Akademik Veri Yönetim Sistemi
Real-World AI Systems, cilt.1, sa.2, ss.45-59, 2025 (Hakemli Dergi)
The research domain of positioning systems has become increasingly critical with the rapid expansion of applications in navigation, asset tracking, robotics, and autonomous systems. Conventional solutions primarily rely on Global Positioning System (GPS), which, despite its widespread use, suffers from high energy consumption and limited reliability in indoor or obstructed environments. To address these limitations, prior studies have explored Inertial Measurement Unit (IMU)-based sensor fusion approaches, including Kalman and particle filters, as well as hybrid schemes with intermittent GPS updates. Although these methods improve robustness, they remain limited in handling sensor drift, nonlinear dynamics, and adaptability to changing conditions. This paper addresses this gap by proposing an AI-augmented sensor fusion framework that integrates a deep learning–enhanced Kalman filter for robust IMU–GPS positioning. The proposed method uses a neural network to model nonlinear motion dynamics and adaptively correct estimation errors, thereby improving positioning performance while reducing reliance on frequent GPS measurements. Experimental evaluations under varying GPS sampling intervals demonstrate that the proposed approach maintains acceptable accuracy and precision while significantly reducing energy consumption and memory usage compared to GPS-only and conventional filtering methods. The main contribution of this work is the introduction of an adaptive, AI-driven sensor fusion architecture for resource-constrained embedded systems. By combining model-based estimation with data-driven correction, the framework offers a scalable and efficient solution for both indoor and outdoor localization. This study advances the state of the art by demonstrating that AI-augmented filtering can effectively balance accuracy, robustness, and resource efficiency in real-world positioning scenarios.