Ph. D. Project
Decentralized control with communication and computation constraints
2016/09/01 - 2019/08/30
Other supervisor(s):
Janneteau Christophe ( , Boc Michael (
1 Scientific context

Embedded systems have become a predominant component of modern engineering systems. An important application domain of
embedded technology is the field of control systems where embedded software is executed in order to control a physical plant. The
software receives information about the plant, processes this information and determines control actions that are applied to the
plant. Moreover, these systems are often interconnected with others forming a network. A typical application of such networked
embedded systems is the decentralized control of fleets of robots.

Figure 1: Each system compute its control action using information received through a time varying interaction graph from its

2 Objectives

Traditionally, the design of embedded control systems assumes a separation of concerns between computation and control. An
integrated approach where the constraints due to limited or shared computational resources are taken into account in the synthesis of
control strategies would enable the development of high quality embedded controllers with guarantees of safety, stability and
performance, while optimizing the usage of the available computational resources. The main objective of this thesis is to design
decentralized controllers that integrate the communication and computation constraints. In other words, we specify the computation
budget and the communication bandwidth and range and we want to design a controller that can be executed under these constraints.
In order to satisfy the communication constraints we will impose a limited number of simultaneous communications per agent. The
computation constraints will be taken into account by designing simple control laws that require small computation loads. This will be
basically done by decoupling the controller in two parts. The first will compute reference trajectories based on standard consensus
algorithms while the second will design tracking controllers for each agent independently from the others.

Moreover, these results will be illustrated both in simulations and on a specific platform that has to be designed for this goal. A part of
this PhD will be dedicated to build this experimental platform composed of a fleet of robots.
Multi-agent systems, control theory, constrained control systems, communication
Control Identification Diagnosis
This project is financially supported by the Regional Council of Lorraine and FEDER