Autonomous Navigation Robotic System to Recognize Irregular Patterns

This paper presents an approximation of navigation problem under unknown environment using reinforcement learning. The motivation is to represent a robot that can move in a world with streets and intersections. Each intersection has a different quantity of streets (irregular blocks). The selected algorithms were Q-Learning and Value Iteration. The robot was programmed only with Q-Learning and we developed a simulation with both of them. This simulation allows making comparisons in order to determinate in which situation each algorithm is appropriate.