An EKF-Based Fusion of Visual-Inertial Odometry and GPS for Global Robot Pose Estimation

Globally accurate and drift-free pose estimation is essential for long-term navigation of autonomous robots. In this paper, we present an efficient method to fuse visual-inertial odometry and GPS measurements using the Extended Kalman Filter (EKF). The filter state is propagated by relative visual-inertial estimates and updated by absolute GPS readings to eliminate the accumulated drift on position and heading. We built the sensor hardware and implemented the algorithm on an embedded computer for real-time computation. The experimental datasets were collected in an outdoor environment with 6-degree-of-freedom ground truth. Evaluation results showed that the proposed algorithm achieved consistent and accurate pose estimation with different motion trajectories.