Towards Driver Adaptive Active Safety Systems
In recent years, active safety systems have become a standard accessory in new cars and trucks. This thesis addresses one of the major flaws of most such systems: they act the same independently of who is operating the vehicle. Driver modeling has been in focus for research the past 60-70 years, but it is only recently that adapting driver assistance systems to the current driver has been one of its main application areas. The first part of this thesis gives an overview of the historical development of driver models and leads up to one of the most popular modern modeling frameworks, hybrid driver models. In particular, the framework of hybrid ARX models, including probabilistic ARX models, is described. It is shown how these models can be used to predicvt a driver's steering behavior and to classify the dominating driving style based on vehicle sensor measurements. An algorithm for online steering angle prediction is designed and validated in Paper 1. The driver behavior is also classified as a normal or aggressive. The concept of driver behavior prediction and classification is analyzed in depth for vehicles driving behind another vehicle in Paper 2. It is concluded that for long prediction horizons, there is no gain using a complex model structure. Instead, one-step ahead prediction can be used for classification of the driver's braking behavior. The thesis is summed up with a discussion on how to integrate driver classification models into existing active safety systems.
EC, vån 5, EDIT-huset, Chalmers
Opponent: Dr. David J. Cole, Department of Engineering, Cambridge University, Cambridge, UK