Arithmetic for Multi-joint Redundant Robot Inverse Kinematics Based on the Bayesian-BP Neural Network

摘要

Based on the combination of Bayesian methods and BP neural network, a Bayesian-BP neural network model is presented to solve multi-joint redundant robot inverse kinematics in the continuous path. After inspecting joint's movement rules of multi-joint robot, the knowledge distribution of nature connection tied in Bayesian methods is used to formalize all kinds of priori information and implement the durative process of learning. With BIC criteria, using a two-stage cross optimization method to amend parameters of network weights and improves the learning speed of neural networks, convergence and accuracy. The simulation shows that Rotations or move changes of per joints are smooth in the multiple working points of the robot continuous path, and the error of the method could be less than 0.001.