This talk will focus on the implementation of distributed control applications on embedded platforms. Many embedded systems now consist of up to 100 processors or electronic control units (ECUs) that communicate using different bus protocols like FlexRay, CAN or Ethernet. Such a setup is common in the automotive domain, as well as in many other domains like avionics, industrial automation and smart buildings. Traditionally, the process of designing control algorithms start with system identification or modeling, followed by determining the appropriate control laws, computing their parameters, and finally verification or simulation to ensure that the required control performance requirements are satisfied.

However, all of these are done at a high level of abstraction where most of the implementation details are either neglected or modeled in a pessimistic manner as a lumped error. However, as embedded systems become more complex, such an approach either leads to a large semantic gap between high-level controller models and their actual implementations, or an unnecessarily expensive or pessimistic design.

We will discuss techniques for addressing this problem, through a cyber-physical systems oriented design approach, which requires a co-design of controllers and the architectures on which they are implemented. In particular, we will show how taking conventional platform resources - such as computation, communication, and memory - into account necessitates a redesign of control algorithms compared to traditional controller design approaches that are agnostic to such resources.