UC San Diego

This thesis investigates new techniques to provide a central requirement for many robotics and extended reality systems -- accurate location in many GPS-denied and indoor environments. Accurate location and the consistent tracking of a device, a person, or an important asset over time in indoor environments is a key enabler for the safer operation of robots, improved user experiences in virtual reality systems, or more efficient logistics in warehouses and factories. However, current systems lack the robustness and accuracy to meet the requirements for these applications. My research seeks to combat the challenges facing current localization systems by leveraging wireless radio signals to sense an object's location in space.

However, a single one-size-fits-all localization solution will not suffice. On the one hand, some applications often demand few-centimeter-accurate localization and millimeter-accurate tracking at a room-scale. Imagine setting up a multi-person virtual reality system in a classroom or delivering realistic spatial audio to users in their homes. On the other hand, some applications necessitate consistent localization at a building-scale and have localization and tracking accuracy requirements relaxed by an order of magnitude. Imagine an indoor security robot or mapping drone that must traverse the entire building. Consequently, this thesis will tackle building independent systems to address these broad localization needs for smarter indoor environments of tomorrow.

First, I present XRLoc, which developed a compact localization module that can be easily deployed to provide fine-grained accuracy in room-scale environments to enable AR/VR applications. Through a novel fusion of time and phase measurements of ultrawideband (UWB) signals within a particle filter, the XRLoc localization module provides localization to approximately 2 cm of median accuracy in dynamic scenarios for UWB tags and mobile devices. Next, I present P²SLAM and WAIS, which leverage WiFi devices in an environment to accurately and efficiently correct for drifts in robot trajectory predictions and provide reliable building-scale localization. Through 1500 m of cumulative travel, I found an end-to-end 4.3x reduction in compute and memory compared to traditional visual-based localization systems. Incorporating WiFi measurements also improved the 90th percentile translation errors by approximately 40% and orientation errors by approximately 60% compared with these visual systems. It further provided a 6x improved trajectory estimate compared to purely relying on wheel odometry. Finally, I present DOLOS, a system to protect enterprise WiFi users from the privacy concerns posed by current WiFi sub-meter accurate localization systems. Through this work, I develop a novel system to intelligently mask the user's location without impacting their connectivity to the system and degrade the location accuracy of state-of-the-art localization systems by 2.5x to 3.8 m.