Research On Walking State Recognition Of Hexapod Robot

With the development of science and technology,the demand of human operation in mountain reconnaissance,resource exploration,rescue and relief and other special environments is increasing.Because of its multi degree of freedom leg structure and discrete landing points,hexapod robot has become an excellent solution for complex environment operation,gradually replacing human to perform tasks in complex environment.Due to the particularity of the working environment,the hexapod robot will inevitably be impacted by the external environment in the process of walking,which may cause many problems,such as leg structural deformation,joint assembly relaxation,electronic component failure,etc.,resulting in the abnormal walking state of the robot,affecting the robot to continue to perform tasks.So it is very important to recognize the walking state of robot.Aiming at the problem of walking state recognition of hexapod robot,this paper collects the acceleration,joint angle and joint moment data of the robot in the process of walking,and uses machine learning algorithm to establish the recognition model to realize the recognition of walking state of hexapod robot.The main contents of this paper are as follows:(1)The structure design,gait planning and kinematics analysis of the hexapod robot are carried out,and the experimental prototype of the hexapod robot is designed.Three legged gait,four legged gait,five legged gait and three corresponding single leg injury gait are planned.The acceleration waveforms of the fuselage in different gait are drawn,and the joint angle and moment of the robot are calculated by the kinematic model.The relationship between the acceleration of the fuselage,joint angle,joint moment and the walking state of the robot is analyzed.(2)The acceleration sensor is used to collect the acceleration information of the fuselage,the joint steering gear is used to collect the torque and angle information of joint,and the data acquisition and storage program is designed.The experiment of robot walking is designed to collect the acceleration,joint angle and joint moment of the robot when walking in normal gait and single leg injury gait.(3)This paper analyzes the characteristics of standardization,normalization,smoothing,windowing and dimensionality reduction methods,and uses mean value,variance,skewness,kurtosis and correlation coefficient as feature extraction quantity,and the number of features is calculated.The principle of support vector machine,k-nearest neighbor,random forest and Ada Boost are analyzed.The parameters are selected by grid search combined with cross validation.The accuracy,confusion matrix,F1 score and recall rate are used as the measurement indexes of model recognition effect.(4)Data preprocessing and feature extraction are carried out by computer.Support vector machine,k-nearest neighbor,random forest and Ada Boost algorithm are used to build the model.Comparing the influence of different data processing methods on the recognition effect of the algorithm,one of the algorithms is chosen as the model of robot walking state recognition and experiments are carried out to verify the effectiveness of the proposed scheme.