PATH PLANNING OF HUMANOID ROBOTS FOR STABLE MOTION USING REINFORCEMENT LEARNING BASED FUZZY LOGIC CONTROLLER

Abstract

The locomotion and route planning of humanoid robots has become one of the utmost promising areas of research as humanoids are used more frequently in various fields of industrial automation and manufacturing. In this study, an effective solution for the navigation of humanoid robots is promised by fuzzy logic controllers. The fuzzy rule-base built and tuned by a human operator must be kept accurate, consistent, and complete, but this is challenging. One kind of machine learning is the reinforcement learning approach. The field of robotics frequently employs this strategy. When we suppose that the sole information collected is a scalar signal which is a reward or punishment, it seeks to automatically learn the fuzzy rules and to build a control law for a humanoid robot in an unfamiliar environment. The robot navigation in this study makes use of fuzzy controllers and the Q-learning algorithm. The outcomes of the simulation demonstrate appreciable improvements in the robot behaviors and learning rate in compare to latest state of art techniques available in recent literature. The outcomes are evaluated and discussed.

Keywords

• humanoid locomotion
• reinforcement learning
• fuzzy logic
• webots