Simultaneous localization and mapping based on particle filter for sparse environment

This paper presents a method for solving simulation localization and mapping (SLAM) in sparse-feature environment, by adopting a concept of particle filter with multiple extended Kalman filters (EKF). Compared with common FastSLAM where each particle is a sample of one vehicle path whereas each EKF is solely a feature estimator, the proposed algorithm includes the vehicle-pose estimate in each EKF whereas the particle is a sample of vehicle motion. Thus, the proposed algorithm ensures dead reckoning in the absence of features. Map construction is based on line features which are extracted from observation of the environment. Finally, simulation results demonstrate the feasibility and performance of the proposed SLAM algorithm.