F2008-02-026
Methodology to Improve ADAS Specification Using Normal Driving Data
Over the past 15 years major technological changes have taken place in the field of automobile driving. Advanced driver assistance systems (ADAS) equip more and more cars. The design of useful and safe ADAS requires real driving behavior data in particular for their specification and their tune-up. These systems, such as Adaptive Cruise Control (ACC), use for their functioning behavioral data (actions of the driver), vehicle dynamic data (speed, acceleration...) as well as information about close traffic in longitudinal regulation situations. To better improve the specification of these systems, it is important to define drivers´ profiles in order to make ADAS functioning suitable with driving task requirements. An experiment on real road was carried out with 126 common drivers on an instrumented car. To ensure that representative road situations are taken into account, data was recorded in ecological conditions, with common drivers using a non-ACC equipped car on a 250 km real road. Four data types were recorded: drivers´ actions, their comments, car dynamic and road environment characteristics. Four main situations of driving (car following, overtaking, cut-in of another vehicle and acceleration during insertion on highways). Some other particular situations were also recorded in order to allow describing the driving in a microscopic way. Two levels of indicators have been calculated: macroscopic ones (mean highways speed...) and microscopic ones (headways before overtaking...). We focus in this paper on the methodology elaborated to extract relevant driving indicators, useful to determine drivers´ profiles. Several methods of multidimensional exploratory data analysis like Principal Components Analysis (PCA), Hierarchical Classification and Multiple Correspondence Analysis (MCA) can be considered. This experimental method has the advantage to allow understanding both the driver´s real need (and not what the technology enables) and his/her real dynamic use of the car. As for any experimental procedure, it is essential to be aware of some biases, which could influence the study conclusions. The data collected from this study and from other ones should enable determining rules for the specification of Adaptive Cruise Control adaptive to drivers´ profiles.
Session: MMI and Driver Performance