This thesis discusses an approach to automatic vehicle guidance on a motorway. The realization of the system has taken place on the Daimler-Benz research vehicle VITA II (Vision Technology Application). The approach presented here is motivated by psychological models of human driving behaviour and by the artificial potential field approach which is known from robotics. The close connection between driving behaviour and the subjective risk of a traffic situation is used to introduce a decision criterion for driving manoeuvres. Each element of the traffic scene is associated with a risk value. This leads to a risk based representation of traffic situations which is used to generate actions that intend to minimise the risk. Two approaches are presented which are based on different risk representations. The first approach builds a risk map for representation of the environment. By computing the risk value and the reaction the vehicle guidance task is reduced to a search problem. The second approach integrates the gained experience and separates risk and reaction. Knowledge about the driver’s behaviour is represented by competing behavioural patterns. Behavioural patterns correspond to action schemes. They represent driving behaviour of different cognitive levels together with a risk value that describes the necessity of their execution. A set of behavioural patterns is arranged in an inhibition network structure. This network is used to combine their results with regard to their associated risk. The objective is the computation of a command vector for the autonomous vehicle. In real world application the main problem is the noisy input data from different image processing modules. This renders autonomous driving very difficult. The main task to solve this problem is to build up a consistent situation representation in real-time. One part of this thesis deals with this problem and presents a sensor fusion method designed for VITA II. As a supporting work a simulation system for system development and testing has been built up. It contains models of the used sensors and deterministic models of vehicle dynamics and provides an appropriate test bed for the subsequent use of the developed programs in real traffic. (Author/publisher)
Samenvatting