Multi-Objective PSO with Variable Number of Dimensions for Space Robot Path Optimization

Abstract

This paper aims to solve the space robot pathfinding problem, formulated as a multi-objective (MO) optimization problem with a variable number of dimensions (VND). This formulation enables the search and comparison of potential solutions with different model complexities within a single optimization run. A novel VND MO algorithm based on the well-known particle swarm optimization (PSO) algorithm is introduced and thoroughly described in this paper. The novel VNDMOPSO algorithm is validated on a set of 21 benchmark problems with different dimensionality settings and compared with two other state-of-the-art VND MO algorithms. Then, it is applied to solve five different instances of the space robot pathfinding problem formulated as a VND MO problem where two objectives are considered: (1) the minimal distance of the selected path, and (2) the minimal energy cost (expressed as the number of turning points). VNDMOPSO shows at least comparable or better convergence on the benchmark problems and significantly better convergence properties on the VND pathfinding problems compared with other VND MO algorithms.

This paper aims to solve the space robot pathfinding problem, formulated as a multi-objective (MO) optimization problem with a variable number of dimensions (VND). This formulation enables the search and comparison of potential solutions with different model complexities within a single optimization run. A novel VND MO algorithm based on the well-known particle swarm optimization (PSO) algorithm is introduced and thoroughly described in this paper. The novel VNDMOPSO algorithm is validated on a set of 21 benchmark problems with different dimensionality settings and compared with two other state-of-the-art VND MO algorithms. Then, it is applied to solve five different instances of the space robot pathfinding problem formulated as a VND MO problem where two objectives are considered: (1) the minimal distance of the selected path, and (2) the minimal energy cost (expressed as the number of turning points). VNDMOPSO shows at least comparable or better convergence on the benchmark problems and significantly better convergence properties on the VND pathfinding problems compared with other VND MO algorithms.