Knowledge Based Reinforcement Learning Robot in Maze Environment

Abstract

A simple approach for knowledge based maze solving is presented for a mobile robot. The artificial intelligence concept like reinforcement learning technique is utilized by the robot to learn the new environment. The robot travels through the environment and identifies the target by following a set of rules. After reaching the target, the robot returns back through the optimum path by avoiding dead ends. For achieving this, the robot uses a line maze solving algorithm which uses a set of replacement rules to replace the wrong paths travelled with the correct ones. The algorithm for this maze solver is qualitative in nature, requiring no map of environment, no image Jacobian, no Homography, no fundamental matrix, and no assumption. The environment is accessible, deterministic and static. The working procedure of this project consists of line path following, mobile robot navigation, knowledge based navigation, reinforcement learning.