A Novel Acoustic SLAM Method Based on Single Cluster Probability Hypothesis Density Filter

Robustness to uncertain source numbers and data association is a key challenge for acoustic simultaneous localization and mapping (ASLAM). To address this problem, a novel acoustic probability hypothesis density (PHD)-SLAM method with an improved single cluster PHD filter is proposed in this article. Specifically, the robot positions and direction of arrival (DoA) observations are modeled as random finite sets (RFSs), and their first-order moments are recursively propagated. Then, the PHD prediction is executed through the particle swarm optimization (PSO) algorithm, wherein a fitness function is constructed to refine the PHD using the latest observations. Next, the bearing-only DoA information with range hypotheses is calculated by the unscented Kalman filter (UKF). Finally, the number and location of sound sources as well as the robot's trajectory are jointly estimated based on the improved single cluster PHD and Rao-Blackwellized filter. The proposed ASLAM demonstrates commendable localization accuracy even under speech inactivity and clutter measurements conditions. Experimental results reveal the validity of the proposed method.