COMPARISON OF ARCHITECTURES FOR NEURAL-BASED MODELING OF LATERAL VEHICLE DYNAMICS1

AbstractNonparametric modeling of an unmanned ground vehicle is performed as part of a larger effort to develop online state estimation techniques based on artificial neural networks. The output of these networks will provide estimation of the states of the lateral dynamics. The architecture of these networks is considered in an effort to provided an adaptable, generalized network that provides accuracy comparble to that of traditional modeling techniques. The feedforward, fully-connected feedforward and cascade architectures are considered within. A Box-Jenkins model is developed to provided a parametric baseline as a means of comparison.