Theses

Abstract:Our research lies at the intersection of Control Theory and Robotics, with a focus on event-driven control and aerial robotics. Control theory typically develops algorithms to drive dynamic systems to desired setpoints using feedback loops. Traditional periodic feedback control can be resource-intensive, especially for embedded systems, whereas event-driven control reduces resource usage by updating the control signal as needed, maintaining or even enhancing the closed-loop system performance. Besides theoretical contributions to this topic, our research work also includes the design and implementation of control strategies (not necessarily event-driven) for embedded systems in robotics, aiming for resource efficiency and covering various robots like unmanned aerial vehicles, aerial manipulators or cable-driven parallel robots, with experimental assessments.
Our research perspectives involve event-driven control for robotics, proposing new trends with preliminary results for a complete event-driven control architecture from perception to actuation. This includes designing an event-driven sensorimotor chain and perception-action algorithms. Prospective projects emphasize applications of the event-driven paradigm to improve efficiency and performance in robotics, such as frugal navigation for quadcopters in dark environments, or efficient flap-and-glide flight control of flapping-wing robots for better agility and endurance.
Overall, this manuscript underscores the potential of event-driven control to enhance the efficiency and performance of robotic systems, paving the way for future innovations in frugal design and control.

Abstract:The demand of electronic components in all embedded and miniaturized applications encourages to develop low-cost components, in term of energy consumption and computational resources. Actually, the power consumption can be reduced when decreasing the supply voltage and/or the clock frequency, but with the effect that the device runs more slowly in return. Nevertheless, a fast predictive control strategy allows to dynamically manage this tradeoff in order to minimize the energy consumption while ensuring good performance of the device. Furthermore, the proposals are highly robust to tackle variability which is a real problem in nanometric systems on chip.
Some issues are also suggested in this thesis to reduce the control computational cost. Contrary to a time-triggered system where the controller calculates the control law at each (constant and periodic) sampling time, an event-based controller updates the control signal only when the measurement sufficiently changes. Such a paradigm hence calls for resources whenever they are indeed necessary, that is when required from a performance or stability point of view for instance. The idea is to soften the computational load by reducing the number of samples and consequently the CPU utilization. Some simulation and experimental results eventually validate the interest of such an approach.

Abstract:The main goal of this training was to modelize the request treatment in a database server in order to reproduce the thrashing event which occurs when the system is overloaded. A fluid approach was used to establish a first simple model. Then, we determined the available capacity of the processor (analyzing the request treatment and parallelisation costs) and established the relation between the number of requests into the server and the output throughput. We finally reproduced a typical thrashing curve and a parametric identification allowed the model to fit with the real system (a postgreSQL database server and TPC-C client generator). At the end, we improved the model and gave some points to work on the future.