Modelling of an AUV with Voith-Schneider vector thruster

First principles physics models are generally used in system identification of Autonomous Underwater Vehicles (AUVs). These models, through different parameters, capture the effects of hydrodynamics, inertial weight and other important characteristics. Due to the large number of parameters, which can number to hundreds, it is difficult to estimate such models. Moreover, AUV capabilities like thrust vectoring increases the non-linearity of the model. We suggest solving the problem of modelling AUVs with the help of a rectifier activated multilayer perceptron, making use of their motion data and control inputs. We also provide details on the optimisation of our model and compare its performance with that of a standard system identification technique. Although the rectifier neural network's performance was tested for a typical streamlined AUV with a Voith-Schneider thruster, the model presented here is general and can be easily extended to other systems.