Compliant Control Of Walking Rehabilitation Training Robot Based On Long Short-term Memory Network

Facing the rapid development trend of population aging and the rapid increase of patients with motor dysfunction,it is impossible to meet the supply demand for medical services to help the elderly and the disabled only by relying on medical trainers to conduct sport rehabilitation training.Therefore,walking rehabilitation training robots and their related technologies emerge at the historic moment.In the research of walking rehabilitation training robots,the compliance control technologies of walking rehabilitation training robots are particularly important for the user's comfort in human-computer interaction.Therefore,from the perspective of compliance control of walking rehabilitation training robot,this thesis is performed the following aspects:Firstly,this thesis introduces a walking rehabilitation training robot independently developed by our laboratory,its mechanical structure and interaction mode.The kinematics model of the walking rehabilitation training robot is established.Combining with the mechanical structure characteristics of the robot,an easy-to-integrate and non-contact multi-channel proximity laser sensor is designed to collect the relative distance of human-computer interaction during walking rehabilitation training.Secondly,to calculate the relative dynamic distance between the legs and the robot in the process of human-computer interaction,four the time series models of kalman filter,autoregressive moving average model,LSTM model and the improved LSTM model are used to solve the interaction distance between the legs.Simulation experiments are conducted for different types of gait data based on the four kinds of models.Combining the absolute error distribution and the performance evaluation index,the pros and cons of the four models in different types of gait prediction effects are analyzed.The result shows that the improved LSTM model proposed in this thesis can effectively improve the prediction accuracy,and has good generalization ability for different types of gait,thus it is suitable for calculating interaction distance of legs at the next moment.Thirdly,the distance-speed conversion algorithm is used to convert the calculated distance information into the user's intention speed.The MPC controller and the PID controller based on the kinematics model of the walking rehabilitation training robot are established respectively.The simulation and comparison of the user's intentional speed when walking at constant speed and variable speed are conducted.The result shows that the PID controller has better control performance.Aiming at the problem of robot vibrate caused by the user's sudden change of speed,the sample interpolation method is used to smooth the input data.The feasibility and effectiveness of the proposed compliant control method are proved by the experiment.Finally,the prediction performance of the improved LSTM model is conducted for the gait phase.The pressure sensor is used to obtain the plantar pressure data in real time.The improved LSTM model is used to divide the gait phase in the process of human walking,which verifies that the improved LSTM model is also feasible for gait phase.A two-dimensional fuzzy system method combining with subjective and objective evaluation is proposed to evaluate the safety and comfort of the method proposed in this thesis,which proves that the method proposed in this thesis can realize the compliant control of the robot and at the same time enhance the experience in the process of human-computer interaction.