Genetic optimization of a vehicle fuzzy decision system for intersections

This paper presents a case study in which an autonomous vehicle must cooperate with a supposedly manually driven one to carry out a cross-roads manœuvre without risk. The main difference with other intersection systems is that the manual vehicle is driven without paying attention to the controlled one, so a cooperative coordination between vehicles is not possible. In this case is the autonomous vehicle the responsible of adapting its speed to the state of the manually driven, for finalizing the manœuvre both in a safe and efficient way. For this purpose, a three layer hierarchical fuzzy rule-based system (FRBS) is developed with the aim of dealing with such a situation: the first layer is in charge of detecting the kind of manœuvre that will be necessary; the second, in the case that an intersection is going to be crossed, is in charge of determining the suitable speed to do so without risk; and the third acts on the vehicle's real speed. The first two layers are implemented by means of fuzzy decision systems, with the second being optimized by a genetic algorithm (GA). The GA evaluates candidates in random simulated scenarios taking into account different factors to calculate the fitness. These factors are: implementing a free collision policy, avoiding unnecessary stops, and terminating the manœuvre as rapidly as possible.