Research Of Model Predictive Control Algorithm For Biped Robot

As an important research object in the field of legged robots,biped robots have broad application prospects.In the future,it is expected to become a human assistant,providing assistance to humans in areas such as emergency rescue and logistics distribution.However,the dynamics of biped robots are very complex,which makes the design of its control strategy extremely difficult.Therefore,designing a stable control strategy for biped robots to enable them to walk and jump on the ground has always been the focus of researchers.In this context,this thesis studies the application of model predictive control algorithm in the field of biped robot control,aiming to realize the stable control of biped robot by virtue of its robustness and adaptability.In this thesis,a forward kinematics model predictive control method is proposed as an inverse kinematics solution method for the biped robot.At the same time,based on the single rigid body model predictive control method,this thesis proposes an improved support leg control method and an improved control framework for biped robots.(1)The forward kinematics model predictive control method is a new inverse kinematics solution method based on numerical iteration.Compared with the numerical iterative method based on the inverse Jacobian matrix,this method does not need to invert the Jacobian matrix,and is suitable for situations where the Jacobian matrix cannot be inverted.At the same time,compared with the geometric analysis method,its design process is fixed and simple,does not depend on the designer’s understanding of the mechanical structure,and has strong versatility.Therefore,the design complexity of the inverse kinematics algorithm of the biped robot can be reduced.(2)The single rigid body model predictive control method based on virtual legs is an improved control method for supporting legs of the biped robot.In the overall control framework of a biped robot,the control method for supporting legs largely determines the stability of the robot.The single rigid body model predictive control method is suitable for quadruped robots with point-shaped feet,while the biped robot has only two legs,resulting in low stability.Especially when standing still,the biped robot cannot adjust the foothold position at this time.Biped robots have line-shaped feet or rectangular-shaped feet,and the contacts with the ground are relatively stable.But this method does not have a high utilization rate of the end joints of the legs,and cannot take this advantage.Experimental results show that the single rigid body model predictive control method based on virtual legs can simultaneously solve the above two problems and enhance the stability when standing.(3)The overall control framework for the biped robot based on learning how to predict disturbances,combines the single rigid body model predictive control method with the deep reinforcement learning algorithm,which is a learning-based control method.This method models the effect of the biped robot’s swinging legs on the torso as disturbances of center-of-mass acceleration and rotational acceleration.And through the deep reinforcement learning algorithm,a policy that can predict the disturbances according to the motion state of the biped robot is obtained.The model prediction control algorithm gives the ground reaction forces based on the disturbances and the single rigid body dynamics,so as to resist the influence of the swinging leg and enhance the stability during walking.The new framework relaxes the restriction on the mass ratio for the legs,and broadens the scope of application of the single rigid body model in biped robots.