Paulo Tabuada
Department of Electrical Engineering
CyPhyLab
University of California, Los Angeles
Event-triggered and self-triggered control
Control task have been traditionally treated as periodic hard real-time tasks for the purpose of real-time scheduling. This periodicity assumption results in a separation of concerns that has been very successful over the years: control engineers can safely ignore implementation details when designing feedback controllers; and software engineers can safely ignore stability and other control considerations when scheduling control tasks on digital platforms. However, the periodicity assumption also leads to conservative designs that are inadequate for existing resource limited platforms such as the ones used in sensor-actuator networks. In this talk I will revisit real-time scheduling of control tasks with the purpose of illustrating the benefits of relaxing periodicity. I will propose event-triggered and self-triggered scheduling policies with guaranteed performance and improved usage of computational resources with respect to traditional periodic schedules.