Planning in driver models using probabilistic networks

The driving task is often described in terms of the strategic, the tactical and the operational level. Long-term conditions, e.g. time pressure or decisions for certain routes, are defined on the strategic level. The tactical level is more short-term oriented and deals with decision making out of several maneuvers like overtaking or following. Finally, on the operational level these decisions are translated into low level actions like steering, braking and accelerating. This paper mainly concentrates on models for the tactical level while keeping in view the interface for including strategic decisions. The driver models are used to simulate realistic driver behavior in a motorway environment. Two kinds of models are introduced, a reactive one which receives its inputs directly from sensors and a model whose decisions are based on a preceding planning step. Both models use Probabilistic Networks, to take into account the high level of uncertainty naturally occurring in behavior modeling. It is shown that the combination of planning and a probabilistic decision model is superior to a conventional non-planning system.