This thesis addresses the problem of providing a methodical framework for highly automated driving on highways, which allows an automated vehicle to execute a safe fallback behavior in case of a system failure. The thesis is however not focusing on the implementation of the whole system. Instead, it focuses on the missing functional components which are needed for the implementation of such kind of fallback system.
Within the thesis a system architecture is presented, in which the methods developed in this thesis can be applied. The methods include algorithms for maneuver recognition, position prediction and trajectory planning.
The main part of the thesis deals with the implementation of the functions which were identified in the system architecture and are needed for planning fallback trajectories. To do so, in a first step classification methods are developed for the recognition of driving maneuvers. The classification problem in the context is limited to the recognition of lane changing and lane following behavior. Extending the state of the art the situation context and possible influencing factors are analyzed systematically.
To predict future positions of other traffic participants, a method is introduced which on the one hand considers the physical state of a vehicle and on the other hand also takes account of the situation context. The presented algorithm does not only deliver information about the future position, but a time dependent probability distribution of future whereabouts.
In order to plan a fallback behavior, a novel method for trajectory planning is introduced. This method introduces a representation of the dynamic free space, which allows to sample a family of trajectories efficiently.
Besides the theoretical investigations presented in this thesis, all methods have proven to be applicable in prototypical vehicles in real traffic.