Robotic system for reactive navigation in dynamic environments

Abstract

We introduce a mobile robotic system able to learn through reinforcement, which allows it to navigate within a dynamic environment (e.g. a room) avoiding any obstacle it might encounter. The robotic system must be able to locate and reach an established pattern, which has been previously learned. The learning system is implemented with two neural networks. Both neural networks use reinforcement learning by means of the Hebb rule. The mobile robot is implemented using a Lego Mindstorm NXT 1.0, with a design of twin-engine vehicle, 2 ultrasonic sensors, a touch sensor and a webcam. The system was programmed in C++ and uses a Bluetooth device to communicate the robot with the computer.