Safe learning-based observers for unknown nonlinear systems using Bayesian optimization

In this paper, a modular observer design methodology is formulated for nonlinear systems with partial model knowledge. Our consists of three phases: (i) an initial robust that enables one to learn the dynamics without allowing state estimation error diverge (hence, safe); (ii) learning phase wherein unmodeled components are estimated using Gaussian process-based Bayesian optimization; and, (iii) re-design leverages learned improve convergence rate error. The potential our proposed learning-based demonstrated on benchmark system. Additionally, certificates guaranteed performance provided.