Research On Continuous Control Technology For Lower Limb Exoskeleton Based On Motion Imagery Decoding And Online Gait Planning

Stroke has become one of the leading causes of death and disability among the Chinese population,and the rehabilitation method combining brain-machine interfaces(BMI)and exoskeletons has received much attention.Among them,the motor imagerybased BMI can provide motor imagery therapy that helps with rehabilitation while controlling the exoskeleton.It has been proven effective,but the weak and noisy electroencephalogram(EEG)signals due to the brain structure make it difficult to accurately classify and recognize lower limb motor imagery,resulting in lower recognition rates for left and right foot motor imagery.In addition,the low granularity of the existing noninvasive BMIs for intention recognition cannot correspond to the fine movement coding of the lower limb exoskeleton,and the existing exoskeleton control algorithm operates at the task level,allowing users to only control the next action of the exoskeleton without the ability to control how it is done.This limits the ability to motivate patients to actively participate in rehabilitation training.To address these issues,this thesis conducted the following research:(1)In order to address the problem of low recognition rates for left and right foot motor imagery in existing models,this thesis proposed a lower limb motor imagery classification algorithm based on the MFCP＿BLSTM model.The algorithm first used Fast Independent Component Analysis(FASTICA)to denoise the EEG signals to improve their purity.Then,the EEG signals were decomposed into different frequency bands using the Wavelet Packet Transform(WPT)algorithm,and the Common Spatial Patterns(CSP)algorithm was used to calculate the spatial distribution feature matrix of the EEG signals in different frequency bands.These feature matrices were superimposed on the channel dimension of the EEG signals to calculate the Power Spectral Density(PSD)features that contain both frequency and spatial distribution information.Finally,these features were input into a BLSTM based neural network for training to learn deep features in the time dimension,and for classification to output the three-class probabilities of left foot,right foot,and idle motor imagery.The performance of the algorithm was validated on a public competition dataset,and the average kappa coefficient increased by 9% to reach 0.69 compared to other award-winning participants.In the online experiment for lower limb motor imagery,the classification accuracy of left and right foot reached 77%,and the classification accuracy of idle and motor imagery reached 83%.This MFCP＿BLSTM-based algorithm for lower limb motor imagery classification can improve the accuracy and stability of lower limb motor imagery recognition and has high practical application value.(2)To enhance patients’ active rehabilitation willingness and promote the recovery of damaged brain nerves,this thesis proposes a continuous control framework for exoskeletons based on lower limb motor imagery and online gait planning.The framework includes two phases,namely the voluntary control phase and the passive control phase.In the voluntary control phase,the classification probabilities of continuous predicted motor imagery are converted into the angle of the hip joint motor rotation,controlling the height of lifting the leg and affecting the step length.In the passive control phase,the DMP algorithm is used to plan the exoskeleton motion curve online,so that after the angle change of the hip joint during leg lifting,the other joints can dynamically coordinate with the hip joint to achieve stable walking.Additionally,a control logic based on a finite state machine is designed to switch between multiple states of exoskeleton walking.This control framework integrates multiple methods and algorithms,which can improve the accuracy and stability of exoskeleton voluntary control and better promote the rehabilitation effect.(3)Furthermore,a brain-controlled lower limb exoskeleton medical system has been developed,which deploys the trained MFCP＿BLSTM model and applies the exoskeleton continuous control technology based on lower limb motor imagery.The system allows patients to independently control the exoskeleton walking while wearing it,with autonomous control manifested in the order of foot movement,height of leg lifting,and size of stride.The correctness of the system’s functionality,feasibility,and safety of the above method have been experimentally verified.