Risk-Aware Guidance of a Fixed-Wing UAV using Neural Network Model Predictive Control

Abstract

This paper presents a guidance algorithm for fixed-wing Unmanned Aerial Vehicles (UAVs) that accounts for risk wrt people at ground in case of failure of the vehicle. Model Predictive Control is used along with neural networks to predict online the ground risk probability associated to future trajectories. Guidance inputs are computed in this way for the UAV to follow a reference path while ensuring a given level of safety for the mission, despite flight conditions that may differ from mission preparation (wind, altitude, speed). Computation time concerns are accounted for in the design of the algorithm with the objective to facilitate a possible future implementation on board of an UAV. More precisely, neural networks are used for fast risk prediction, as well as systematic search for resolution of the MPC problem corresponding to the risk avoidance component of the control. Simulation results are proposed to illustrate the proposed approach.