Human Driving Behavior Analysis and Model Representation with Expertise Acquiring Process for Controller Rapid Prototyping

Driving car means to control a vehicle according to a target path, e.g. road and speed, with some constraints. Human driving models have been proposed and applied for simulations. However, human control in driving has not been analyzed sufficiently comparing with that of machine control system in term of control theory. Input - output property with internal information processing is not easily measured and described. Response of human driving is not as quicker as that of machine controller but human can learn vehicle response to driving operation and predict target changes.

Driving behavior of an expert driver and a beginner in an emission test cycle was measured and difference in target speed tracking was looked into with performance indices. The beginner's operation was less stable than that of the expert. Transfer function of the vehicle system was derived based on linearized model to investigate human driving behavior as a tracking controller in the system. A driver's response delay is too large to correct speed error with feedback. Operation in advance to the target change, i.e. feed forward control, is concluded to be an optimal control comparing with predicted control theory and optimal path search programming results. It is proposed that an expert driver sets a new target for operation through learning during repeated driving and that inverse model of vehicle property also is formed within a driver. An expert driving with an optimal target is shown as behavior of a driving agent.