Online map building for terrain scanning robots using a hybrid neurofuzzy Kalman filter

This paper presents a proposed online map building for a terrain-scanning robot that is essentially a mobile manipulator built for landmine detection. Map building of the robot is a generic dynamic modeling process to provide a terrain map in real time. It is expected that the proposed map building process is equally applicable to other autonomous mobile manipulators. Map building provides an a priori terrain model for obstacle free path planning in parallel to other tasks of the robot. Map filtering, as the core of map building, updates the model in real time using the range measurements obtained from laser and ultrasonic rangefinders. Filtering takes place by means of a proposed adaptive neurofuzzy Kalman filter. The filter is capable of dealing with uncertainties associated with sensor data and modeling through the learning capability of artificial neural networks in incorporating real life experiments for more efficient terrain typing. Experimental trials validate the effectiveness of the proposed map building process.