Navigation of multiple mobile robots using a neural network and a Petri Net model

Abstract

This paper describes the use of a hybrid system comprising a multi-layer perceptron and a Petri Net model to control the navigation of multiple mobile robots in an unknown and cluttered environment. The multi-layer perceptron is trained with examples representing the typical static obstacles to be encountered by the robots and the evasive actions to be taken. The Petri Net model embodies the rules describing how the robots should move in order to avoid colliding against one another. The hybrid system has been implemented in simulation software as well as in actual mobile robots. The paper presents the results of simulation and practical tests that demonstrate the ability of groups of robots incorporating the proposed navigation system to negotiate various types of obstacles and find targets successfully.