Abstract
This paper presents a control strategy with neural networks for the all terrain vehicle robot during the development by the authors. This robot consists of two modules: a normal vehicle with wheels or tracks, and a controllable mass with parallel links to control the vehicle motion. Since it is not easy to control the parallel link mechanism with a feed forward controller, a neural network is employed and its effective learning is discussed through numerical simulations. Furthermore, the use of another neural network is proposed to determine the desired trajectory of the controllable mass. The desired position of the mass is also discussed with respect to falling moment of the vehicle.
