Real-Time Optimal Control via Deep Neural Networks: Study on Landing Problems

Recent research has shown the benefits of deep learning, a set of machine learning techniques able to learn deep architectures, for modelling robotic perception and action. In terms of a spacecraft navigation and control system, this suggests that deep architectures may be considered now to drive all or part of the onboard decision-making system. In this paper, this claim is investigated in more detail, training deep artificial neural networks to represent the optimal control action during a pinpoint landing and assuming perfect state information. It is found possible to train deep networks for this purpose, and the resulting landings, driven by the trained networks, are close to simulated optimal ones. These results allow for the design of an onboard real-time optimal control system able to cope with large sets of possible initial states while still producing an optimal response.