Neuro-adaptive tracking control of a hypersonic flight vehicle with uncertainties using reinforcement synthesis

In this paper, the neuro-adaptive optimal nonlinear control approach and algorithm are proposed for the tracking control of an air-breathing hypersonic aircraft considering uncertainties. Based on the reinforcement learning mechanism, the neuro-adaptive control agent is constructed using actor-critic architecture, which consists of two interacting neural networks, one for optimal control protocol, known as the actor NN, and the other for policy evaluation, known as critic NN. The optimality conditions for this adaptive controller are derived by using the discrete minimum principle. In the meanwhile, the parametric and mismatched uncertainty and unmodeled nonlinearity are handled by using another neural network named as UNN with aid of the concept of a virtual plant. The output of the network in the virtual plant helps to estimate the optimal control with varying dynamics of air-breathing hypersonic flight aircraft. Simulation results are presented to verify the effectiveness of this design method for the tracking control of the air-breathing hypersonic aircraft in the presence of uncertainties.