Genetic multiobjective fitness assignment scheme applied to robot path planning

This paper proposes a new adaptive Pareto-ranking for multiobjective genetic algorithms. The ranks are assigned after splitting the population in several groups, based on the current weak nadir point and the average objective values. This grouping supplements the sorting provided by the dominance analysis and gives the possibility to encourage certain valuable solutions recommended by the particular landscape of the objective space. Additionally, the preliminary grouping allows a more effective diversity control during the evolutionary loop. The effectiveness of the suggested fitness assignment scheme is shown on a robot path planning problem. The study cases consider continuous working scenes with known non-convex and/or disjoint obstacles.