Model of Human Vehicle Driving - a Predictive Nonlinear Optimization Approach

When driving a vehicle the human acts as a controller in a highly dynamic environment. Thus human behavior in that control loop has to a large extent been described using control theoretical methodology. We develop a driver model, in which driving is seen as a model predictive control task in such a way that the driver accumulates knowledge about his/her vehicle's handling properties. He/she builds a model out of that knowledge and uses it to predict the vehicle's future reactions on his/her control inputs. The human's behavioral optimization is reflected in the driver model by using that prediction model in order to optimize control inputs such, that a set of criteria, which reflect human well-being, are minimized. Prediction models and criteria depend on the current driving situation and on personal driver preferences. The principal properties of the driver model are discussed using very simple standard maneuvers like driving straight and cornering under different preferences. The method is then applied to a more complex track. The findings from that are backed up by experiments done in real world and in a driving simulator.