Using Particle Filters to Solve the Problem of Symmetric Multiple Solutions in Robot Inverse Kinematics

Abstract

Encountering multiple solutions is the most severe problem in inverse kinematics of robot arms. Multiple solutions usually appear in symmetrical forms; it means that the joint angles vary greatly between two symmetrical postures. During the operation of the robot arm, the path through any joint space should be smooth. Otherwise, there will inevitably be situations where the joint angle jumps significantly between two adjacent timesteps. In this paper, we propose a novel method to solve robotics inverse kinematics for manipulators. This method utilizes particle filters to track the possible postures of planar robot arms given the positions of the end effectors. In this way, the poses of the robot arms are modeled as particles within these filters. The particle filter algorithm is an iterative process, which tracks the angles of joints by averaging the particles and regenerates populations so that particles can converge to the arm poses. In addition, particles can remember the previous values after regeneration, so that the inverse poses do not follow a non-differentiable path in the joint spaces. To verify the effectiveness of the proposed method, we implemented a simulator and tested the performance of the particle filters in a nonlinear end terminal path.