Adaptive Machine Learning for Efficient Anomaly Detection in Autonomous UAVs Swarm Operations

Unmanned Aerial Vehicles (UAVs), commonly known as drones are aircraft without a human pilot, crew, or passenger on board. An Unmanned Aerial Vehicle swarm (UAV swarm) usually consists of three or more drones and can execute complex tasks and missions. UAV and UAV Swarm have proliferated these years and have significantly impacted vari-ous fields, including military, agriculture, disaster relief, rescue supplies, and energy management. However, these advantages require enhanced safety measures to prevent group collisions, secure communication within the swarm, and detect abnormal behaviours, including sensor faults and cyber-attacks (e.g., GPS spoofing and command hijacking). This paper proposes an anomaly detection method based on Bi-LSTM with an attention mechanism to identify abnormal behaviours within a UAV swarm. The proposed method uses a supervised approach, training and validating the model with data collected and labelled from self-designed simulated UAV swarm flight experiments. These swarm flight tests include normal flights, flights in windy environments, flights with simulated cyber-attacks and fault injection. These multiple flights enable the model to learn to detect anomalies during formation missions across different conditions. Experimental results demonstrate that the anomaly detection model effectively processes time series data and learns its features from UAV swarm. The evaluation confirms that this method is a sensible solution for detecting anomalies within UAV swarm flights.