A Robust Mobile Robot Navigation System using Neuro-Fuzzy Kalman Filtering and Optimal Fusion of Behavior-based Fuzzy Controllers

Abstract

This study proposes a control system model for mobile robots navigating in unknown environments. The proposed model includes a neuro-fuzzy Extended Kalman Filter for localization task and a behaviorbased fuzzy multi-controller navigation module. The neuro-fuzzy EKF, used for estimating the robot’s position from sensor readings, is an enhanced EKF whose noise covariance matrix is progressively adjusted by a fuzzy neural network. The navigation module features a series of independently-executed fuzzy controllers, each deals with a specific navigation sub-task, or behavior, and a multi-objective optimizer to coordinate all behaviors. The membership functions of all fuzzy controllers play the roles of objective functions for the optimizer, which produces an overall Pareto-optimal control signal to drive the robot. A number of simulations and real-world experiments were conducted to evaluate the performance of this model.