Driver assistance and traffic systems have an enormous potential to improve road safety by targeted support of the human user (in this case the driver) in overly demanding or hazardous situations. To realize this potential and to achieve driver compliance, the systems need to act in accordance with the driver’s own driving intentions and not at cross-purposes. This requirement is particularly important in urban traffic, due to its great complexity, diversity of situations and options for driver response; the limiting factor is usually the time budget for decisions, which is generally smaller than in freeway traffic and must be used as effectively as possible. If the intervention strategy of a vehicle system clashes with the intentions and actions of the driver, the resulting ambiguous situation could lead to delays, missing the window of opportunity for mitigating a traffic conflict. Hence, inferring the driver’s intentions and predicting his response to a hazardous situation as early as possible are of central importance for coordinated driver assistance; intention inference and behavior prediction should be adapted to the individual driver and the situation.
The central aims of the sub-project are:
After analysis of data requirements, the sub-project will work on detailed intention inference and behavior prediction algorithms for urban areas. The quality of the algorithms will be systematically tested and demonstrated in project test cars. Demonstration of the algorithms represents the central result of the sub-project.
The consortium structure will contribute to cooperation between university and industry partners in development of algorithms utilizing different input signals and in their comparative quality assessment. The practical benefits of the algorithms will emerge through their integration into applications. Direct integration is planned by the industrial partners, particularly in cognitive assistance applications.
In the sub-project “Human-Machine Interaction for Urban Environments”, additional applications that are limited in scope to informing or warning the driver, but that do not intervene, will be demonstrated. In this way, the cooperation in UR:BAN will enable utilization of these algorithms as well.