Neural-Network-Based Soaring Flight in Randomized Environments

What is it about?

Biologically-inspired flight techniques such as dynamic soaring can extend the flight endurance of such aircraft if implemented on autonomous aerial vehicles. This research presents a method of training neural networks that can steer an aircraft along dynamic soaring trajectories while subject to disturbances, uncertainties, and noise.

Featured Image

Why is it important?

Our methods present a way of evaluating neural network control systems in stochastic environments, comparing topologies of differing networks, and generating simple but robust neurocontrollers.