@misc {pub2985,
	title = {Foresighted Driving Assistance Systems},
	author = {Thomas H Weisswange AND Christian Goerick},
	year = {2015},
	month = {September},
	abstract = {Experienced humans drive foresighted. They choose their behavior based on a prediction of most probable behaviors of other traffic participants and incorporate small adaptations the help avoid potential risks. The environment is up to a certain degree compliant, i.e. also reacts to the behavior of the ego vehicle. That means, pure prediction is not sufficient for true foresighted driving because a large class of hazardous events has a low probability of occurrence. Additionally, trying to avoid any potential hazard would usually favor a static behavior, i.e. not moving at all. Instead, it is necessary to combine the detection of potential risks with a measure describing the trade-off between behavior changes and risk reduction.
When thinking about developing advanced driver assistance systems (ADAS) which support a driver in foresighted driving, we also have to consider how to communicate results to a driver. This is particularly challenging since the potential events do often not occur and might reduce the acceptance and comprehensibility.  

This work will describe approaches for all three of the above mentioned challenges (potential risk detection, trading off behavior changes, human machine interface), show some application results and also discuss future challenges on the way to a potential use of the general idea of foresighted driving in future ADAS and autonomous driving products.
One risk that is not directly observable in a driving environment originates from static traffic participants, which could start moving at any time and could interfere with the future trajectory of the ego car. In our framework we use the orientation and relative position with respect to e.g. the ego car and the road in general to determine the most relevant entities. To evaluate potential foresighted driving maneuvers we use a spatial grid to represent information about the current environment, including the space occupied by obstacles but also the spatial road and lane layout as well as the space that could be affected by a potential movement of the relevant entities. To also include the most likely movement of other dynamic traffic participants without increasing the complexity of the representation, those future locations will only be considered with respect to the coarse future position of the ego car. With this information potential future behaviors are iteratively evaluated, where it is e.g. favorable to stay inside the lane or pass as little {\textquotedblleft}potential area{\textquotedblright} as possible. Given a result favoring a certain lateral change of behavior, this is finally communicated to a driver via a subtle suggestive car yaw angle change in the respective direction. These {\textquotedblleft}Car Gestures{\textquotedblright} were also investigated in a accompanying simulator user study, to assess their effectiveness, safety and acceptance for regular car drivers in realistic situations.

The presented work shows the first attempt to address the support of foresighted driving via a driver assistance system including possible solutions for dealing with uncertain or potential events. Many of these ideas should also be relevant for the advance of autonomous driving applications.
},
	publisher = {IEEE},
	booktitle = {18th IEEE Intelligent Transportation Systems Conference 2015, Workshop on Interaction of Automated Vehicles with other Traffic Participants},
	city = {Las Palmas de Gran Canaria}
}