Research On Personalized Gait Generation And Human-like Control For Lower Limb Rehabilitation Robot

With the gradual increase in the aging of our population.The physical functions of the elderly decline with age,making them prone to predisposing diseases,especially strokes,neurological diseases and limb injuries that lead to lower extremity motor dysfunction in patients.Research indicates that long-term and effective rehabilitation reshape motor nerves,gradual restoration of partial or full motor function.At present,there is a serious mismatch between the rehabilitation medical resources and the people in need of rehabilitation.As an "assistant" to rehabilitation doctors,the lower limb rehabilitation robot has to some extent alleviated the shortage of rehabilitation medical resources,ensure the effectiveness and practicality of rehabilitation exercises.However,the degree of injury is different for each patient,so how to tailor the reference trajectory,ensure the effectiveness and safety of the design controller for each patient has become a key aspect of the study of lower extremity rehabilitation robots,so the following research is done in this paper:(1)Starting from the gait characteristics of the human body,the gait datasets of different genders and different human bodies are constructed.Firstly,a 3D motion capture system is used to build the gait acquisition experimental space,and 21 volunteers are allowed to walk freely20 times each in the experimental space according to three speeds: fast,medium,and slow.Finally,through a series of data pre-processing steps such as filtering and uniform data length,the lower limb joint angles of the volunteers in the experimental space are calculated,and a gait data set of 21(12 males and 9 females)volunteers with a total of 726 sets of gait data is constructed.(2)The gait dataset is used to train neural networks to create mapping relationships between gait parameters and body parameters to gait trajectories.A gait generation model based on generalized regression network,a gap loss function-based gait generation model for LSTM network,and a GLS network-based gait generation model are designed.The patient’s body parameters and gait parameters are used as the input of the gait generation model,and the output is a continuous joint trajectory.The generated joint trajectories are compared with the actual joint trajectories,and the experimental results are analyzed using MAD.The GLS networkbased gait generation method is superior to the previous two methods,and the generated joint trajectories are closer to the actual joint trajectories,and continuous gait trajectories could be generated.(3)To ensure the effectiveness of rehabilitation training,the joint trajectory generated by the gait generation model is used as the reference trajectory of the robot,sliding mode neural network control and human-like slide neural network control are designed.In response to the inability to obtain an accurate robot model,this paper utilizes the nonlinear approximation capability of RBF neural networks to approximate the unknown terms in the robot model to reduce the difficulty of modeling.The robot model is susceptible to external disturbances,and the anti-disturbance of sliding mode control is used to suppress the external disturbances to achieve fast and effective tracking of the reference trajectory.After Matlab simulation to verify these two methods,the design of humanoid neural network sliding mode control is more suitable for patients to perform rehabilitation training.