Advanced Driving Assistance System (ADAS) provides warning and automatic control intervention to a human driver for accident avoidance and collision mitigation. To simultaneously minimize the human-machine conflict and enhance human trust in the driving assistance system, ADAS should take personalized driving characteristics into account to support a particular driver in an optimal manner. Typical Personalized Driving Assistance System (PDAS) trains a driver model offline via complicated machine learning algorithms and assumes that the trained driver model maintains reliable until the next model retraining phase. However, even for a specific driver, its driving style can vary substantially due to various factors, e.g., daily mood variation, weather condition, and hardware adaption. To account for the time-varying feature of the driving style, the physics-based driver-model parameter identification has been applied to achieve an authentic PDAS. The commonly employed online driver model parameter identification approach is the Least Square (LS) method. However, LS, leaning on the classical Persistence of Excitation (PE) condition, may have slow estimation convergence. Alternatively, this paper proposes a purely algebraic method for driver model parameter identification. This new approach can achieve almost instantaneous multiple parameter identification. Moreover, by periodically resetting the proposed algebraic identifier, time-varying parameters of the driver model can also be captured online. Simulations demonstrate the excellent performance of the proposed algorithm.

This content is only available via PDF.
You do not currently have access to this content.