A Novel Approach to Coupling Terms to Avoid Obstacles in a Manipulator Movement Reproduction

Many people have attempted to generate specific movements based on the concept that neural networks in the brain and spinal cord create multiple sets of temporal templates. Dynamic movement primitives (DMPs) are inspired by the motion control of biological systems and can be mathematically represented as stable nonlinear dynamic systems in the form of motion primitives. One way to improve the work efficiency of robots in various industries is to leverage the ability of DMPs to generalize learned trajectories to enable them to perform a wider range of tasks. This study discusses obstacle avoidance techniques using DMPs and proposes a novel approach to obstacle avoidance. DMPs have the ability to generalize and have extensions that make them valuable in generalizing in unforeseen situations. Obstacle avoidance in DMPs has been approached in various ways, with previous research utilizing potential field methods as typical obstacle avoidance techniques. We added a formulated coupling term to DMPs to avoid obstacles. This novel approach proposes modeling obstacles as point clouds, objects surrounded by bounding boxes or smooth standard shapes, and adding a new coupling term to smoothly avoid obstacles without disrupting the existing reference movement’s topology while closely following a reference trajectory. This study also discusses the determination of the magnitude and direction of a desired repelling force against obstacles. Overall, this study discusses obstacle avoidance techniques using DMPs and introduces a novel approach that improves obstacle avoidance in DMPs. The goal of this study is to confirm the effectiveness of the proposed approach by implementing previous and newly proposed algorithms for semi-elliptical trajectories and applying them to robot manipulators.