Performance-based adaptive assistance for different subtasks of walking in LOPES II

1. INTRODUCTION
Robotic gait training is a promising tool to improve walking ability after stroke, however, therapeutic effect might largely depend on the type of robotic gait trainer and control algorithm that is used [1]. Therapy should be task-specific and promote active participation as this is crucial for motor learning. Some algorithms have been proposed to encourage patients to actively participate [2], however, the amount of assistance for different subtasks of the gait cycle is chosen by the therapist, and therefore, therapy could be affected by subjective decisions.
This contribution presents a novel controller that is able to automatically adapt the assistance for specific subtasks of gait based on patients’ performance. This approach is expected to improve robotic gait therapy.
2. MATERIALS AND METHODS
2.1 Robotic platform
LOPES II was used to develop and test the algorithm. LOPES II is a treadmill-based, admittance controlled robotic gait trainer that has eight actuated degrees of freedom to control the motion of the lower limbs and pelvis [2].
Currently, LOPES II is 'manually' controlled through a graphical user interface in which the operator can adjust the level of support for several gait subtasks (i.e. step length, step height, stance, prepositioning and weight shift).
2.2 Adaptive controller
A new adaptive controller that is able to automatically change the assistance provided to the patient has been developed. The controller evaluates patients’ performance with respect to reference values in a defined number of preceding steps. This evaluation is done for each subtask of walking separately. Based on this, the assistance is automatically adjusted for each subtask of gait and might be affected in different ways: first, if performance is within a specific range around a previously selected threshold, amount of assistance in the particular subtask will remain constant; and second, assistance will be increased or decreased depending on whether the performance is below or above the specified range respectively.
3. RESULTS AND DISCUSSION
A preliminary evaluation of the algorithm in healthy users showed the potential of the controller to adjust the assistance for different subtasks of gait depending on the user's needs. (e.g. simulated stiff knee gait resulted in increased assistance in step height or knee flexion; meanwhile simulated crouch gait was corrected providing assistance in knee and hip extension).
In coming studies, authors plan to validate the algorithm in stroke survivors in two ways: First, to evaluate its benefits compared to the current 'manual' assistance where the operator selects the amount of support; and second, to study the effect of different amounts of body weight support and walking speed on the amount of assistance provided by the algorithm.
In the future, an optimized version of this algorithm might not only improve robotic gait training for stroke survivors, but could also be used for assessment of patients’ abilities and to monitor progress during rehabilitation.