Research On The Integration And Active Control Technologies Of Lower-limb Exoskeleton For Rehabilitation Training

With economic development and aging population growth,the lower-limb motor dysfunction caused by cardio-cerebrovascular disease,and accidental injuries,has fre-quently occurred,which seriously affects the daily life and work of the patients.As a new type of rehabilitation equipment,the lower-limb exoskeleton robot can help people with lower-limb motor dysfunction in rehabilitation treatment,and has advantages in improving the rehabilitation effects and training economically.At present,the lower-limb exoskeleton robot are gradually being marketed,however,there are still many common problems in practical applications,such as a large bearing load provided by the exoskeleton system,the uncoordinated motion between human and robot,ignor-ing human motion intention in exoskeleton control and non-personalized gaits adopted by the lower-limb exoskeleton robot,etc'The above problems can lead to users' low participation and poor compliance,thus affecting the rehabilitation effects in practical applications.In order to solve some common problems existing in the current research,this research has designed serialized lower-limb exoskeleton systems,and provided the key technical scheme in different active control.The details are as follows:(1)Aimed at the problem of uncoordinated motion between human and robot,we have proposed the synergy-based control of lower-limb exoskeletons by skill transfer from the perspective of human-robot stiffness matching.First,the joint-stiffness estima-tion model modulated by muscle activity level is established to realize the real-time esti-mation of joint stiffness by surface electromyography(sEMG)signals.Then,the model of exoskeleton dynamics and joint variable impedance are established,the unknown dynamic parameters are estimated by fuzzy estimators,and a human-robot cooperative controller is designed.The Lyapunov direct method is used to prove the boundedness of errors in human-robot stiffness matching and dynamic model estimation.Finally,the above scheme is verified by experiments,and the results show the feasibility of the control scheme and the effectiveness of human-robot coordination improvement.(2)Aiming at the problem of ignoring human motion intention in exoskeleton con-trol,we have proposed the active human-following control of exoskeleton robots based on continuous motion intention recognition.First,the 3D tracking information of hu-man body is obtained by combining the binocular camera and the OpenPose algorithm,which solves the complicated feature point matching problems in depth acquisition of the binocular camera.The kinematics model of the assistive walker with non-holonomic constraints is established,and a speed controller is designed to track the position of the human shoulder in real time.Then,to obtain human motion intention,the neural net-work based on long short-term memory(LSTM)is designed for continuous mapping of multi-channel sEMG signals to joint angles.Moreover,in order to identify the gait phase during walking,a gait phase recognition method based on distribution character-istics of plantar pressure is proposed.A trajectory tracking controller is designed based on the barrier Lyapunov function,which is used to ensure that the unknown system non-linearity is bounded.Finally,the effectiveness of the overall control scheme is verified through experiments.The comparative results show that that the LSTM network model has high accuracy in predicting joint angles,and smaller tracking errors utilizing the trajectory tracking controller.(3)Aimed at the problem that the gait of the exoskeleton robot are not personalized,we have proposed a method that combing virtual reality(VR)technology and electroen-cephalogram(EEG)signals decoding to generate the active gaits for the exoskeleton robot.First,a method based on gaussian process regression(GPR)is proposed to pre-dict the key gait parameters,which are used to construct the constraint conditions to generate human single-step personalized gait trajectories by the polynomial interpola-tion.Then,in order to generate continuous movement gaits,a EEG-triggered method for single-step movements under motor imagination is proposed.By combing the VR technology,the visual fatigue caused by the use of visual stimulation to collect EEG signals can be avoided.On the one hand,a VR scene with forest and grass as themed environment is developed,and relevant markers are set to remind users to image clas-sified movements.On the other hand,the common spatial pattern(CSP)algorithm is used to extract the features of EEG signals that are classified based on the support vector machine(SVM)learning afterwards.Finally,the effectiveness of the method is verified through some experiments.The results show that the generated GPR models has pre-diction error rates of about 5%for the key gait parameters.Compared with spontaneous motor imagination,it is easier to generate different EEG signals in the VR scene.