Neural networks to aid the autonomous landing of a UAV on a ship

Abstract

This paper proposes to examine the possible uses of Artificial Neural Networks (ANN) to aid the landing of an Unmanned Aerial Vehicle (UAV) on a ship. Three distinct phases are proposed. The dataset required for training and testing was produced by simulating a ship's motion at sea using Unity. Phase 1 converts video images from a UAV on-board camera to numeric data. Phase 2 utilizes Phase 1 data and calculates the current relative orientation and distance of the UAV to the landing platform. Co-ordinate pairs representing screen positions of particular areas of a ship's landing pad were normalized and used to train the Phase 2 ANN. Orientation has been calculated to an accuracy of +/-1% and distance +/-2%. Phase 3 determines future landing windows. Phase 3 uses the orientations produced in Phase 2 and calculates future periods when a landing, within a time limit, could be attempted. This paper proposes strategies and current research into Phases 1, 2 and 3 and suggests development of an indicator of optimal landing times for Manned Aerial Vehicles (MAV).