Quadcopter Trajectory Prediction and Wind Estimation Using Machine Learning

Abstract

Small unmanned aerial systems are heavily impacted by wind disturbances. Wind causes deviations from desired trajectories, potentially leading to crashes. In this thesis, we consider two inherently related problems: predicting quadcopter trajectory deviations due to wind disturbances and estimating wind velocity based on quadcopter trajectory deviations. The former is addressed using linear difference equation identification as well as neural network (NN) modeling. Simulations validate the use of linear difference equation identification as a tool to predict trajectory deviations in crosswinds and machine learning (specifically, long short-term memory (LSTM) NNs) as an approach to predict trajectory deviations in multidimensional wind. We approach the wind estimation problem from a machine learning perspective due to easier generalization of the NN to multidimensional winds. As in the trajectory prediction case, we use LSTM NNs to identify a model. The trained NN is deployed to estimate the turbulent winds as generated by the Dryden gust model as well as a realistic large eddy simulation of a near-neutral atmospheric boundary layer over flat terrain. The resulting NN predictions are compared to a wind triangle approach that uses tilt angle as an approximation of airspeed. Results from this study indicate that the LSTM NN based approach results in lower errors in both the mean and variance of the local wind field as compared to the wind triangle approach.