Extended Kalman Filter Enhanced by Neural Network to Solve the SLAM Problem

This work presents a way for online estimation of the location and mapping of a non-holonomic robot by means of an algorithm that uses EKF and in the output of this algorithm, a multilayer perceptron neural network (MLP) has been added that aims to improve the estimation of the robot pose in an unfamiliar environment. The effectiveness was proven through the comparison between the EKF-SLAM and the EKFMLP-SLAM, where it was evidenced a significant improvement in relation to the location of the poses of the robot.