Considering the major participants in the automotive industry, it is hard to find one with no development connected to autonomous driving. This study provides insight to my work I did in collaboration with Knorr-Bremse Fékrendszerek Kft. Given the profile of the company, our topic focused on heavy duty commercial vehicles, although the discussed algorithms are not truck-specific.
As an input a predefined abstract structure of data is given, describing the environment and the situation that the ego vehicle is in. Using these information I tried to find feasible solutions to some problems which could improve high level decision making in highway scenarios. The theme of my thesis, the problem of behaviour planning has to be handled in real-time, therefore the runtime of the algorithm, the optimization of the solution is a critical point of the specification.
Behaviour planning for autonomous vehicles is a complicated task. Behaviour planning is a more abstract problem than simple vehicle control. Our aim is not merely to stay in a traffic lane at a certain speed, we are thinking at a higher level: it is aimed to change lanes or perform overtakings in order to make an optimal decision in some respects (for example travelling time, consumption, navigation reasons, etc.).
Consciously, we do not concentrate on the the two main controlling task of driving: keeping the vehicle in the middle of the lane and maintaining a steady speed by adjusting the position of the accelerator pedal. Instead, we observe the surrounding vehicles and other objects, we are aware of our own goals and decisions are made to achieve
them. In a given situation our actions are determined by our driving knowledge and our previous experience. In my thesis, therefore, I decided to use artificial intelligence, and I designed the behaviour planner algorithm using a statistical method, probabilistic networks and decisions. After the implementation phase I verified the completed module in SiL simulation.