Distributed Extended Kalman Particle Filter for Acoustic SLAM

Abstract

In complex environments, the background noise and room reverberation seriously degrade the performance of acoustic SLAM systems. To remedy the problem, an acoustic SLAM method based on distributed extended Kalman particle filter (DEKF-PF) is proposed. First, the fixed-position microphones are considered as nodes in a distributed microphone array. The direction of arrival (DOA) between the robot and the sound source, and the time difference of arrival (TDOA) between the robots are simultaneously used as observations for the DEKF-PF. Then, the particle filter (PF) integrated with extended Kalman filter (EKF) is utilized to perform local and joint state updates of the SLAM system. The distributed state update mechanism enables the sharing and fusion of state information among nodes, which can effectively improve the overall estimation accuracy of the system and suppress the accumulated error. Next, the average consensus algorithm efficiently fuses the local and joint state information to obtain the sound source positions and robot trajectories. Finally, the acoustic SLAM with the DEKF-PF is performed based on distributed microphone arrays. The proposed method combines the nonlinear modeling capability of the particle filter and the linear estimation advantage of the extended Kalman filter, improving the localization accuracy of acoustic SLAM systems. Simulation and real-world experimental results verify the effectiveness of the proposed method.