Dynamic control of a remotely operated vehicle using a neural network

Abstract

The control of a remotely-operated underwater vehicle to maintain a prescribed depth in shallow water under irregular surface waves is realized through the application of the Robust Adaptive Neuro Controller, a composite control system incorporating—together with the conventional control algorithm—a neural network controller. This network bestows a learning capability on the system, allowing it to deal with unanticipated disturbances that would otherwise cause erroneous behavior of the vehicle. The effectiveness of this application is verified through mathematical simulation of a model vehicle's behavior, through experiment in a model basin, and through simulation of the behavior of an actual remotely operated vehicle in shallow water under irregular surface waves. Graphic data representing the learning process undergone by the neural network distinctly indicate the rising output from the network with the progression of learning, and the vehicle's depth variation traced in terms of the mean square error vividly show the diminution of deviation from the prescribed depth obtained with application of the neural network. Thus controlled to maintain constant depth, under-water vehicles with power supplied externally through a tether for propulsion and for heavy-duty operations should consolidate their advantage for such activities as maintenance of submarine structures and surveys in deep or hazardous water.