Inverse Kinematics Solution Of Manipulator Based On Adaptive Feature Weights Chain Neural Network

Intelligent robot technology is an important indicator to measure the national scientific and technological level and high-end manufacturing processes.Its research and industrial development has risen to a national strategy.The manipulator as a key component for robots to perform task operations has become one of the most widely used robot structures.As a key component for robots to perform task operations,the manipulator has become one of the most widely used robot structures,and its inverse kinematics solution has also become a hot research direction for researchers.Inverse kinematics has the problems of complicated solution process and low accuracy.With the continuous development of artificial intelligence technology,neural network has become an important tool for scientific researchers to solve the inverse kinematics problem of manipulator by virtue of its own advantages.However,the existing neural network solution has some defects: 1.It ignores the characteristics of the manipulator data itself,such as the serial characteristics of the joints;2.The network structure does not fit the actual scene,and most of the existing network structures are applied.In view of the above,this paper proposes an adaptive feature weights chain neural network to solve the inverse kinematics problem of the manipulator.The network mainly includes two core characteristics:1.Adaptive feature weights.The idea is mainly aimed at the complex spatial changes of the manipulator joint,and the importance of each input feature is also correspondingly changing.By continuously learning the changes in feature weights duringthe network training process,the network is always focused on more important features,thereby reducing error;2.Chain structure.This structure mainly aims at the serial characteristics of the joints of the manipulator,and uses the prediction information of the previous joint variables for the prediction of the subsequent joint variables,thereby taking advantage of the correlation between adjacent joints to speed up the network convergence speed.The neural network proposed in this paper is designed for the data characteristics of the manipulator,which is more suitable for the actual scenario and more interpretable.The final experimental results also show that the network has higher accuracy and faster convergence speed.The Mean Euclidean Distance(MED)between the actual position of the end of the manipulator and the predicted position is reduced to about 1cm,which effectively solves the problem of inverse kinematics of the manipulator.