Assist-as-needed control strategy for upper-limb rehabilitation based on subject's functional ability

Abstract

Assist-as-needed (AAN) robotic rehabilitation therapy promotes neural plasticity and motor coordination through active participation in functional task. The technique significantly improves recovery of lost neural function following neurological impairment such as stroke. A key component of this strategy is to provide robotic assistance to the patients only as it is needed. Thus, accurate estimation of the patients’ functional or movement ability (FA) is required to adequately evaluate the patients’ need for robotic assistance or the amount of assistance torque to be provided. However, several strategies have shown inconsistencies in their estimation techniques and often significantly influenced by interferences and disturbances from the robotic device, making them unsuitable for clinical applications. The previous methods are also not related to clinical assessment scales in clinical practice. This study proposes an assist-as-needed (AAN) control strategy for robotic rehabilitation based on a new formulated Functional Ability Index (FAI) which adequately estimates patients’ functional ability task. The formulated FAI starts with statistical normalization function (FAIN) and then extended to apply a z-spline curve (FAIs) to estimate patient’s movement ability based on the quality of movement (QoM) and the time score (TS) of the patient in each functional task. The approach is defined to be consistent with popular clinical method such as Wolf Motor Function Test (WMFT) and Action Research Arm Test (ARAT). A baseline low-level Linear Quadratic Gaussian (LQG) torque controller with integral action for robustness is integrated to the AAN controller to physically provide low-level torque assistance on the elbow and shoulder joint during reaching tasks. The overall strategy is automated by a hybrid finite state automaton to smoothly coordinate the patients’ activities. Simulation and experimental study have been conducted to validate the proposed AAN control strategy with the newly formulated FAI. Eighteen patients completed the experimental study which involves two clinical tasks: a pick-and-place reaching task (Task 1) involving shoulder adduction/abduction movement and a table-to-mouth task (Task 2) involving elbow flexion/extension movement. Results showed that the proposed FAI algorithm could estimate patients’ functional movement ability consistently with repeatability. And also allowing the proposed AAN control strategy to regulate the amount of torque assistance as needed, in accordance to the patients’ functional ability and following the clinical assessment scales. The average torque assistance across trials for the mild and severely disabled patients were found in the range 3% - 6% and 42% - 47% for Task 1, respectively; and in the range 9% - 14% and 48% - 55% for Task 2, respectively; which are consistent with their FAI. The results show that the proposed AAN control strategy with the newly formulated FAI can successfully provide the assistance torque as needed to the patients in accordance with their patient’s functional ability.