Driver modeling

Models of the human driver are needed as a substitute to live studies in order to drive simulated vehicles in simulated environments. Normally, those models are controller models that are based on differential equitation just like the models of vehicle dynamics.  This type of driver model can simulate the “driving” as such pretty well; however, they are not suitable to simulate real human behaviour.

HFC has been working for years to develop “manlike” driver models. This is realized by taking context-related dead times into consideration, by combining steering- and controlling strategies and finally, by turning away from the “reference-trajectory paradigm”.

This leads to 4 basic assumptions of the “manlike” modeling.

• A human being does not plan trajectories he wants to follow, but rather action sequences which are related to consequences. The action sequence offers the possibility of correction.
• A human action consists of cycles of constant planning, controlling and correction.
• A task is solved by correcting ( regulating ) behavioral parameters.
• The regulation takes place on the parametric level as long as possible; an adjustment only takes place if there are unexpected influences or huge deviations occuring. If that happens, the driver reacts with a lag.

Every driver has his own strategy; they become visible in the way tasks are handled if the drivers have the appropriate liberties.

A discreet action model results that is comprised of a discreet series of periods dedicated to prediction and action plus a parallel process of regulation. The illustration below depicts the time scheme for the maneuver “double lane change”.

One core theme of the development of a model is the inclusion of knowledge from human sciences, especially psychology, into the mathematical modeling.  That’s exactly why the development is carried out by a consequently interdisciplinary team.  An example of a “non-technological” approach to modeling is the development of an optimal reference route for the company “Tesis Dynaware”, Munich. Starting from interviews with experienced rallye and racing drivers,  strategies of choosing routes were identified and translated into predictable algorithms. Through the concept of discreet entry points, apex and exit points, with a classification system of curves and curve combinations and a widespread set of transition rules an algorithm was created that produced reference routes resembling those of real drivers. Different driver strategies and the influence of different motorizations can be controlled via a corresponding parameter. That way, one can specify “Combat courses” (direct line) and “Race courses” as well as in-between strategies.