A collection of robotics problems for benchmarking evolutionary computation methods

článek ve sborníku ve WoS nebo Scopus

en

The utilization of benchmarking techniques has a crucial role in the development of novel optimization algorithms, and also in performing comparisons between already existing methods. This is especially true in the field of evolutionary computation, where the theoretical performance of the method is difficult to analyze. For these benchmarking purposes, artificial (or synthetic) functions are currently the most widely used ones. In this paper, we present a collection of real-world robotics problems that can be used for benchmarking evolutionary computation methods. The proposed benchmark problems are a combination of inverse kinematics and path planning in robotics that can be parameterized. We conducted an extensive numerical investigation that encompassed solving 200 benchmark problems by seven selected metaheuristic algorithms. The results of this investigation showed that the proposed benchmark problems are quite difficult (multimodal and non-separable) and that they can be successfully used for differentiating and ranking various metaheuristics.

The utilization of benchmarking techniques has a crucial role in the development of novel optimization algorithms, and also in performing comparisons between already existing methods. This is especially true in the field of evolutionary computation, where the theoretical performance of the method is difficult to analyze. For these benchmarking purposes, artificial (or synthetic) functions are currently the most widely used ones. In this paper, we present a collection of real-world robotics problems that can be used for benchmarking evolutionary computation methods. The proposed benchmark problems are a combination of inverse kinematics and path planning in robotics that can be parameterized. We conducted an extensive numerical investigation that encompassed solving 200 benchmark problems by seven selected metaheuristic algorithms. The results of this investigation showed that the proposed benchmark problems are quite difficult (multimodal and non-separable) and that they can be successfully used for differentiating and ranking various metaheuristics.

Evolutionary computation; Metaheuristics; Benchmarking; Robotics

09.04.2023

Springer

Cham

978-3-031-30229-9

0302-9743

26th International Conference on Applications of Evolutionary Computation, EvoApplications 2023, held as part of EvoStar 2023

13989

364 –379

16