Robust localization and map updating based on Euclidean signed distance field map in dynamic environments

Classical Simultaneous Localization and Mapping (SLAM) approaches are designed to map the static environments. In order to reuse the created map in the real world, this paper presents an approach based on Euclidean signed distance field (ESDF) map to cope with dynamic environments. To tackle the robust localization problem, the laser and odometry sensors are tightly coupled to estimate the mobile robot poses, which achieves superior performance over adaptive Mento Carlo Localization (AMCL) in terms of localization accuracy and robustness in both static and dynamic environments. As for map updating, an obstacle classification algorithm based on similarity is proposed to classify the new merged low-dynamic and high-dynamic obstacles. Then, the proposed algorithm based on wavefront propagation is used to update the prior ESDF map. Numerous experiments are carried out in realistic scenarios to validate the proposed approach.