Research On Design And Stand Control Optimization Of Quadruped Robot

As a multi-articulated robot,quadruped robots can replace humans in various complex unstructured environments to complete some work,such as desert management,no man’s land rescue,outer space exploration,etc.In recent years,quadruped robot technology has developed rapidly,but there are shortcomings in control accuracy and effective energy consumption.Therefore,the research on quadruped robots mainly focuses on intelligent control and effective energy consumption improvement.Intelligent control of quadruped robots needs to be carried out with the help of a virtual environment,but it is difficult to obtain parameters such as dynamics.The effective energy consumption improvement is mainly carried out from the two aspects of structural optimization and control optimization,structural optimization hopes to ensure the movement ability of quadruped robots while using a more streamlined design scheme,and control optimization hopes to achieve quadruped robot walking with minimal energy cost.This paper studies the design simplification and standing control optimization of quadruped robots,and the main contents can be summarized as the following three points:Firstly,in order to obtain the relevant parameters of quadruped robot required for theoretical modeling and simulation model,a lightweight quadruped robot structure is designed based on the body structure of the quadruped robot,and kinematic parameters such as the coordinate position between each component of the single leg of the quadruped robot are obtained,and the theoretical calculation model of the foot end coordinate position of the quadruped robot is established.The kinetic parameters such as mass,center of mass coordinate position and inertia tensor of each rigid component are calculated by using Solidworks software,and a simplified dynamic model of quadruped robot is established.A geometric collider was designed for the Mujoco simulation environment,and a quadruped robot simulation model was built in the Mujoco environment.Secondly,in order to improve the effective energy consumption of quadruped robot,a simplified neural network structure of force system is proposed by replacing the autoencoder decoding network with the spatial arbitrary force system balance equation by drawing on the principle of spatial force system simplification.Based on the network structure,the components of the dataset suitable for the distribution of virtual centroid force systems are analyzed,and the force system is formulated to simplify the neural network training process.Through experiments,the loss function and learning rate update method suitable for the task of centroid force system assignment are obtained,and finally the neural network that can be used for centroid force system distribution is trained.The results of suspension control experiment and trajectory tracking control experiment show that the trained simplified neural network can be used to assign the centroid force system,which can reduce the excess force by40% compared with quadratic programming.Finally,in order to improve the accuracy of virtual model control of quadruped robot,an experiment of controlling quadruped robot based on the simplified neural network model of force system is carried out.Based on the virtual model algorithm,the centroid force system distribution method based on the simplified neural network of the force system is integrated,and the centroid force system of the quadruped robot is assigned to each leg as compensation for the end force of the single leg.At the same time,the method of assigning the centroid force system by the quadratic programming method is compared.Experimental results show that the simplified neural network of force system can assign the centroid force system to compensate for the foot force and effectively improve the control accuracy.