Slow-Drift GPS Spoofing in UAV Swarms Using LSTM-Based Sensor Fusion -- 2025 3rd International Conference on Foundation and Large Language Models (FLLM)

Unmanned Aerial Vehicles (UAVs) increasingly rely on civilian Global Positioning System (GPS) signals for autonomous navigation, making them vulnerable to spoofing attacks. Among these, slow-drift spoofing is especially dangerous: by injecting small, noise-level biases into GPS measurements, an adversary can gradually displace a UAV or swarm without triggering traditional anomaly detectors. This paper presents a simulation framework that models a UAV swarm under slow-drift spoofing, incorporating realistic disturbances such as wind bias, GPS jitter, and IMU imperfections. To detect such attacks, we design a lightweight detection system based on a bidirectional Long Short-Term Memory (LSTM) network, which classifies raw IMU and gyroscope sequences in real time. We evaluate the system’s performance across multiple input window sizes to understand how temporal context influences detection accuracy. While simulation constraints limit full generalization, the findings demonstrate the feasibility of temporal deep learning for spoofing detection and highlight the need for future work in field validation, swarm-level collaboration, and adversarially robust defenses.

Slow-Drift GPS Spoofing in UAV Swarms Using LSTM-Based Sensor Fusion -- 2025 3rd International Conference on Foundation and Large Language Models (FLLM) Academic Article