Novelty Detection Based on NCD for Navigation Control of Mobile Robots

Abstract

In the path planning and its execution, a path must be defined from a source point to a destination point and then the navigation control task is executed. This study proposes a new method of novelty detection, named NDN (Novelty Detection with Normalized Compression Distance), to contribute in the navigation control of mobile robots. First, a genetic algorithm for path planning, based on a topological map, is used to generate a set of actions which the robot must perform to achieve the goal. Each action is a different reactive behavior designed for characteristic places of the environment, such as corridors, curves or intersections. Next, NDN is used to recognize such different environmental characteristics, activating the action whenever the method recognizes a context change. The experiments were performed in the Player/Stage robotics simulator and in a real indoor environment. NDN performance is compared with a Neural Network trained to recognize context changes in the same environment. The results reported indicate that NDN is a promising approach to be used in navigation control for mobile robots with the advantage of detecting a context change just knowing a initial state (corridor) from the environment.