A multi-objective evolutionary algorithm for the tuning of fuzzy rule bases for uncoordinated intersections in autonomous driving

This paper focuses on the application of Multi-Objective Evolutionary Algorithms (MOEAs) to develop a Fuzzy Rule-Based System (FRBS) dedicated to manage the speed of an autonomous vehicle in an intersection scenario.
Compared to other intersection scenarios, the main point here is that the autonomous vehicle is approaching an intersection that is being crossed by a row of manual vehicles those are not paying any attention to the presence of the autonomous vehicle, thus making coordination impossible. In this case, the autonomous vehicle bears sole responsibility for adapting its speed to the state of the other vehicles, with the aim of completing the maneuver safely and efficiently.
The specific conditions of this problem make it complex because of the large time requirements needed to consider multiple criteria (which enlarge the solution search space) and the long computation time required in each evaluation. In addition, the large number of variables involved increases the complexity of the scenario.
In this paper, a MOEA is proposed to obtain a more compact and efficient FRBS. The proposal is based on the well-known Strength Pareto Evolutionary Algorithm 2 (SPEA2) technique, but uses different mechanisms for guiding the search towards the desired Pareto zone. The MOEA uses specific operators to deal with the problem, to inherit fitness values from one generation to the next, thus arranging that it is only necessary to execute one scenario per generation to obtain an FRBS that works fine in many situations. In addition, the most important rules are identified in each FRBS, with the aim of realizing balanced crossovers.