Inverse kinematics solution method for robot based on teaching learning

Abstract

The invention belongs to the field of robot kinematics solution, and particularly discloses an inverse kinematics solution method for a robot based on teaching learning. The method comprises the following steps of collecting the joint angle of N sets of robots and the Cartesian position and Euler angle of an end effector, and obtaining a data set f; optimizing the data set f to obtain the data setf1; iteratively solving the data set f1 to obtain Gaussian mixture model parameters with 2-50 Gaussian models; calculating Bayesian information criterion values corresponding to Gaussian mixture models, drawing a Bayesian information criterion curve, and determining a Gaussian number k according to the curve; solving the Gaussian mixture model with the Gaussian models of which the number is k; using the model parameters to conduct Gaussian regression treatment on the Cartesian position and the Euler angle of the data set f1, and solving an inverse kinematic joint angle value; comparing the joint angle value with the joint angle value in the data set f1, and correcting the number of the Gaussian models to obtain desired solution accuracy. The method has the advantages of fast calculation speed, strong applicability, good real-time control effect and the like.

Description

technical field [0001] The invention belongs to the field of solving robot kinematics, and more specifically relates to a method for solving robot inverse kinematics based on teaching and learning. Background technique [0002] The solution of inverse kinematics of robot is the basis of real-time control of robot. It is directly related to robot offline programming, trajectory planning, motion control and other tasks. For robots of any configuration such as nR robots, PR robots and flexible robots, only the traditional robot kinematics expression is used to solve the problem by using data-driven inverse kinematics. In order to realize the mapping relationship between Cartesian space and joint space of any robot configuration, it is of great significance for real-time control. [0003] Patent CN201410121131.5 discloses a general solution method for inverse kinematics of a multi-degree-of-freedom robot, using the common space theory to establish the general kinematic equation...

AI Technical Summary

Problems solved by technology

[0004] Aiming at the above defects or improvement needs of the prior art, the present invention provides a method for solving robot inverse kinematics based on teaching and learning. By adopting the machine learning algorithm of Gaussian mixture model, the robot inverse kinematics dependent on analytical expressions can be solved. Kinematics solves the problem of slow calculation speed, has the advantages of fast calculation speed, strong applicability, and good real-time control effect, and can be applied to the control of various types of robots

Images