Real-Time Control of Autonomous Underwater Vehicles Based on Fuzzy Neural Network

A real-time control scheme based on fuzzy neural network (FNN) is proposed for the motion control of autonomous underwater vehicles (AUVs) in this paper, for which the dynamics of the controlled system need not be completely known. A real-time desired state planning (DSP) based a sigmoid reference model is introduced to assist the FNN to keep the track error in a low level, and that also can serve as teaching signal to guide the training of the network, which makes it possible to implement the real-time motion control with FNN. The designed multilayered neural network architecture involves a modified error back propagation (EBP) as the learning algorithm, which is implemented by using the error at the output of the vehicle instead of that of the network so that the weights can be effectively adjusted to maximally decrease the system error. Results of simulation studies on the "AUV-XX" simulation platform are performed to illustrate the effectiveness of the presented scheme.