Modelado e identificación de vehículos móviles usando modelos de baja complejidad basados en datos

Autonomous vehicles are attractive platforms for several applications such as aerial, terrestrial, aquatic and underwater applications. The system modeling and identification is paramount to the success of the model-based controllers. Reliable control strategies require faithful models to achieve a good performance. Classical modeling represents the system dynamics by ordinary differential equations and often requires extensive human knowledge. Many times, the dynamics are complex and nonlinear and also many simplification assumptions are made during system modeling. In this paper we compare different data-driven techniques to model the system dynamics. Particularly, we use the well-known artificial neural networks, multilayer perceptron and radial basis functions, as well as Gaussian process regression to model the vehicles dynamics. These techniques learn the underlying structure of the vehicles dynamics from the experimentally measured data offering a natural framework to incorporate the unknown nonlinearities. In this paper a terrestrial vehicle is identified, the Pioneer 3 at and the obtained model is validated with the real vehicle.