Advanced Iterative Learning Algorithm for Control of Rehabilitation Robots

Abstract

Stroke and sports injuries are both major contributors to temporary and permanent disabilities. A well designed, patient specific rehabilitation programme has been shown to be a very important tool to improve patient outcome and to prevent repeat injuries. Robotic rehabilitation offers benefits such as precision, force accuracy and task specific training when compared to traditional manual manipulation method as used by physiotherapists. In this research, a more robust control technique for robotic rehabilitation devices is presented. Currently, most rehabilitation robots use some form of impedance control involving cascading outer position and inner force control loops, and are usually tuned to cater for a range of users. However, different capabilities among patients such as their ranges of motion and force capacities result in unoptimised rehabilitation programmes. In order to obtain the best performance possible, both control loops need to be manually adjusted, often by trial and error or empirical knowledge. Iterative learning control in the form of iterative feedback tuning (IFT) is explored to mitigate this shortcoming as it offers a model-free method to optimally obtain a local optimum of the controller parameters. The performance of the IFT technique depends on several factors, one of which being the optimisation algorithm. Three algorithm candidates are examined in detail through simulation and experiment, which yields the Gauss-Newton algorithm as the winner when considering tuning performance and stability. The design criterion is another factor that influences IFT’s performance. A normalised version of the design criterion is proposed along with an optimal value range for the weighting factor. The newly presented normalised design criterion is system agnostic and is demonstrated to improve the robustness of the IFT technique. IFT with the proposed system agnostic design criterion is then implemented on two novel robotic rehabilitation devices using various configurations. Impedance control with iterative learning capabilities is explored on a wearable knee joint rehabilitation orthosis powered by pneumatic muscle actuators. IFT is used to obtain optimal controller parameters on various parts of the control system. A multiple degrees-of-freedom IFT technique is proposed and examined on a redundantly actuated parallel ankle rehabilitation platform to control its end effector orientations in joint space. For both rehabilitation robots, human participant trials are also conducted in order to investigate the efficacy and robustness of the IFT technique under real life rehabilitation scenarios. The results of this research can be applied to a wide variety of control system structures. With its model-free and robust nature, the techniques proposed and demonstrated have the possibility to provide tangible benefits to both patients and physiotherapists alike.