Research On Human-robot Harmony Theory Of Flexible Lower-limb Rehabilitation Training Robot

Motor rehabilitation has an important impact on the neural plasticity of patients with limb dysfunction.Human-robot harmony is the key to achieve individualized rehabilitation treatment.Because of different special hemiplegic gait of patients,rigid mechanism of rehabilitation training robot,inadequate combination of patient initiative and robot automatic control,and lack of interaction between physiotherapists and robots,it is difficult to adopt traditional robot motion control methods to solve the behavior coordination and natural interaction of patients,robots and physiotherapists in the dynamic treatment process.These factors lead to a gap with the clinical needs.In the study,the key scientific problems of behavior,task and intelligence harmony are solved to realize safe and effective individualized rehabilitation treatment.The overall structure of the human-robot harmony system for gait rehabilitation training is proposed,which is different from the conventional structure composed of physiotherapists,patients and rehabilitation robots.It adds the manipulated exoskeletons worn by physiotherapists,so that they can further integrate into the task of rehabilitation training.Considering the combination of rigid support and flexible drive,a flexible bionic mechanism is designed to achieve human-robot walking behavior blending.Based on the walking characteristics of human body and the needs of rehabilitation training,the configuration design of a BWSTT?body-weight-support-treadmill-training?robot is completed from the point of view of structure and motor bionics,which includes a 4-DOF?degree of freedom?exoskeleton of both lower limbs.In addition,a pneumatic proportional servo system matching with the robot is designed to provide driving flexibility.Based on the principle of friction braking and proportional pressure control of diaphragm cylinder,a new type of pneumatic joint with controllable torque is developed.The gravity compensation algorithm is used to provide stable control moment for physiotherapists who wear the multi-joint manipulation exoskeleton.Appropriate rehabilitation training tasks should be formulated for patients with different physiques and conditions and therefore a three-dimensional spatio-temporal gait planning method under the interaction of physiotherapists is proposed to achieve task harmony,breaking through the traditional methods of online generation,prediction and manual setting.The continuous trajectory is obtained by the l₁-constrained Huber loss minimization machine learning algorithm.The kinematics model and velocity model are established by POE?product of exponentials?formula,and the heel motion mapping method is proposed to coordinate the spatio-temporal gait parameters of robot,the center of gravity displacement of weight-reducing mechanism and the speed of treadmill.The integrated planning method enables patients to coordinate and comfortably walk.It adds a new idea of physiotherapist's participation to the human-robot interaction.Starting from the twist and Jacobian matrix of centroid velocity,the Lie group representation for dynamic modeling of serial robots is deduced by combining Lagrange equation.The dynamic model of gait rehabilitation training robot is established in joint space by this method.Based on the biomechanical characteristics of walking,the dynamic models of swinging leg and supporting leg are established for the alternating gait phases.Considering human-robot interaction,a dynamic model of the coupling system is established,and the dynamic characteristics are analyzed in passive,active assisted and active resistance rehabilitation training mode.In order to solve the problem of adapting the strength and difficulty of rehabilitation training to the needs of patients,a method of obtaining control parameters in accordance with the characteristics of patients'condition by the reinforcement learning algorithm is proposed,which is different from the method of setting parameters dynamically by the designer's constructed function and the rule of intelligence or the empirical estimation of specific tasks.A two-layer nested control structure is complemented with force-position interaction algorithm in outer loop and joint position control in inner loop.Considering the control difficulty and modeling error of the pneumatic system,an independent joint decentralized controller with feed-forward compensation for dynamic characteristics is designed to track the gait trajectory in the joint space.A variable stiffness/damp admittance model is established to adapt to patients.Therefore,a two-dimensional mesh representation of state variables is proposed by discretizing the human-robot contact force and joint angle errors,and an algorithm for transforming the reinforcement learning optimal strategy to admittance parameters is proposed to quantify the patient's personalized characteristics to realize the intelligent harmony of human-robot adaptation in rehabilitation training.Three scientific issues are focused in the study,that is,“physiotherapist's interactive gait planning method”,“human-robot coupling dynamics characteristics study in gait rehabilitation training”and“human-robot cooperative control strategy adapting to patient's personalized characteristics”.The human-robot harmony system is constructed by the bionics design of mechanical structure and flexible driving system.The goal gait trajectory is provided by the gait planning method under the interaction of physiotherapists,and the rehabilitation training is realized by the human-robot interactive dynamics and adaptive control strategy.The above research results form a complete realization method of individualized gait rehabilitation training.Through the interaction with the physiotherapist in the experiments of gait planning and motion mapping,the planed gait trajectory and the center of gravity adjustment trajectory are in accordance with the physiological characteristics of human body and the biomechanical characteristics of walking.The maximum overshoot of step response is less than 2 mm in the joint position servo control experiments,which meets the requirement of the allowable range of position error in rehabilitation training.The tracking error is less than 5 mm and there is no time delay in the passive rehabilitation training,which ensures the gait tracking accuracy and real-time performance.The joint trajectory is adjusted adaptively with the change of human-robot contact force in active rehabilitation training experiments to optimize the strength and difficulty of rehabilitation training,and the trajectory adjustment is limited within 20%FS.Therefore,it is verified that the robot can provide the assisted-as-needed?ANN?rehabilitation training in the normal walking gait range.