Adaline neural network for online self-learning and adaptive control of a vehicle with thrusters in a fluid

We describe a simple artificial neural network, made up of Adalines that is able to learn to imitate very accurately the complex nonlinear dynamics of a thruster propelled vehicle that moves in a fluid. This is made possible by providing these Adalines with judiciously chosen nonlinear inputs. Such Adalines can be used as an internal model of the vehicle's behavior in an adaptive model motion controller. The vehicle thus becomes able of intelligent behavior, in that it can predict ahead of time what its dynamical state will be and can adjust its reactions accordingly. As we show, it is also able to adapt to the incidents that can modify its behavior or its environment. The well-known ability of Adalines to adapt rapidly endows this controller with all the features that can be wished for in artificial as well as in biological systems. We show that it can learns rapidly, by itself, to control the vehicle, which then flawlessly drives the vehicle; and it can further adapt to any change in the parameters that regulate its motion. In particular, we demonstrate its adaptation to changes in the current, drag constant, mass, buoyancy and maximum thruster forces. We also show that it is able to keep the vehicle perfectly still, without developing the limit cycle observed in most marine vehicles.