Evolutionary constrained self-localization for autonomous agents

In this paper, a new model for robotic self-localization using constrained evolutionary computation techniques is proposed. The approach uses a previous stage of location estimation based on Kalman filters in order to redefine the search space for a Genetic Algorithm that finds the most accurate current position of a robot. Genetic Algorithms (GAs) have the advantage of being non-gradient-based optimization methods that have an important role in non-linear systems with high noise to signal ratio. The set of solutions is modified according to natural evolution mechanisms. In addition, GAs are used as a parallel, global search technique, and it evaluates many localization solutions simultaneously, improving the probability of finding the global optimum. Experiments using the approach show the promise of the method to predict the correct position in a robotic soccer field with a error margin better than other state-of-the-art techniques.