A comprehensive review of AI-based UAV navigation techniques in GPS-denied environments
TL;DR
It is found that sensor fusion and learning-based perception significantly improve navigation in GPS-denied environments, however, challenges such as high computational requirements, limited onboard processing capabilities and data inefficiency in reinforcement learning remain key barriers.
Abstract
This paper aims to provide a comprehensive review of artificial intelligence (AI)-based navigation techniques for unmanned aerial vehicles (UAVs) operating in GPS-denied environments. It highlights the limitations of traditional satellite-based navigation in indoor, urban and dense environments, and explores how AI-driven approaches can enhance autonomous navigation reliability and accuracy. The study reviews literature published between 2020 and 2025, focusing on AI-based navigation frameworks. It analyzes methods such as deep learning, reinforcement learning, visual/inertial SLAM, Kalman filtering and multi-sensor fusion. A comparative evaluation is conducted based on algorithm design, sensor configurations, computational requirements and validation methodologies used in existing research. The review identifies a growing trend toward hybrid navigation architectures combining traditional estimation techniques with AI-based models. It finds that sensor fusion and learning-based perception significantly improve navigation in GPS-denied environments. However, challenges such as high computational requirements, limited onboard processing capabilities and data inefficiency in reinforcement learning remain key barriers. This paper provides an up-to-date and structured review of AI-based UAV navigation specifically focused on GPS-denied environments. It offers a detailed comparison of existing approaches and highlights emerging trends such as hybrid architectures and Edge AI integration, providing valuable insights and future research directions for researchers and practitioners.