Uncertainty-aware optimal flight state selection for a transitioning UAV via simulation-based learning

BITU is a transitioning UAV that can switch between fixed-wing, VTOL and hover-capable multi-rotor states, and other states in-between. Earlier work has developed a conceptual aerodynamic design and the fundamental models of BITU flight dynamics. This paper builds on this earlier framework, and aims to analyze the robust-optimal states that BITU should assume under different mission and flight environment scenarios – w.r.t. offering mission success and energy efficiency. Flight states are defined in terms of cruising speed, cruising altitude, and tilt arm angle (which allows BITU to assume hybrid flight states). A supervised learning approach is taken that maps the mission requirements (payload being carried) and wind conditions (wind speed, direction and gust, and their variations) to the optimal state for the given current flying environment. Neural Network models of the cruising speed and the arm tilt angle are trained using simulation based flight state optimization for a DOE of scenarios. While the cruising speed and tilt-arm angle are found to be both sensitive to the mission and wind conditions and strongly impact the flight performance (range and mission success rate), cruising altitude was found to have a marginal impact under the current simulation settings.