On-board Deep-learning-based Unmanned Aerial Vehicle Fault Cause Detection and Identification

With the increase in use of Unmanned Aerial Vehicles (UAVs)/drones, it is\nimportant to detect and identify causes of failure in real time for proper\nrecovery from a potential crash-like scenario or post incident forensics\nanalysis. The cause of crash could be either a fault in the sensor/actuator\nsystem, a physical damage/attack, or a cyber attack on the drone's software. In\nthis paper, we propose novel architectures based on deep Convolutional and Long\nShort-Term Memory Neural Networks (CNNs and LSTMs) to detect (via Autoencoder)\nand classify drone mis-operations based on sensor data. The proposed\narchitectures are able to learn high-level features automatically from the raw\nsensor data and learn the spatial and temporal dynamics in the sensor data. We\nvalidate the proposed deep-learning architectures via simulations and\nexperiments on a real drone. Empirical results show that our solution is able\nto detect with over 90% accuracy and classify various types of drone\nmis-operations (with about 99% accuracy (simulation data) and upto 88% accuracy\n(experimental data)).\n