Research On High Dynamic Motion Control Method Of Quadruped Robot Based On Force Control

In nature,quadrupeds such as cheetahs can run at a maximum speed of close to 110km/h,antelopes can jump across cliffs of 6 to 7 meters long and stand on palisades close to 90 degrees,and mules can deliver weights of 150 to 200 kg.Due to the characteristics of the quadruped's flexibility and terrain adaptability,the research on bionic quadruped robots has attracted more and more researchers' attention.Inspired by the Boston Dynamics' quadruped robots such BigDog,SpotMini,the high dynamic performance and strong robustness have led people to explore the great interest of quadruped robots.This paper is based on the SCalf series of hydraulic quadruped robots and electric quadruped robots developed by the Center for Robotics,Shandong University.This paper studies the modeling and control of quadruped robots from hydraulic actuator modeling to single-leg compliance control to quadruped robot overall control.Based on the specific robot platform and motion performance requirements,three types of quadruped robot control methods based on inverted pendulum model,optimal foot force distribution control and model prediction control are proposed.The main research contents are as follows:1.Model the hydraulic cylinder execution unit controlled by the flow valve.The non-linear hydraulic output force affected by multi-parameters is linearized to obtain a simplified model of output force control to achieve good joint torque servo.Furthermore,based on the hydraulic single-leg test platform,the method of active compliance control based on position impedance and virtual model control were studied.The single-leg compression experiment and free-fall experiment verified the good effect of the proposed method on active compliance and reducing foot impact force.2.In order to solve the shortcomings of large foot impact force and inflexible movement caused by position control of the SCalf robot,a simplified control method based on the inverted pendulum model was proposed.Integrating the attitude angle of the robot's torso into the design of the foot trajectory makes the robot more adaptable to external force disturbances and uneven terrains.At the same time,the active compliance control method of the leg virtual model control method is used instead of the position control to improve the compliance of the interaction between the foot and the ground.In order to improve the robot's ability to adapt to uneven terrain,such as slopes,a decoupling terrain adaptation and torso adaptation strategy is proposed to improve the robot's climbing ability.The effectiveness of the proposed method is verified by the SCalf-II robot experiments3.In order to improve the robot's dynamic movement ability,the trot and flying trot gait control methods based on the optimal foot force distribution are proposed The relationship between the six-dimensional force and moment at the torso center and the reaction force of the support legs is established.The torso motion is regarded as a spring damping model to design the virtual force and moment at the center of mass.The virtual force and torque are distributed to the supporting legs through a quadratic optimization method to achieve the optimal foot force distribution with the target torso movement constrain.Virtual model control of the leg is used in the swing phase to achieve stable trot gait movement.On this basis,a periodic jump motion of the torso is designed based on the kinetic potential energy conversion principle to achieve flying trot gait motion.For the application of physical platform,the torso motion is mapped to the foot motion and the inverted pendulum model control method is used to maitain the stablility of the roobt.Based on this method,the SCalf-III robot achieves stable flying trot motion in place4.In order to achieve the multi-gait stable motion ability of a quadruped robot,a simplified dynamic equation at the center of mass of the robot's torso is established,and a model predictive controller is constructed with the robot's torso position,velocity,torso angle and angular velocity as state variables.Each optimization process is based on the current robot state information to predict the robot state for the next 10 robot phases.The MPC problem is discretized into a multi-constrained QP problem,and the ground reaction force is solved as the feedforward force of the supporting legs Because each leg performs independent force analysis during the modeling process,the support phase and swing phase of each leg can be designed freely during the gait design to obtain different motion gaits.For the electric quadruped robot pre-researched by the research group,the stability and robustness of gaits such as trot,flying trot,bound,gallop,and walk were verified through dynamic simulation.At the same time the terrain adaptability was verified by up and down 20-degree slopes.