Research And Application Of Gait Perception Prediction And Control Method Of Lower Limb Exoskeleton

As one of the current popular research techniques,the lower extremity exoskeleton robot plays an important role in enhancing the body's exercise capacity and assisting the rehabilitation training of patients with lower limb diseases.How to rationally plan and control the trajectory of exoskeleton of lower extremities is one of the key issues to solve the coordinated movement of exoskeleton and human body.In this thesis,a study of the gait prediction algorithm and the lower limb exoskeleton control method for the gait control problem of lower extremity exoskeleton based on inertial measurement unit sensing system have carried out.The research content of this thesis mainly includes the following three aspects:(1)To solve the problem that the lower limb exoskeleton actuator lags behind the human motion,a gait single step prediction method based on the Gradient Boosting algorithm was studied.In this thesis,the sliding window was used to extract the joint angle sequence from the collected human gait data to construct the dataset,and use XGBoost and LightGBM based on the gradient boosting algorithm to train the dataset.Finally,the prediction model trained by XGBoost and LightGBM was used to predict the test samples,and compared with the results of Kalman filter gait prediction algorithm.Experimental results show that LightGBM is more accurate than Kalman filter in gait prediction,and has faster training speed than XGBoost.(2)To solve the problem that the gait single step prediction can not completely offset the delay of the lower limb exoskeleton actuator,two multi-step gait prediction methods based on LightGBM are proposed.By analyzing the defects of LightGBM gait single step prediction algorithm in exoskeleton gait predictive control,taking the prediction model trained by LightGBM algorithm as the core,the gait multi-step prediction of the lower limb exoskeleton is realized by using the single prediction model to calculate the target value iteratively and the multi prediction model to calculate in parallel.Experimental results show that two kinds of multi-step prediction based on lightgbm can accurately predict the angle change of gait joint.In this thesis,the online training prediction model process is designed for multi-step prediction of single-predictive model target value iteration,and the feasibility of online training is verified by actual gait data simulation.(3)To solve the problem of controlling the exoskeleton of lower limbs with multiple joints,a multi-motor closed-loop control method based on gait sensing feedback was studied.In this thesis,a hierarchical model of gait-aware feedback is designed,by analyzing the PWM speed regulation characteristics of a single motor and the PID control algorithm,the gait-sensing feedback model is combined with multi-motor cooperative control to achieve coordinated control of multiple joints of lower extremity exoskeleton.Finally,the feasibility of multi-motor closed-loop control based on gait-sensing feedback model for multi-joint coordinated control of lower extremity exoskeleton is verified by analyzing the response time and system response time of each motor in multi-motor system.