Towards a learning autonomous driver system

A system is to be implemented, which is able to learn and develop completely on its own the ability to steer different vehicles in different environments. Both the environment (road patterns of middle complexity) as well the vehicles are part of a real-time simulation. The system processes the image video stream coming from a video camera and creates a parametric description of the current scene. By means of processing the records of different repeated simulation runs, the behavioural patterns are developed and optimised. Based on these, the capability and efficiency of this system increases in handling the task of steering the vehicle. Similar traffic situations are also being managed, even if they are unknown, based on the accumulated knowledge and experience. The paper describes the first step towards a system that is able to learn to steer different vehicles on different courses quite optimally