LAUSR

Learning from human driver’s strategies for undertaking complex traffic scenarios has the potential to improve decision-making methods for designing ADAS systems, as well as for design self-driving rules for automated vehicles. This paper proposes a human-like decision-making algorithm built up from human drivers experiential naturalistic driving. The approach of this work consists of exploring two main techniques. Firstly, the use of “think aloud protocol” to build a dataset based on naturalistic driving, capturing driver’s intentions. Afterwards, the technique of decision tree is used to generate an algorithm to categorize driving patterns as a function of circumstantial driving parameters. The study is focused on simple roundabouts in presence of other vehicles. The decision tree is translated into algorithmic rules, where the tree pathways are represented as ‘if-then’ clauses, resulting in a model of driver behavior at roundabouts. Finally, the accuracy of the driver behavior model has been assessed, yielding promising results.