sought to determine the relative role of kinematic errors and dynamic criteria in promoting recovery from the adapted state. In particular, we compared kinematic and dynamic measures of performance when kinematic errors were allowed to occur after removal of the viscous fields, or prevented by imposing a simulated, mechanical "channel" on movements. Hand forces recorded at the handle revealed that when kinematic errors were prevented from occurring by the application of the channel, recovery from adaptation to the novel field was much slower compared with when kinematic aftereffects were allowed to take place. In particular, when kinematic errors were prevented, subjects persisted in generating large forces that were unnecessary to generate an accurate reach. The magnitude of these forces decreased slowly over time, at a much slower rate than when subjects were allowed to make kinematic errors. This finding provides strong experimental evidence that both kinematic and dynamic criteria influence motor adaptation, and that kinematic-dependent factors play a dominant role in the rapid loss of adaptation after restoring the original dynamics.

Mechatronic assessment of arm impairment after chronic brain injury

Significant potential exists for mechatronic devices to improve assessment and treatment of individuals with a movement disability following stroke, traumatic brain injury, or cerebral palsy. We report the use of a mechatronic device for evaluation of the arm after chronic brain injury. We performed a series of experiments with the device in order to identify the relative contribution of three different motor impairments to decreased active range of motion of reaching in five brain-injured subjects. Our findings were that passive tissue restraint and agonist weakness, rather than antagonist restraint, were the most common contributors to decreased active range of motion. These results demonstrate the feasibility of objective assessment of functional movement using a mechatronic device, and could provide the basis for improved, individualized treatment planning and monitoring following brain injury.

Assessment of active and passive restraint during guided reaching after chronic brain injury

We report the use of a mechatronic device for assessing arm movement impairment after chronic brain injury. The device, called the "Assisted Rehabilitation and Measurement Guide," is designed to guide reaching movements across the workspace, to measure movement and force generation, and to apply controlled forces to the arm along linear reaching paths. We performed a series of experiments using the device in order to identify the contribution of active muscle and passive tissue restraint to decreased active range of motion of guided reaching (i.e., "workspace deficits") in a group of five chronic, spastic hemiparetic, brain-injured subjects. Our findings were that passive tissue restraint was increased in the spastic arms, as compared to the contralateral, nonparetic arms. Active muscle restraint, on the other hand, was typically comparable in the two arms, as quantified by measurements of active arm stiffness at the workspace boundary during reaching. In all subjects, there was evidence of movement-generated weakness, consistent with a small contribution of spasticity to workspace deficits. These results demonstrate the feasibility of mechatronic assessment of the causes of decreased functional movement, and could provide a basis for enhanced treatment planning and monitoring following brain injury.

Guidance-based quantification of arm impairment following brain injury: a pilot study

This paper reports the design and preliminary testing of a device for evaluating arm impairment after brain injury. The assisted rehabilitation and measurement (ARM) Guide is capable of mechanically guiding reaching and retrieval movements across the workspace and of measuring constraint forces and range of motion during guidance. We tested the device on four hemiplegic brain-injured individuals and four unimpaired control subjects. During guided movement, the brain-injured subjects generated distinct spatial patterns of constraint force with their impaired arms that were consistent with the standard flexion and extension "synergies" described in the clinical literature. In addition, the impaired arms exhibited well-defined workspace deficits as measured by the ARM Guide. These results suggest that constraint force and range of motion measurements during mechanically guided movement may prove useful for precise monitoring of arm impairment and of the effects of treatment techniques targeted at abnormal synergies and workspace deficits.

Feedforward stabilization in a bimanual unloading task

When one hand removes a load from the other hand, feedforward motor commands stabilize the position of the unloaded hand. We studied the stabilization of the postural hand using a novel apparatus that allowed unloading at different rates, and unexpected uncoupling of the unloading force from the postural hand. Feedforward stabilization of hand position was observed in all subjects. This stabilization was achieved both by deactivation of postural agonist muscles and by activation of postural antagonist muscles. The neural feedforward command apparently increased with unloading rate. However, the command only partially canceled the interaction torque generated by removing the load, and stabilization became less effective as unloading rate increased.