Ph. D. Project
Title:
Control design for nonlinear systems using Model-based and AI approaches. Application to robots
Dates:
2025/10/01 - 2028/09/30
Supervisor(s):
Other supervisor(s):
Mr. OUDANI Mustapha, MCF-HDR (mustapha.oudani@uir.ac.ma)
Description:
The thesis topic, in co-supervision between the International University of Rabat and the University of Lorraine, concerns the control of nonlinear
systems using hybrid approaches based on Models and Artificial Intelligence (AI) [1]-[2]. The practical aspect focuses on the control of robots. The
obtained results will be tested and validated through numerical simulation, and then on a robotic platform.
In classical control theory, implicit or explicit use of the system's dynamic model is made to construct a control law ensuring specific performance.
However, in many cases of physical systems, even when the model is known, it is very difficult to establish control laws based solely on the model.
Examples include underactuated and/or constrained nonlinear systems. Employing hybrid approaches based on both models and Artificial Intelligence
techniques opens up a very promising field but remains relatively unexplored. One of the application fields could be control design of cable robots.
In a recent study [3] based on reinforcement learning technique, with specific activation functions, the authors demonstrate the existence of solutions to
optimization problems under constraints for a class of linear dynamic systems, unlike the classical approach that has not provided a solution so far.
In [4], we developed a technique for solving a class of nonlinear PDEs based on physics-informed neural networks (PINNs) with remarkable precision
and speed compared to Eulerian discretization techniques. These techniques will be partly used in designing controllers for nonlinear systems.
The dynamic systems considered in this project represent a wide range of physical systems: nonlinear systems in a general form with asymmetric
saturation type constraints.
The thesis topic is divided into two main components:
- A methodological part aimed at developing generic approaches based on Models and AI for controlling nonlinear systems, with a particular focus on
real-time implementation and deployment.
- A part dedicated to validating the obtained results, initially through numerical simulation and then on the cable robot carrying an effector to perform a
given task. It is worth noting that such a system poses multiple scientific and technological challenges. The dynamic behavior is described by complex
nonlinear differential equations of large dimension with six degrees of freedom and very strong constraints. The development of control laws must take
into account the continuous cable tension constraints over time, to both keep them taut and prevent their breakage, while ensuring trajectory tracking.
Références
[1] I. Goodfellow, Y. Bengio, and A. Courville. Deep learning. MIT-Press, 2016.
[2] Y. LeCun, Y. Bengio, G. Hinton. Deep learning. Nature 521, pp. 436-444, 2015.
[3] F. A. Yaghmaie and D. J. Braun. Reinforcement learning for a class of continuous-time input constrained optimal control problems. Automatica, 99
:221⬓ 227, 2019.
[4] A. Farkane, M. Ghogho, M. Oudani, M. Boutayeb. Enhancing physics informed neural networks for solving Navier-Stokes equations. International
Journal for Numerical Methods in Fluids, 2023.
[5] R. Pannequin, M. Jouaiti, M. Boutayeb, P. Lucas et D. Martinez. "Automatic tracking of free-flying insects using a cable-driven robot". Science
Robotics, Juin 2020.
systems using hybrid approaches based on Models and Artificial Intelligence (AI) [1]-[2]. The practical aspect focuses on the control of robots. The
obtained results will be tested and validated through numerical simulation, and then on a robotic platform.
In classical control theory, implicit or explicit use of the system's dynamic model is made to construct a control law ensuring specific performance.
However, in many cases of physical systems, even when the model is known, it is very difficult to establish control laws based solely on the model.
Examples include underactuated and/or constrained nonlinear systems. Employing hybrid approaches based on both models and Artificial Intelligence
techniques opens up a very promising field but remains relatively unexplored. One of the application fields could be control design of cable robots.
In a recent study [3] based on reinforcement learning technique, with specific activation functions, the authors demonstrate the existence of solutions to
optimization problems under constraints for a class of linear dynamic systems, unlike the classical approach that has not provided a solution so far.
In [4], we developed a technique for solving a class of nonlinear PDEs based on physics-informed neural networks (PINNs) with remarkable precision
and speed compared to Eulerian discretization techniques. These techniques will be partly used in designing controllers for nonlinear systems.
The dynamic systems considered in this project represent a wide range of physical systems: nonlinear systems in a general form with asymmetric
saturation type constraints.
The thesis topic is divided into two main components:
- A methodological part aimed at developing generic approaches based on Models and AI for controlling nonlinear systems, with a particular focus on
real-time implementation and deployment.
- A part dedicated to validating the obtained results, initially through numerical simulation and then on the cable robot carrying an effector to perform a
given task. It is worth noting that such a system poses multiple scientific and technological challenges. The dynamic behavior is described by complex
nonlinear differential equations of large dimension with six degrees of freedom and very strong constraints. The development of control laws must take
into account the continuous cable tension constraints over time, to both keep them taut and prevent their breakage, while ensuring trajectory tracking.
Références
[1] I. Goodfellow, Y. Bengio, and A. Courville. Deep learning. MIT-Press, 2016.
[2] Y. LeCun, Y. Bengio, G. Hinton. Deep learning. Nature 521, pp. 436-444, 2015.
[3] F. A. Yaghmaie and D. J. Braun. Reinforcement learning for a class of continuous-time input constrained optimal control problems. Automatica, 99
:221⬓ 227, 2019.
[4] A. Farkane, M. Ghogho, M. Oudani, M. Boutayeb. Enhancing physics informed neural networks for solving Navier-Stokes equations. International
Journal for Numerical Methods in Fluids, 2023.
[5] R. Pannequin, M. Jouaiti, M. Boutayeb, P. Lucas et D. Martinez. "Automatic tracking of free-flying insects using a cable-driven robot". Science
Robotics, Juin 2020.
Department(s):
| Control Identification Diagnosis |