Abstract. The visual localization problem in robotics poses a dynamically changing environment due to the movement of the robot compared to a static image set serving as environmental map. We develop a particle swarm method adapted to this task and apply elements from dynamic optimization research. We show that our algorithm is able to outperform a Particle Filter, which is a standard localization approach in robotics, in a scenario of two visual outdoor datasets, being computationally more effective and delivering a better localization result.