The Authors present a new approach to the modelling of human driving behaviour, which describes driving behaviour as the result of an optimization process within the formal framework of hybrid automata. In contrast to most approaches, the aim is not to construct a (cognitive) model of a human driver, but to directly model driving behaviour. We assume human driving to be controlled by the anticipated outcomes of possible behaviours. These positive and negative outcomes are mapped onto a single theoretical variable - the so called reinforcement value. Behaviour is assumed to be chosen in such a way that the reinforcement value is optimized in any given situation. To formalize our models we use hybrid automata, which allow for both continuous variables and discrete states. The models are evaluated using simulations of the optimized driving behaviours. A car entering a freeway served as the scenario to demonstrate our approach. First results yield plausible predictions for car trajectories and the chronological sequence of speed, depending on the surrounding traffic, indicating the feasibility of the approach.