Session A1: Alternatives, Backups, Complements to GNSS

Enhanced Navigation for Autonomous Vehicles in GNSS-Denied Urban Environments: Integrating GIS with Deep LSTM/Robust Kalman Filter-Based Landmark Navigation System
Sagar Dasgupta, Mizanur Rahman, Muhammad Sami Irfan, and Minhaj Uddin Ahmad, The University of Alabama
Location: Beacon A

Autonomous Vehicles (AVs) significantly enhance urban mobility by improving safety and reducing environmental impacts, but they heavily rely on the Global Navigation Satellite System (GNSS), which is vulnerable to signal interference such as jamming and multipath effects, particularly in dense urban areas. This study introduces a robust navigation framework for AVs operating in GNSS-denied urban environments, integrating Geographic Information Systems (GIS) with a landmark-based navigation system that employs advanced Deep Long Short-Term Memory (LSTM) networks and a Robust Kalman Filter in parallel to estimate the distance the AV has travelled between two timestamps. This approach enhances data processing from Inertial Measurement Units (IMUs) and speedometers. The Deep LSTM networks effectively analyze sequential sensor data for accurate prediction of vehicle dynamics, while the Robust Kalman Filter manages non-linearities and mitigates noise and outliers in real-time navigation data. The system also utilizes sophisticated computer vision techniques to detect dynamic landmarks such as traffic intersections, which serve to continuously refine and correct the vehicle's trajectory. This landmark-based approach not only improves the reliability and accuracy of navigation in urban settings where GNSS signals are compromised but also demonstrates significant enhancements in operational reliability and navigation precision. Our experimental results have shown that this framework can guide an AV for extended periods without substantial compromise in localization accuracy, effectively maintaining high localization accuracy throughout trips in GNSS-contested urban environments. During our evaluation, we compared the performance of the Deep LSTM networks and the Robust Kalman Filter and found that the Kalman Filter performs better in terms of accuracy and reliability in real-time navigation data processing. This study highlights the practical application of integrating multiple advanced technologies to solve critical navigation issues in autonomous driving, setting a new standard for robust navigation systems in environments with unreliable GNSS signals. This framework promises to significantly enhance the operational reliability and precision of AVs in urban landscapes.