Sub-Meter Accurate Pedestrian Indoor Navigation System with Dual ZUPT-Aided INS, Machine Learning-Aided LTE, and UWB Signals

TLDR: The PINDOC system achieved a position Root-Mean-Squared Error (RMSE), maximum error, and loop-closure error of 0.93 m, 2.23 m, and 1.28 m over the 600-meter trajectory, respectively.

Abstract: Developing a universal pedestrian navigation framework that operates through extreme environmental conditions is essential. Such a navigation framework can enable Location-Based Services (LBS) in many applications, and one application in high demand of accurate and reliable positioning solutions is firefighter localization, primarily for navigating in indoor environments where signals of Global Navigation Satellite Systems (GNSS) might degrade or fail, visibility is poor, and infrastructure dedicated to navigation is often not accessible. Jao et al. (2022a) reported a Pedestrian Indoor Navigation system integrating Deterministic, Opportunistic, and Cooperative localization approaches (PINDOC). The deterministic localization is a Zero-velocity-UPdaTe (ZUPT)-aided Inertial Navigation System (INS) enhanced with self-contained aiding approaches, including altimeter measurements and foot-to-foot ranging measurements. The opportunistic approach uses pseudorange measurements extracted from cellular Long-Term Evolution (LTE) towers and implements a Deep Neural Network (DNN)-based Synthetic Aperture Navigation (SAN) to spatially mitigate multipath. This approach operates in a base/rover framework, where a GNSS receiver and a "base" LTE receiver, both installed stationary in an outdoor environment, are used to estimate clock bias drifts of LTE towers, and the estimated clock biases are transmitted to "rover" LTE receivers equipped on agents navigating in indoor environments. The cooperative localization approach uses UWBs for inter-agent range measurements and differentiates Line-Of-Sight (LOS) and NLOS components using a power-metric-based detector. In this paper, we experimentally investigate the navigation performance of the PINDOC system. Two experiments were conducted. The first experiment involved three agents, with one agent traversing in an indoor environment a trajectory of 600 meters in 14 minutes, during which the other two agents remained stationary. The traversed trajectory included terrains of flat surfaces, stairs, ramps, and elevators. The PINDOC system achieved a position Root-Mean-Squared Error (RMSE), maximum error, and loop-closure error of 0.93 m, 2.23 m, and 1.28 m over the 600-meter trajectory, respectively. In the second experiment, all three agents traveled in the indoor environment for 12.5 minutes, and the navigation solutions estimated by the PINDOC system showed loop-closure errors of 0.35 m, 0.82, and 1.15 m for the three agents. In all cases, access to signals of opportunity and cooperative exchange of information between agents were available less than 20% of time for duration of the experiments.