Research On Gait Planning And Repetitive Learning Motion Control Of Biped Robot

Biped robots have complex mechanical structure and redundant degrees of freedom,and their dynamic models are usually multivariable,nonlinear,time-varying and high-order,which makes it difficult for existing biped robots to achieve stable and reliable motion planning and existing motion control strategies to achieve accurate trajectory tracking.Therefore,this dissertation conducts research on the trajectory planning and motion control of bipedal robots,aiming to improve the walking stability of the current bipedal robots.This dissertation proposes an energy optimized trajectory generation method based on zero moment point(ZMP)stability domain constraints for general walking gait,as well as a repetitive learning motion control method based on radial basis function(RBF)neural network.This method can improve the accuracy of motion tracking for bipedal robots based on repetitive motion trajectories.The main work of this dissertation is as follows:(1)This dissertation proposes a gait trajectory generation method based on robot energy optimization based on ZMP constraints.Specifically,given the target parameters of the motion trajectory and the target motion task,a group of variable parameters(N,V)can be used to describe a gait trajectory cluster.At the same time,the generated foot trajectory can be used to obtain the ZMP stability region of the motion cycle,and its ZMP stability region constraints and robot dynamics model can be used to transform the energy of the center of mass into a quadratic form form,Furthermore,the trajectory of the lowest energy centroid motion within the reachable range of a set of variable parameter pairs(N,V)is obtained.Finally,the variable parameter pair(N,V)is iterated,and the joint motion trajectory is inversely solved through the robot kinematics as the reference trajectory for motion control.Obtain the lowest energy joint motion trajectory to complete the target motion task.(2)The walking motion of a bipedal robot has periodic repeatability.This paper designs a repetitive learning controller based on RBF neural network to address this characteristic.Due to the widespread occurrence of periodic disturbances in periodic walking movements,this paper also uses the function approximation characteristics of RBF neural networks to compensate for periodic disturbances,and provides periodic negative disturbance outputs in joint control to compensate for them.Furthermore,the disturbance error of the periodic repetitive motion of the bipedal robot is reduced,and the accuracy of the periodic repetitive motion of the bipedal robot is improved.(3)In order to verify the effectiveness of the energy optimization trajectory generation method based on ZMP stability domain constraints and the repetitive learning controller based on RBF neural network proposed above,a series of experiments were conducted on the webots simulation platform and the UBTECH bipedal robot platform Walker,respectively,to verify the effectiveness of the proposed method.Specifically,periodic repetitive motion simulations and experiments,including squatting in place,energy optimization trajectory generation method based on ZMP stability domain constraints,and walking,were conducted.The experimental results showed that the controller proposed in this paper can effectively reduce tracking errors,proving the effectiveness of the proposed method.