Self-learning visual path recognition

Abstract

In this paper the feasibility of equipping a mobile robot with the ability to learn a path, in real time, using principles based on insect-based navigation skills and a biologically plausible neural network model inspired by the "Conjunction of Local Features Network" (CLF network) through both real-world and controlled environment experiments is presented. Results shown for experiments on a prototype LEGO robot indicate that memory-based navigation using the proposed navigation network is suitable for simplified environments. 1 Introduction Until recently most approaches to robot visual navigation were based on symbolic representation of the world in terms of known structural information. One of the most fundamental problems with the symbolic paradigm is that a high degree of prior knowledge of the robot's environment must be known, thus reducing the structural complexity of the environment for which the robot can be operated in. The dependency on structural information makes this a...