Adaptive assistive control of a soft elbow trainer with self-alignment using pneumatic bending joint

Abstract

For safe and effective robot-assisted rehabilitation, natural inherent compliance and self-alignment of rehabilitation devices completed with assistive behavior are assumed to be the essential properties. To provide required human joint stability each joint can be separately supported using exoskeleton-like devices. However, the necessity of exact adjustment to the individual extremity is very time-consuming for physiotherapists and strongly reduces the effective treatment time. In this paper a soft elbow trainer based on pneumatic bending joint using skewed rotary elastic chambers (sREC) is presented as first specific solution. This shaftless actuator is placed under the elbow joint and allows for implicit self-alignment to the polycentric movement of human joint axis without elaborate adjustments. Position estimation is performed using two accurate inertial measurements units (IMUs) and four less accurate but robust cost-effective resistive bend sensors (flex sensors). Sensor fusion of flex sensor and IMU signals is used to obtain a robust control feedback. An artificial neural network (ANN) is applied to combine flex sensor signals. The adaptive assistive controller learns online using dynamic model function approximation and takes into account the patient's behavior, effort and abilities while maximizing the patient's voluntary effort. Practical tests with healthy subjects confirm the effectiveness of the controller.