Angular Memory and Supervisory Modules in a Neural Architecture for Navigating NOMAD
This paper presents a neural modular architecture for navigating mobile robots. The proposed architecture, based on functional task division, has been shown to be efficient in previous work [1, 2].
The traditional difficulties associated with monolithic neural networks are circumvent by introducing a neural modular architecture, developed for the NOMAD mobile robot. The modularity introduced retrieves all the available information, minimizing the incoherence in the training sets, that arises from conflicting training patterns.
After presenting the modular architecture, two additional modules are introduced: Supervisory and Angular Memory. The combination of all defined modules was first tested in simulation and then with the real robot, providing a solution to the problem of navigating NOMAD mobile robot to an objective in an unknown environment.
KeywordsMobile Robot Real Robot Ultrasonic Sensor Modular Architecture Neural Architecture
Unable to display preview. Download preview PDF.
- C. Silva, M. Crisostomo, B. Ribeiro, “MONODA: A Keural Modular Architecture for Obstacle Avoidance Without Knowledge of the Environment”, in Proceedings of the 2000 International Joint Conference on Neural Networks, 24–27 July 2000.Google Scholar
- C. Silva, M. Crisóstomo, B. Ribeiro, “A Modular Learning Architecture for Navigating NOMAD Mobile Robot”, 8th International Conference on Information Processing and Management of Uncertainty in Knowledge Based Systems, 3–7 July 2000.Google Scholar
- A. Schmidt, Z. Bandar, “A Modular Neural Network Architecture with Additional Generalisation Abilities for Large Input Vectors”, in Proceedings of the ICANNGA97, Springer-Verlag, pp 40–43, April 1997.Google Scholar
- A. Sharkey (Ed.), “Combining Artificial Neural Networks”, Springer-Verlag, 1999.Google Scholar
- S. Haykin, “Neural Networks — a comprehensive foundation”, Prentice Hall, 1999.Google Scholar