Improving EKF-based solutions for SLAM problems in Mobile Robots employing Neuro-Fuzzy Supervision

The determination of the solution for simultaneous localization and mapping (SLAM) problems has gained significant research momentum in recent times. Although extended Kalman filters have been extensively employed to solve these problems in mobile robots, the performance of the EKF can degrade significantly, if the correct a priori knowledge of process and sensor/measurement noise covariance matrices (Q and R respectively) is not available. Hence, the present paper proposes the development of a new robust solution method for SLAM problems where we employ a neuro-fuzzy system to supervise the performance of the EKF for SLAM problems and take necessary corrective actions by adapting the sensor statistics online, so that the degradation in system performance can be arrested. The free parameters of the neuro-fuzzy system are learned offline, by employing particle swarm optimization in the training phase. The system hence proposed is successfully evaluated by employing it to localize a mobile robot and simultaneously acquire the map of the environment, under benchmark environment situations with varying landmarks and with wrong knowledge of sensor statistics