Deep reinforcement learning reveals fewer sensors are needed for autonomous gust alleviation

Although both the public sector and defense agencies are interested in urban uncrewed aerial vehicle (UAV) mission performance, fixed winged aircraft are still incapable of adapting to the complex aerodynamics within a city environment [1, 2, 3, 4, 5, 6]. Currently, the most dynamic environments are dominated by multirotor flight vehicles; however, the highly maneuverable and responsive quadrotor design suffers from substantial weight and power constraints, limiting the operational range and on-board computational capabilities needed for autonomy [7, 8, 9, 10]. Current fixed wing UAVs have greater range but are not as maneuverable [11]. Counter to both rotorcraft and traditional fixed wing UAV design, birds can adapt their wing shape as the environment changes to achieve both efficient and maneuverable flight [12]. This ability supports birds of prey in navigating through complex environments [13], or rejecting perturbations in a gusty environment [14, 15].