Experience-based Problem Solver for Robot System Design

Abstract

In this study, an experience-based problem-solving method was developed to design robotic systems, including conveyors, bases, sensors, and robots. Experience reuse involves selecting the most “useful” experience from a dataset and reusing it to query new problems. To solve this robot system design problem, the environmental components are arranged appropriately, and the path length planned by the motion-planning algorithm is considered as the evaluation criterion. Therefore, a case-injected genetic algorithm (GA) is introduced as an experience-based optimization problem solver for robot environment design. The motion and path length of the robotic arm calculated from the experience-driven random tree (ERT) algorithm are considered performance indices in the environment arrangement of the robot system. In this study, standard and experience-based optimization methods and motion planning methods were combined to solve the proposed robot system design problem. These four combinations of methods were compared in terms of computation time and path length. Simulation results demonstrate that experience reuse in different aspects has different focuses, the optimization aspect has a more significant impact on the reduction of calculation time, and the motion planning aspect has a greater impact on path length.