Model Structure and Learning Process for a Driver Model Capable to Improve Driving Behavior

Vehicle electrification has been extended rapidly in recent few years and development work for that has been added to conventional vehicles development. Model based development (MBD) methodologies have been adopted widely. A dynamic driver model is required for controller design considering driver’s behavior and for verification with SiLS and HiLs in the MBD process. Some higher response and multi-variable control systems can be constructed with electronic devices. However, human control is not so quicker and not capable to handle multi states. There have been a lot of published papers regarding to driver models. Structure of the driver model with constrains of human property and learning process seems to be under study. Authors have investigated driver models for target speed tracking driving in emission test cycles in which the target is clearly defined. Taking account of constrains with driver’s response and information processing capability, a driver model structure, with feed forward operation based on prediction and additional error feedback correction, is introduced. A learning algorithm to obtain inverse vehicle property for the feed forward control is proposed. Knowledge which enables to select features to be learned and condition for stable learning process are discussed. Numerical simulation illustrates driving behavior from a beginner to an expert with the driver model. Further, it is shown that speed tracing driving performance with a novice driver model could be improved when vehicle property is changed, e.g. an internal combustion engine is replaced by an electric motor. It is supposed that the proposed method is also applicable to development process for a lower order and rower sample rate controller with adaptation functionality.