"A Virtual Co-Driver for Improved Driving Safety on Country Roads"
A Virtual Co-Driver for Improved Driving Safety on Country Roads
Sprache des Titels:
Safety is a topic of huge importance in road traffic. A key component in preventing
potentially fatal accidents is a correct perception of the vehicle environment, including
other traffic participants, which subsequently enables a safe interaction. This applies
to human drivers as well as to Advanced Driver Assistance Systems (ADAS) and, also,
to fully Autonomous Vehicles (AVs) that may become available in the future. Different
environments present different challenges as far as safety is concerned. Therefore, the
development of well-tailored safety concepts for specific environments promises a more
significant impact than a one-for-all solution. In Austria, country roads contribute at
the largest extent to the fatality rate in traffic. This is mostly due to high (relative)
velocities, bidirectional traffic and narrow curves which not only restrict the observable
range but also the permissible forces acting on a vehicle.
In this thesis, a Virtual Co-Driver (VCD) which improves traffic safety on country
roads is proposed. It exploits available data of the environment, analyzes the behavior
of other traffic participants, and identifies the role of the vehicle in the scenario as
well as its options to remain in a global safe state. A global safe state is defined as a
state where a collision can be avoided in any case. The system is designed to either
communicate the safety of available maneuver options to a driver or to take over
control if the intended action of the responsible driver, ADAS, or AV tends to result
in an accident. Efficient ways to assure the safety of states and transitions within an
environment are developed, and the system is validated on various traffic situations in
simulation as well as by application on actual measured drives along a country road.
The evaluation of basic maneuver options by the VCD yields sensible outcomes, and
the system intervenes accordingly at critical occasions when applied to real-world
driving data. Hence, an improvement of driving safety on country roads is clearly
given by the VCD developed within this thesis.