Deep Learning-Enhanced Sensor Fusion for Autonomous Unmanned Aerial Vehicle Navigation

Abstract

The proliferation of Unmanned Aerial Vehicles (UAV) in the civil and military sector has demanded the development of more potent yet precise navigation systems capable of operating under the hostile and mobile environments. The stability of the autonomous is obligatory to be the aspect of precision state estimation, path planning and obstacle avoidance. However, each sensor suffers independent failures such as drift of Inertial Measurement Unit (IMU) and signal jamming in a Global Navigation Satellite System (GNSS). In this review paper, the state-of-the-art of the two sensor fusion designs on autonomous UAVs and navigation sensors will be discussed in detail. We are critical of core sensor principles of operation and constraints like IMUs, GNSS, Vision, LiDAR, and Magnetometers. Besides, we review the available literature on the sensor fusion algorithms, both conventional types of Kalman Filters and the more recent types of Simultaneous Localization and Mapping (SLAM) and the newer Deep Learning-related approaches. According to the review, the trend has been towards edge-computing solutions and using multi-modal sensor redundancy to resolve the GNSS-denied environments. We end up by establishing important challenges and recommending the future research path of hybrid AI based fusion and collaborative swarm localization.