Model based Kalman Filter Mobile Robot Self-Localization

Abstract

Monotonous position error growth is inherent characteristic of every mobile robot navigational system based solely on proprioceptive sensors. In order to deal with various sources of uncertainties in mobile robot localization it is necessary to establish a representative model of its internal states and environment and use perceptive sensors in the pose estimation. In this chapter we have demonstrated those properties on a differential drive mobile robot by localizing it in a 2D environment by using sonar ring as the perceptive sensor and in a 3D environment by using a mono camera as the perceptive sensor. In both cases we have applied nonlinear Kalman filtering for pose estimation and have compared results with the Monte Carlo localization based on a laser range finder, which is much more accurate sensor than sonars and cameras. Achieved localization accuracies with sonar ring and with mono camera are comparable to those obtained by the laser range finder and Monte Carlo localization. The applied calibration of mobile robot kinematic model also contributed to the increased accuracy.