Learning the nonlinear inverse flight dynamics using radial basis functions

In this paper, we propose to use different optimization objectives to train a neural network to approximate the nonlinear inverse dynamics of a system. We implement and test this approach using radial basis function (RBF) networks to approximate the nonlinear inverse dynamics of a simulated high performance aircraft. The synthesised inverse dynamics controller performs well in tracking simulated trajectories. The use of optimization objectives allows us to deal with the issues of non-invertibility and stability of the inverse dynamics and to synthesise a nonlinear controller which has the desired performance characteristics