Although driver steering control models and on-line model identification have been studied extensively, the application of the driver models to driver state assessment is seldom investigated. Furthermore, the validity level, or confidence index, of the on-line modeling and assessment of driver behavior is not reported in the literature. In this paper, on-line system identification techniques are applied to the determination of driver model parameters and model validity estimation. The driver steering control model is estimated on-line using system identification techniques. The on-line driver state assessment is achieved using probabilistic neural network (PNN), and the validity of the assessment is derived from the likelihood function inside the PNN. Preliminary results show that the computed validity indices agree with expectation reasonably well. More driving simulator experiments will be conducted to validate the proposed indices.
