Event-Triggered Control for Networked Systems with Delays

Controlling uncertain systems through online learning is a promising approach, but resource-intensive learning algorithms often introduce significant computational delays, especially in systems with limited local processing power. To mitigate this, a common strategy is to deploy learning-based controllers on remote computation nodes, connected via communication channels, forming an in-network control architecture. This research establishes a control performance guarantee by deriving a tracking error bound for such in-network architectures, explicitly considering computational delays. This bound allows for diverse communication and computation strategies, including both time-triggered and event-triggered mechanisms. The study also highlights the trade-off between communication and computation performance for a given desired control outcome. To further enhance efficiency, an asynchronous event-triggered mechanism is devised for both control and online learning, specifically designed for scenarios with computational delays. This proposed event-triggered strategy is proven to achieve the same control performance as time-triggered scenarios while avoiding Zeno behavior, which is a common issue in event-triggered systems. The paper concludes by providing an explicit expression for the event-trigger condition for exponentially stabilizable systems and validating its effectiveness through simulations.