The control of networked embedded systems poses challenging new problems questioning standard assumptions in systems and control theory. Control design has traditionally benefited from a useful abstraction that enabled the separation of continuous mathematical models of control systems and feedback laws from hardware/software and real-time implementation details such as sensor quantization, sampling rates, real-time scheduling, etc. The prevailing assumption that control loops were implemented in highly engineered dedicated hardware systems can no longer be made. Networked embedded systems have limited resources that drastically reduce the quality of control. In the context of this project, formal methods for the integration of control design with real-time scheduling are being developed in order to overcome some of the limitations imposed by the separation of these two aspects of embedded control design.
Based on new stability abstractions for control systems, computational tools for automated synthesis of real-time schedulers enforcing control, timing, scheduling and power consumption requirements are being built. The resulting schedulers satisfy the desired specifications by construction thus providing formal guarantees of operation and performance. The results of this project are being evaluated by implementing stabilizing feedback control laws on MICA motes for two familiar benchmark problems in control engineering: the double inverted pendulum, and the ball and beam.
This integration between control and real-time scheduling allows practicing engineers to fully exploit the limits of existing embedded technology and results in cost reductions for the industry relying on embedded hardware. A particularly relevant example is the automotive industry where a small reduction in the cost of embedded hardware multiplied by the large number of produced units results in a considerable reduction in production costs.
