Multi-objective PSO-PS application to humanoid path following optimization

This paper optimizes multi-objective particle swarm optimization with preference based sort by discovering optimal exploration and exploitation variables. In the proposed algorithm, user´s preference is taken into account along with mutual dependencies and priorities of objectives while selecting the global best particle. The user´s preference is represented as degree of consideration for each objective. This is achieved by using the fuzzy measure which integrates the partial evaluation of each objective with respect to the degree of consideration. The randomly chosen global best attracter is among the non-dominated particles selected, and has a relatively higher global evaluation value than the other particles. While updating the particles, the optimal inertia weight, cognitive acceleration and social acceleration constants are selected simultaneously based on user´s preference. Inspired by reinforcement learning, exploration v.s. exploitation strategy is incorporated to find the optimal solution. The experiments done on humanoid robot foot-step following a path demonstrate the effectiveness of the proposed algorithm.