Adaptive Tuning of Event-Triggered Control based on Machine Learning: Application to Mobile Robotics

This paper aims to optimize the frugal navigation of mobile robots by combining event-triggered control with artificial intelligence techniques. Event-triggered control updates the control signal only during significant changes in the system dynamics or its environment, thereby reducing the usage of limited onboard resources. This work is based on a discrete-time linear quadratic event-triggered control approach, which comes with a constant tuning parameter to set the tradeoff between resource utilization, performance, and control effort. Machine learning methods (linear regression and neural network) are explored to make this parameter dynamically adaptive to enable better and sparser trajectory tracking. Both simulation results on a digital twin and experimental results on a real two-wheeled robot show that combining event-triggered control with machine-learning techniques can efficiently improve robot navigation, offering adaptive and robust solutions across various configurations.