Alopex algorithm for adaptive control of dynamical systems

An adaptive control scheme for dynamical systems using Alopex neural networks is investigated. Using a linear underwater vehicle dynamics model, it is shown that the Alopex algorithm can be effectively used as a learning algorithm for adaptive control of complex, unknown dynamical systems. The algorithm is computationally simpler, faster and better suited for hardware implementation than other algorithms such as backpropagation. Simulation studies with an underwater vehicle model show that these networks perform well as adaptive controllers. The algorithm is used only in its very basic form. Since the algorithm works on the basis of a global performance measure, the position of the controller in the system is irrelevant. This makes it better suited for complex system architectures