2D Pose Optimization with Genetic Algorithms and Particle Swarm Optimization: A Comparative Analysis for Robotic Localization

Resumo

Accurate localization of mobile robots—particularly under conditions of initial uncertainty, known as the Global Localization Problem (GLP)—remains a fundamental challenge in robotics, as it directly impacts autonomous navigation, precise mapping, and safe interaction with the environment. This paper presents a comparative study of two optimization metaheuristics—Genetic Algorithms (GA) and Particle Swarm Optimization (PSO)—applied to the scan-to-map matching problem for 2D localization with LiDAR sensing. A controlled simulation framework was developed, within which 150 independent experiments were conducted on a high-resolution synthetic occupancy map, employing a cost function based on a Euclidean distance field with penalties assigned to invalid poses. The methodology encompassed automated data generation, parameterized execution of the algorithms, and comprehensive statistical evaluation of key performance metrics, including position error, orientation error, computational time, and convergence behavior. The experimental results demonstrate that both algorithms achieve high accuracy in most trials, yet with notable differences. PSO exhibited faster convergence and attained a lower median angular error (0.18° versus 0.45° for GA), though it displayed a greater tendency to converge to local minima, occasionally resulting in substantial localization errors. In contrast, GA, while slower and with a higher median angular error, proved more robust in avoiding large-magnitude failures, thereby yielding more consistent solutions across the simulated scenarios. These findings suggest that the choice between GA and PSO should be dictated by application-specific requirements: PSO is preferable in domains where rapid convergence is essential, whereas GA is advantageous in safety-critical contexts demanding reliability and fault tolerance. Future directions include the design of hybrid approaches that combine the efficiency of PSO with the robustness of GA, as well as validation in real-world robotic systems, extension to 3D localization tasks, and integration with multi-sensor data.