An online gain tuning proxy-based sliding mode control using neural network for a gait training robotic orthosis

Abstract

In the last decades, wearable powered orthoses have been introduced in the state of the art for lower-limb rehabilitation. Most of these applications are driven by electric motors. Comparing with electric motor actuators, pneumatic artificial muscle (PAM) actuators are compliant because of the elasticity of PAMs. Consequently, for more safety and comfort of lower-limb rehabilitation, a compliant robotic orthosis powered by PAMs is developed. Based on safe control, a new control method, proxy-based sliding mode control (PSMC), was introduced into rehabilitation robotics a few years ago. It combined safety and accuracy of tracking to make it suitable for the safe control of PAM actuators. As the reason of low frequency response of PAM actuators and variable loads caused by different human subjects, the fixed parameters of PSMC makes the tracking performance vary from subject to subject, and lacks robustness. This paper presents a modification of PSMC by using neural network to tune PSMC gains online, and implements both PSMC and modified PSMC control schemes in the robotic orthosis. Experimental results demonstrate that the improved PSMC method performs better on tracking with little degradation on safety for different loads and human subjects.