Motivated by the need to adapt to changing models and reduce modeling errors in model-based control and reinforcement learning, we develop data-driven, adaptive system identification methods to identify the unknown parameters in a system model. The methods we develop can achieve parameter identification without restrictive persistence of excitation assumptions typically utilized in the online parameter identification literature and without numerical differentiation of the state that is typically required in modern data-driven parameter identification techniques. We are also able to generalize this technique to achieve simultaneous online state and parameter estimation using input-output measurements.


Get in touch
(405) 744-5900