Saturday , August 18 2018

Evolutionary Method for Designing and Learning Control Structure of a Wheelchair

Imen Ben Omrane
Institut National des Sciences Appliquees et de Technologie INSAT, Centre Urbain Nord BP
676, Tunis, 1080, TUNISIA

Abderrazak Chatti
Institut National des Sciences Appliquees et de Technologie INSAT, Centre Urbain Nord BP
676, Tunis, 1080, TUNISIA

Pierre Borne
Ecole Centrale de Lille ECLille, Cite Scientifique Villeneuve-d’Ascq
Lille, 59650, FRANCE

Abstract:

This article describes an aspect of evolutionary robotics for trajectory tracking. We will combine genetic algorithms with neural networks for modelling and controlling a wheelchair for disabled people. The interest of the hybridization of Neural Networks (NN) with Evolutionary Algorithms (EA) in robotics is based on the observation that a local search by a gradient descent method is replaced by a global search performed by EA. The gradient descent methods are subject to variations in performance due to the initial position of the NN, which sometimes leads to a convergence towards local minima. In contrast, the proposed evolutionary methods provide a global research of both the structure and the weights of the neural net. The control structure used for robot trajectory tracking control is based on the Internal Model Control (IMC) which direct neural model was learned with our new EA.

Keywords:

Evolutionary Robotics, trajectory tracking, evolutionary algorithms, Neural Networks, direct neural model, mobile robots, wheelchairs, Internal Model Control.

>>Full text
CITE THIS PAPER AS:
Imen BEN OMRANE, Abderrazak CHATTI, Pierre BORNE, Evolutionary Method for Designing and Learning Control Structure of a Wheelchair, Studies in Informatics and Control, ISSN 1220-1766, vol. 21 (2), pp. 155-164, 2012.

1. Introduction

The most common technique for training neural networks is by studying the variations of the gradient descent. This technique suffers from well known problems, essentially local minima. Hence, there is a need for more efficient and effective methods to determine network weights and structure of NN. These methods combine another biologically inspired technique, the technique of genetic algorithms with neural networks.

Developed by John Holland [3], a genetic algorithm is a biologically inspired search technique. In simple terms, the technique involves generating a random initial population of individuals, each of which represents a potential solution to a problem. Members of the population are then selected for reproduction based upon fitness function, and a new generation of potential solutions is generated. The process of evaluation, selection, and recombination is iterated until the population converges to an acceptable solution.

Several hybridization of genetic algorithm and neural network exist; the most common among them are the determination of network weights by the use of genetic algorithms [7], [11], [14] and the evolutionary design of the network architecture [1], [4], [l0], [12], [15].

Evolutionary methods have found applications that span the range of architectures for intelligent robotics. For example, evolutionary algorithms have been used to learn rule sets for rule-based autonomous agents, topologies and weights for neural nets for robotic control [8], [9], [14], [15] fuzzy logic control systems [19], and rules for behaviour-based robots [13], [2].

In this paper, the NN is learned with EA and is using Internal Model Control (IMC) structure in order to benefit from performances of each one of them to control a wheelchair for disabled people. Inverse and direct neural models of the wheelchair are elaborated and a trajectory tracking is realised.

We have chosen IMC strategy of control because it constitutes a powerful strategy of control for complex systems, thanks to its simplicity of implementation, its robustness toward the errors of modelling and its facility of adjustment. [20]

We make lateral control for position and direction of the wheelchair for disabled people from angle control u. The speed of the wheelchair will be considered low and constant.

This is justified by the fact that the wheelchair is not subject to run at variable and high speed. Thus, dynamic model will not be considered.

Learning for modelling will be performed by two techniques:

  • Standard: back propagation gradient.
  • Evolutionary: hierarchical genetic algorithm.

The evolutionary method has simultaneously to determine the structure of NN and to learn it by minimizing the squared error

a05f01

Sd is a desired output and Sr a network output.

We will compare these two methods of learning by simulating prediction error. Finally, we apply the IMC using the models already developed.

