SDF-Loc: Signed Distance Field based 2D Relocalization and Map Update in Dynamic Environments

Abstract

To empower an autonomous robot to perform long-term navigation in a given area, a concurrent localization and map update algorithm is required. In this paper, we tackle this problem by providing both theoretical analysis and algorithm design for robotic systems equipped with 2D laser range finders. The first key contribution of this paper is that we propose a hybrid signed distance field (SDF) framework for laser based localization. The proposed hybrid SDF integrates two methods with complementary characteristics, namely Euclidean SDF (ESDF) and Truncated SDF (TSDF). With our framework, accurate pose estimation and fast map update can be performed simultaneously. Moreover, we introduce a novel sliding window estimator which attains better accuracy by consistently utilizing sensor and map information with both scan-to-scan and scan-to-map data association. Real-world experimental results demonstrate that the proposed algorithm can be used for commercial robots in various environments with long-term usage. Experiments also show that our approach outperforms competing approaches by a wide margin.