Global state estimation by an initial state observer: theory and experiments for a two-wheeled mobile robot

When we control a mobile robot, it is very important to measure its state. Estimation methods for a mobile object can be classified into dead reckoning and star reckoning. However, there are some problems with these methods. To conquer such weaknesses, many researchers have studied 'sensor fusion.' Kalman filters are a main technique for sensor fusion. But this technique requires an initial state and error variance matrices. However, it is difficult to obtain error variance matrices, and the initial state is often unknown. Against such a problem, in this paper, we propose a new sensor fusion method for position and posture estimation that does not require error variance matrices and an initial state. We confirm the effectiveness of the proposed method by experiments.