References:

  1. ANGELINE, P. J., G. M. SAUNDERS, J. B. POLLACK, An Evolutionary Algorithm that Constructs Recurrent Neural Networks, IEEE Transactions on Neural Networks, Vol. 5, No. I, 1994.
  2. IMEN, A., A. CHATTI, Reactive Control Using Behavior Modelling of a Mobile Robot, International Journal of Computers, Communications & Control, Vol. II, No. 3, 2007, pp 217-228.
  3. HOLLAND, J., Adaptation in Natural and Artificial Systems. Ann Arbor: The University of Michigan Press, 1975.
  4. ESPARCIA-ALCDZAR, A., K. SHARMAN, Evolving Recurrent Neural Network Architectures by Genetic Programming. In Genetic Programming 1997: Proceedings of the Second Annual Conference.
  5. FATTOUH, A., Y. DADAM, D. PAHM, Matlab Based 3D Dynamic Model of a Powered Wheelchair, Innovative Production Machines and Systems Conference, 2008.
  6. BONCI, A., S. LONGHI, A. MONTERIU, M. VACCARINI, Navigation System for a Smart Wheelchair, Journal of Zhejiang University SCIENCE, 2005.
  7. KRISHNAN, R., V. B. CIESIELSKI, 2DELTA-GANN: A New Approach to Training Neural Networks Using Genetic Algorithms. In Proceedings of the Fifth Australian Conference on Neural Networks, 1994.
  8. MEYER, J. A., Evolutionary Approaches to Neural Control in Mobile Robots, Systems, Man, and Cybernetics, IEEE International Conference Vol. 3, 1998, pp. 2418-2423
  9. GREFENSTETTE, J. J., Evolutionary Algorithms in Robotics. In International Symposium on Robotics and Manufacturing, New York, 1994.
  10. MILLER, G. F., P. M. TODD, S. U. HEGDE, Designing Neural Networks using Genetic Algorithms. In Proceedings of the Third international Conference on Genetic Algorithms, 1989 (ICGA-89).
  11. MONTANA, D. J., L. D. DAVIS, Training Feedforward Networks using Genetic Algorithms. In Proceedings of the international Joint Conference on Artificial intelligence, 1989 (IJCAI-89).
  12. MORIARTY, D., R. MIIKKULAINEN, Hierarchical Evolution of Neural Networks. Technical Report A196-242, Department of Computer Sciences, The University of Texas, at Austin, Texas, USA, 1996.
  13. PASSOLD, F., Applying RBF Neural Nets for Position Control of an Inter/Scara Robot, International Journal of Computers, Communications & Control, Vol. IV, No. 2, 2009, pp. 148-157.
  14. NADI, A., S. S. TAYARANI-BATHAIE, R. SAFABAKHSH, Evolution of Neural Network Architecture and Weights Using Mutation Based Genetic Algorithm, Proceedings of the 14th International CSI Computer Conference (CSICC09),
  15. BEN OMRANE, I., A. CHATTI, Training a Neural Network Using Hierarchical Genetic Algorithm for Modeling and Controlling a Nonlinear System of Water Level Regulation, Nonlinear Dynamics and Systems Theory, Vol. 10, N° 1, 2010, pp. 65-76.
  16. OSORIO, R., J. A. ROMERO, M. PEÑA, I. LÓPEZ-JUÁREZ, Intelligent Line Follower Mini-Robot System, International Journal of Computers, Communications & Control Vol. I, No. 2, 2006, pp. 73-83.
  17. CHATTI, A., I. AYARI, P. BORNE, M. BENREJEB, On the Use of Neural Techniques for Path Following Control of a Car-like Mobile Robot, Studies in Informatics and Control: Vol. 14, No. 4, 2005, pp. 221-234.
  18. TANGOUR, F., P. BORNE, Presentation of Some Metaheuristics for the Optimization of Complex Systems, Studies in Informatics and Control: Vol. 17, No. 2, 2008, pp. 169-180.
  19. ALASTY, A., H. N. PISHKENARI, S. H. MAHBOOBI, Trajectory Tracking of a Mobile Robot Using Fuzzy Logic Tuned by Genetic Algorithm, International Journal of Engineering Vol. 19, No. 1, November 2006, pp. 95-104.
  20. BEL HADJ, S. A. NAOUI, D. JARBI, M. BENREJEB, Neural Internal Model Control of a Mobile Robot, Journal of Automation & Systems Engineering, Vol. 2, Issue 3, September 2008.

https://doi.org/10.24846/v21i2y201205