Parameter Filter-based Event-triggered Learning

Model-based algorithms are deeply rooted in modern control and systems theory. However, they usually come with a critical assumption—access to an accurate model of the system. In practice, models are neither perfect nor time-invariant. Even precisely tuned estimates of unknown parameters will deteriorate over time. We propose to combine statistical tests with dedicated parameter filters that track unknown system parameters from state data. These filters yield point estimates of the unknown parameters and, further, an inherent notion of uncertainty. This allows us to detect changes in the dynamics. When the point estimate leaves the confidence region, we trigger active learning experiments. Thus, models are only updated when necessary and statistically significant while ensuring guaranteed improvement, which we call event-triggered learning. We validate the proposed method in numerical simulations of a DC motor in combination with model predictive control.