A Data-enabled Dual Learning based Online Receding Horizon Safe-Critical Control for Nonlinear Systems under Uncertainty

In this paper, a real-time reliable receding horizon control (RHC) with a guaranteed safe adaptation mechanism is developed for uncertain complex nonlinear systems. Ensuring receding horizon optimality and safety, particularly in the presence of uncertain nonlinear system dynamics, poses a significant challenge in both control and learning societies. To tackle this challenge, a novel safe-critical RHC framework has been developed to enhance classical RHC with the capability of prioritizing system safety by timely recognizing and adapting environmental uncertainties. Specifically, the developed frame-work utilizes a novel dual-learning approach with slow learning to recognize environmental uncertainties and further refine RHC along with a situation-aware physics-informed neural network (SA-PINN), and fast learning to ensure system safety by using a safe-critical control with fast learned adaptive control barrier (FA-CBF) function. Therefore, slow learning in the developed dual-learning approach can provide optimal RHC albeit with longer computation time, while the fast learning component provides safe control effectively adapting to the uncertain environment in real-time.