Using Particle Filter to Solve Problem of Symmetric Multiple Solutions in Inverse Kinematics of Manipulator

Abstract

In the inverse kinematics of the robot arm, the problem that needs multiple solutions is often encountered [1]. The multiple solutions often appear symmetrically. In practical applications, a set of solutions must be selected from the multiple solutions as the robot arm pose. Although the multiple symmetric solutions have reasonable poses reachable by the robot arm, the robot arm rapidly changes from one solution to another solution in two adjacent time samples in the process of arm movement if the solution is not selected correctly. The posture of the symmetric solution sampled last time causes the robot arm to rapidly change its posture. This causes damage to the mechanism and makes it oscillate wildly. In order to avoid the interference of multiple symmetric solutions, we abandon the traditional inverse kinematics method of algebraic or geometric and use the nonparametric Bayesian filter (that is, particle filter) based on the Monte Carlo method to process the robot arm inverse kinematics problem. The particle filter uses many random sample points and the design of importance weights to make the sample points converge to the optimal solution in the iterative process. We show how to use the design of importance weights so that the oscillation problem of multiple symmetrical solutions does not occur in adjacent time sampling during the movement of the robot arm.