Personalized driver assistance systems based on driver/vehicle models

2014-10-15Driving, while a necessary and sometimes enjoyable experience, can become stressful and often challenging in city environments due to congestion, difficult environmental conditions, and the mistakes of other drivers. The latest trend of improving the urban driving experience is equipping cars with Advanced Driver Assistance Systems (ADAS). These systems aim to make driving safer and more comfortable. They can advise the driver when performing certain maneuvers is safe and even assist the driver in performing these maneuvers. An ADAS can also monitor driver and vehicle response in order to detect and avoid potentially dangerous situations by comparing current values of parameters with expected ones. In addition, an ADAS can assume control of the vehicle. ❧ The assistance system can be effective only if it captures personal driving style as well as the dynamics of the driver/vehicle system. However, the conventional way of manually adjusting car systems to tailor them to a driver’s personal desires is hardly appropriate for an ADAS. This type of performance requires a large number of parameters that would need to be manually adjusted by the driver over many iterations of trial and error before driving begins. Most ADAS are designed having the average driver’s characteristics in mind. This may lead to a system that aggressive drivers find annoying and passive drivers find too aggressive. Consequently, by adding stress and disappointment, this could dissuade drivers from using the very system that is supposed to make driving easier and more joyful. ❧ In this work we present modeling techniques that allow extracting knowledge of driver/vehicle behavior for the most stressful and accident-provoking driving tasks: vehicle following, lane change and intersection crossing. The models capture the personal characteristics associated to a particular driver and vehicle. The models, along with the corresponding parameter learning techniques, recommendation personalization algorithms and diagnostics methodology can be used as a kernel for a system that can provide personalized driving assistance and behavior monitoring. In addition, the presence of skewed model parameters can be used to evaluate a driver’s level of aggression. ❧ We demonstrate the effectiveness of the proposed techniques by applying the methods to experimental data collected on a customized vehicle. The results demonstrate that the models are able to capture personal habits (such as following distance and lane change duration) and distinguish between different drivers. In addition, we show that the developed intersection assistance algorithm that takes into account driver’s habits and preferences helps to reduce fuel consumption