Abstract
Wheel odometry has been a commonly utilized method of self-contained localization since it is readily available in wheeled platforms. However, this limits against non-wheeled platforms. LiDARs and cameras are suggested as alternative sensors for odometry but suffers from being exteroceptive sensors, computationally intensive and expensive. Furthermore, these limitations can be extended to the case of Simultaneous Localization and Mapping (SLAM). Given the ubiquity of WiFi in this era, in this dissertation, WiFi is proposed for addressing these challenges in sensor implementation for localization and SLAM. First, the methods for computing similarity measures are proposed whereWiFi similarity is utilized as a feature for obtaining odometry and SLAM. Radio fingerprints are used to represent WiFi signals and their nature is analyzed for formulating these methods. Next, theWiFi Similarity-Based Odometry method is proposed for obtaining odometry fromWiFi signals. It involves reducing the noise of the signals for improved reliability and feature extraction for obtaining a model which is used to incrementally estimate position. Next, the WiFi Similarity-Based SLAM method is proposed for performing SLAM using WiFi signals. This method builds upon WiFi Similarity-Based Odometry. A second model is introduced and obtained for loop closure detection used with a pose graph. Finally, the WiFi Similarity-Based collaborative SLAM method is proposed for performing collaborative SLAM using WiFi signals in a multi-robot system. This method builds upon WiFi Similarity-Based SLAM. The same loop closure detection model is used for interloop detection across multiple platforms. The method demonstrates scalability with its models and efficiency in performing SLAM over large areas. Experiments were conducted in different locations with each proposed method which has demonstrated the feasibility and good performance of using WiFi signals primarily for odometry, SLAM and collaborative SLAM.