I. Introduction
Dynamical models are used in planning, control and state estimation of robotic platforms; however, finding a model which describes a real physical platform well is nontrivial. Despite the amount of effort one might put into designing a high-fidelity dynamical model, there exists yet another challenge that arises: such models typically have a large number of parameters which must be tuned very accurately. For instance, when operating a ground vehicle around a turn at high speed, a small inaccuracy in the coefficient of friction could send the vehicle into slip through a discontinuous change in dynamics. Such behavior would be deleterious and potentially hazardous unless it were predictable, in which case it might be used to an advantage in e.g. guiding the vehicle through a tight corner. Even if an accurate set of parameters are chosen via expert tuning, some of these parameters would inevitably change in time due to mechanical degradation or environmental disturbances.