発見と学習によるロボットの動作最適化の研究発見と学習によるロボットの動作最適化の研究

抄録

This study proposes a method of generating the optimized motion of robots by heuristic and learning program. Previously, we proposed a heuristic program that can determine changes of the displacement of a robot with respect to time, in order to reduce the electrical energy or peak current value occurred in motors. As the method can optimize the motion of a robot by evaluating the measured values, it does not need the kinetic and dynamic models of a robot, which are generally difficult to be obtained. However, the proposed method needs to drive robots every time, when determining the appropriate motion. Thus, this study proposes the method for robots to learn the appropriate motion from the data occurred during heuristic process. In concrete, robots construct neural network by using the data obtained with heuristic program as learning data. As a result, robots can determine the appropriate motion without heuristic process.