Derivation of Optimal Parameters According to Path Curvature by Policy Gradient Reinforcement Learning

Machine learning technology has become important in many fields. In robotics, for example, the robot learns the parameters autonomously according to the environment and runs. In the case of a robot that moves autonomously on wheels, the travel path and parameters such as speed and amount of turning must be adjusted. In this paper, we propose to find parameters that allow fast and smooth running along each path and switch between them according to the curvature. To find the optimal running parameters, we define the running state of the robot and propose a learning method that applies policy gradient reinforcement learning. We implement and verify the effectiveness of the proposed method on a real robot.