The appearance of self-driving vehicles projects one of the most significant changes in road transport, similar to the one experienced last time in the history of automotive industry with the appearance of mass production.
Due to emerging opportunities in the market development of autonomous vehicles is one of the most popular and at the same time the most challenging field of research in the industry.
My thesis provides insight into the research and development activities of Knorr-Bremse Fékrendszerek Kft. in this field.
Due to the profile of the company the main purpose of development is related to the automation of heavy duty commercial vehicles, however, the discussed research and development solutions so far are not all truck-specific. The goal of the current development is to create an autonomous system for highway operation.
During my work I was provided with abstract environmental data (e.g. certain parameters of traffic lanes, detected objects) from the environment perception module, which do not contain any information on the actual traffic scenario, its density or its hectic nature. However, to make higher level decisions, these are essential information, as drivers behave in a different way in a stop-and-go traffic jam, in dense traffic (which goes ahead with almost full speed) or on an empty highway.
The aim of my thesis is to make the detection and classification of traffic situations possible based on the data provided by the environment perception subsystem in order to supply the behaviour planner component with traffic classification information. Following the establishment of a concept, I compared the performance of different classification algorithms to predict the density of the surrounding traffic, then I worked on further improving the classification results by fine-tuning the hyperparameters of the best performing Support Vector Machines algorithm. For the efficient generation of training data it was also necessary to develop a simulator application supplemented by a graphical viewer.