Enhancing human operator performance with long short-term memory networks in adaptively controlled systems

Abstract

The focus of this letter is developing a Long Short-Term Memory (LSTM) network-based control framework that works in collaboration with the human operator to enhance the overall closed-loop system performance in adaptively controlled systems. The domain of investigation is chosen to be flight control, although the proposed approach can be generalized for other domains such as automotive control. In accordance with this choice, an adaptive human pilot model is used as the mathematical representation of the pilot during the technical development of the method. An LSTM network is designed in such a way that it predicts and compensates for the inadequacies of the human operator's decisions while they fly an aircraft that has an adaptive inner loop controller. The simulation results demonstrate that the tracking performance is improved, and the pilot workload is reduced.