A multi-sensor fusion framework for autonomous robots localization in repetitive environments

Abstract

It is generally recognized that the widely utilized Global Navigation Satellite System (GNSS) signals may be severely challenged or even unavailable in enclosed or partially enclosed repetitive and ambiguous environments (e.g., offices, hotels, hospitals, airports, industrial warehouses, container seaports). With the surge of autonomous robots applications, a robust localization algorithm that is capable to accurately localize the robots within such challenging environments is a critical pre-requisite to enable functionalities such as mapping, navigation, intelligent services, and collaborative works. Current localization solutions in such environments can be classified into two main categories: infrastructure-based and infrastructure-free. For infrastructure-based solutions (e.g., Wi-Fi based, UWB based, RFID based, etc.), the costs for deploying and maintaining the infrastructures can be high, which is also neither scalable nor flexible. As for infrastructure-free localization, many of the approaches only rely on a single-modal sensor, such as Light Detection and Ranging (LiDAR), camera, or magnetometer. Moreover, methods such as simultaneous localization and mapping (SLAM) have much lower cost and can be easily adapted to almost any environment. Normally, SLAM-based approaches work well in ordinary environments with rich geometric features. However, aforementioned repetitive environments contain lots of similar and ambiguous settings with few distinct geometric features, thus making SLAM-based methods inaccurate or even fail. Over the past few years, the ambient magnetic field (MF) based positioning and localization algorithms have drawn significant attention. The MF has exhibited high distinctiveness and ubiquity at different location, which makes it a viable alternative to the GNSS signal for localization in the local environments. To take the aforementioned challenges and the advantages of MF into account, this thesis proposes a systematic multi-sensor fusion probabilistic localization framework for autonomous robots operating in repetitive environments.