Estimation of Maneuver Risk with Neural Network

OData support
Supervisor:
Csorvási Gábor
Department of Automation and Applied Informatics

The development of solutions improving the road safety, including increasingly diverse automated functions, are treated in a priority position by all the remarkable participants in the automotive industry. In the light of this, it is hard to find any related major company without development and research connected to highly autonomous driving. Due to the complexity of the task, engineers and research scientists face several serious challenges in this area. Some of these are identifying safety threats, quantifying risk related to possible decisions and making high level decisions accordingly. Problems like these don't have and might never have a widely adequate solution, as severe moral and legal questions surround them. In my master thesis I deal with a specific part of these problems. First I survey risk calculation methods used in the field of autonomous driving, then I suggest and create my own solution that is able to predict the detailed risk of different achievable manoeuvres in real time conditions. My research and work was done within the Highway Pilot (HP) project at the Knorr-Bremse R&D Institute Budapest. This project aims to pre-develop a Level3 autonomous system for heavy duty vehicles in highway environment. As a result of the project's earlier stages, a deliberate framework, including an abstract numerical representation of the environment and a specific discretized manoeuvre space for possible decisions, was available for my task. Emphasis in this thesis is placed on the question, weather Neural networks are applicable for manoeuvre risk prediction in such a situation.

Downloads

Please sign in to download the files of this thesis.