Dual multi-objective particle swarm optimization and its application to on-line navigation of humanoid robots

Abstract

This thesis proposes a novel multi-objective evolutionary approach for on-line navigation of humanoid robots. The proposed approach helps a humanoid robot to optimize its footsteps with local information while walking in an unstructured environment. In humanoid robot navigation, it is required to consider the multiple objectives such as elapsed time, safety, and energy consumption at the same time, where two or more of them conflict with each other. Therefore, this kind of problem should be formulated as a multi-objective problem. In other words, multiobjective evolutionary algorithms (MOEAs) are necessary. In addition, the MOEAs should be able to take the priorities or the dependencies of the objectives into account and determine the relative merits of the nondominated solutions. To deal with these issues, dual multi-objective particle swarm optimization (DMOPSO) is proposed. As a preliminary, multi-objective particle swarm optimization with preference-based sort (MOPSO-PS) is developed. And then, by extending the concept of preference-based sort into a secondary multi-objective optimization, DMOPSO is proposed. DMOPSO utilizes the secondary objectives, global evaluation value and crowding distance. The global evaluation value of a particle is calculated by the fuzzy integral that integrates the partial evaluation value of each objective with respect to the degree of consideration where the user’s preference is represented as the degree of consideration for each objective using the fuzzy measure. The crowding distance of a particle implies the crowdedness around the particle. Through the secondary objectives, the user’s preference can be properly reflected and thus the relative merits of the nondominated solutions can be clearly figured out as well as the interactions among the objectives can be considered. The effectiveness of the DMOPSO is demonstrated by the empirical comparison with the other algorithms. The results indicate that the user’s preference is...