Many-Objective Motion Generation Method for Redundant Manipulators by Solving Pathwise Inverse Kinematics

Abstract

Modern robots are required to operate in complex environments and perform diverse tasks, resulting in redundant degrees of freedom (DoF) for flexibility. However, managing redundancy is challenging due to the high-dimensional and non-convex nature of robotic kinematics. When executing complex tracking tasks, redundant robots must handle non-convex constraints while maintaining many objectives, such as balancing and obstacle avoidance. This paper models the pathwise inverse kinematics of redundant mechanisms as a multi-objective nonlinear optimization problem. We propose an efficient gradient-free optimization method named MoeIK, which demonstrates strong multi-objective balance, rapid global convergence, and adaptability. Our approach enhances the method by integrating relaxation dominance, adaptive interval search strategies, and a restart strategy, significantly improving performance in overcoming many-objective optimization challenges. We compared MoeIK with RelaxedIK, Trac-IK, and BioIK across multiple trajectories on various redundant robots, and the experimental results demonstrate that our algorithm exhibits better multi-objective balance capabilities and supports real-time computation.