Efficient EKF-SLAM Algorithm Based on Measurement Clustering and Real Data Simulations

In this paper, we propose an efficient extended Kalman filter based simultaneous localization and mapping (EKF-SLAM) algorithm based on measurement clustering. When we apply conventional EKF-SLAM to the real data, because there are lots of landmarks, the amount of computation increases quadratically as the size of the state vector grows, resulting in the inability to guarantee real-time performance. To reduce the computation amount, we propose the measurement clustering technique before augmenting the landmark position variable to the state vector to prevent indiscriminate increase of length of the state vector. The real data was obtained by driving a vehicle with five radars on the same route twice. Our result shows the estimated path via EKF-SLAM is more accurate than estimated path using odometer only.