Event-Triggered Learning for Linear Quadratic Control

When models are inaccurate, the performance of model-based control will degrade. For linear quadratic control, an event-triggered learning framework is proposed that automatically detects inaccurate and triggers a new process model when needed. This achieved by analyzing probability distribution cost designing trigger leverages Chernoff bounds. In particular, whenever empirically observed signals located outside derived confidence intervals, we can provably guarantee this with high due to mismatch. With aid numerical hardware experiments, demonstrate bounds tight algorithm effectively distinguishes between probabilistic effects, such as noise. Thus, structured approach obtained decides beneficial.