Short-term visual mapping and robot localization based on learning classifier systems and self-organizing maps

Abstract

Ground wheeled autonomous robots like driverless cars are examples of applications which would assist humans on different tasks. From an explicit or emerging need, these systems have come to replace or assist drivers. Estimating the position is a primary function for intelligent vehicle navigation. Different existing solutions use high-end sensors. This paper proposes to augment the autonomy level of a mobile robot based on learning classifier systems and self-organizing maps. From a simple monocular system, whilst the classifier system leads the robot for topological localization tasks, the neural network is applied as a short-term visual memory for internal representation of the environment. These two concepts are presented as separate approaches, wherein each method performs a specific task for the robot's trajectory control.