Fuzzylot II: a novel soft computing approach to the realisation of autonomous driving manoeuvres for intelligent vehicles

Driver error being the primary cause in road accidents provides a strong motivation for developing intelligent systems that can improve vehicle safety and efficiency. This paper describes the use of self-organising neuro-fuzzy rule-based systems to realise intelligent vehicles capable of autonomously performing specific manoeuvres such as parallel and reverse parking, three-point turns, etc. The approach consists of automatically capturing human driving expertise by objectively extracting from training data recorded in simulation an appropriate set of IF-THEN rules mapping sensory input to control output. The method alleviates the need for either complex modelling or knowledge engineering, and produces a fuzzy rule base that constitutes a viable, highly intuitive and easily comprehended linguistic model of the driving process. The driving simulator and neuro-fuzzy systems employed, as well as examples of successful manoeuvres, are presented and discussed.