Towards improved adaptive control: human pilot models & memory-augmented architectures

To facilitate the implementation of adaptive control methods, this dissertation introduces novel solutions to key problems that hinder the employment of adaptive controllers in industrial applications. We present techniques that are inspired by humans’ versatility in the control loop, where we focus on understanding how humans adapt in the face of anomalies, and how they use their memory to better recover from them. Towards that end, we propose adaptive human pilot models suited for the prediction of human behavior in the loop with an adaptive controller. These models serve as valuable tools to test the interactions between human pilots and adaptive control systems in the simulation environment in order to ensure safe operation in the presence of an anomaly. Furthermore, the development of the models is carried out based on rigorous Lyapunov stability analyses, which can provide analytical insights into how to better design adaptive con-trollers for manned applications. Apart from their unfavorable interactions with human pilots, another issue that accounts for the scarce employment of adaptive controllers in piloted applications lies in their transient characteristics. While numerous works are devoted to improving the transients of adaptive control systems, in this dissertation, we focus on taking advantage of it first by providing adaptive controllers with human-like memory capabilities. We propose a memory architecture that can make use of stored data about the transients of previously experienced anomalies to aid in obtaining a resilient system against uncertainties. Thus, the proposed memory architecture enables adaptive controllers to rely on memory rather than exploration to better recover from familiar anomalies. The effectiveness of the architecture is validated through numerical simulations, and a rigorous Lyapunov stability analysis is provided.