One of the most demanding skills for a mobile robot is to be intelligent enough to know its own location. The global localization problem consists of obtaining the robot's pose (position and orientation) in a known map if the initial location is unknown. This task is addressed applying evolutionary computation concepts (Differential Evolution). In the current approach, the distances obtained from the laser sensors are combined with the predicted scan (in the known map) from possible locations to implement a cost function that is optimized by an evolutionary filter. The laser beams (sensor information) are modeled using a combination of probability distributions to implement a non-symmetric fitness function. The main contribution of this paper is to apply the probabilistic approach to design three different cost functions based on known divergences (Jensen-Shannon, Itakura-Saito, and density power). The three metrics have been tested in different experiments and the localization module performance is exceptional in regions with occlusions caused by different obstacles. This fact validates that the non-symmetric probabilistic approach is a suitable technique to be applied to multiple metrics.
Jensen-Shannon divergence; Itakura-Saito divergence; density power divergence; Differential Evolution; global localization; mobile robots