Learning to soar in strongly turbulent flows

A glider moving in a turbulent flow will continuously lose energy via drag. To balance this loss in energy and soar, energy must be continuously extracted from the flow, either by localizing in stable ascending currents (thermal soaring) or in a stable shear region(dynamic soaring). Recent observations of soaring birds show convoluted trajectories distinct from characteristic patterns exhibited during thermal and dynamic soaring, raising the intriguing possibility that energy can be extracted purely from transient ascending currents or shear. In this work, we simulate gliders navigating in a turbulent flow, which use their past experience to infer a strongly fluctuating flow field and actively make decisions. We build the decision-making component using a Monte Carlo tree search (MCTS), which exploits an adaptive filtering and prediction system to consider many paths into the future and execute a trajectory that maximizes the energy gained. We demonstrate the ability of gliders to extract energy from the flow, and identify the significant factors necessary for effective turbulent navigation.