Modular neural networks architecture for navigating mobile robot in changing environments

Abstract EN

This paper addresses the navigation problem of a mobile robot in unknown indoor environments.

Most neural network (NN) approaches to this problem focus on a monolithic system, i.

e., a system with only one neural network that receives and analyses all available information, resulting in conflicting training patterns, long training times and poor generalization.

The work presented in this article circumvents these problems by the use of modular architecture (“divide and conquer” strategy( combining behavior based environment classification and several behaviors based reactive navigation.

The behaviors are learned by modular neural networks (MNN), coordination between these various behaviors is done at the same time in a cooperative and competitive way.

To check the validity of our approach, a graphic interface is developed.

It enabled us to test the proposed architecture in several different situations which approach reality.

In all the cases, the results obtained are very encouraging, and illustrate the effectiveness of this architecture.