Markov Neural Network For Guidance, Navigation and Control
Sungyung Lim, Matthew R. Stoeckle, Brett J. Streetman, Matthew Neave · AIAA Scitech 2020 Forum · 2020
The paper proposes a Markov neural network for guidance, navigation, and control (GNC) designs. The proposed neural network is a temporal neural network evolving through the state that is the outcome of the Markov property of a stochastic dynamic system. At a specific time, a neural network represents either the complicated nonlinear behaviors of a dynamic system or a complicated GNC design problem. This neural network influences its future neural network only through the state, which is the special property to distinguish itself from other artificial neural networks. The proposed Markov neural network is directly solved by deep learning techniques. This paper demonstrates that the proposed Markov neutral network results in more accurate system identification than conventional approaches. It also discusses benchmark GNC design examples to be efficiently solved by the proposed Markov neural network, indicating that GNC designs and state of the art artificial neural networks of deep learning are systematically fused.