Robust Model Predictive Control of a Lower Limb Exoskeleton Robot with Suspension Gravity-Assist using Deep Neural Network

The robotic exoskeleton system with an accurate model can help the subjects improve motor function rehabilitation, but the performance of model-based controllers is often affected by the accuracy of model parameter identification and uncertainty during locomotion. In this paper, a deep learning-based algorithm is proposed to identify the dynamics parameters of the exoskeleton system to assist the user in walking, and based on this, a robust model predictive control is further presented. Firstly, the dynamics of the exoskeleton system are modeled and then the dynamics parameters are learned online by deep learning algorithm. By using the desired trajectory of the exoskeleton and mobile platform generated by polynomial interpolation, the robust model predictive control is designed to ensure the user and exoskeleton with suspension gravity-assist system stable and continuous motion. The user experiments are carried out to verify that the proposed deep learning method can effectively identify the dynamics parameters and the overall system can achieve good tracking performance with the developed controller using identified parameters. These results show the potential to promote gait rehabilitation combining neural-learning based precise model control with the exoskeleton with suspension body weight support system and to be applied in clinical routine.