Stage: Guaranteed learning-based control of dynamical systems

Type de recrutement
laboratoire des signaux et systèmes (L2S), CentraleSupelec
Fin de l'affichage
Détails (fichier)

Work place: Laboratoire des signaux et systemes, CentraleSupelec, Gif-sur-yvette, France


Application: Master’s students with good academic results, a solid background in control theory, and excellent

writing and communication in English. Furthermore, experience in machine-learning

techniques is preferable.


Keywords: Control theory, Learning-based control.


Context and objectives of the internship: In recent years, machine-learning approaches

have made a remarkable impact on the society as they have been successfully employed in

many areas including transportation [6], networking [12], healthcare [8], and education [7].

The obtained results have inspired the control community to use machine-learning techniques

to design safety-critical Cyber-Physical Systems (CPS) such as surgical robots [4],

autonomous vehicles [10, 6, 3] and air traffic collision avoidance systems [9]. Classical approaches

for demonstrating the safety of learning-based systems rely on the use of extensive

simulation testing, which are usually costly and incomplete. At the same time, recent rare

event failures such as the Tesla and Uber [1] autonomous driving crashes, have shown the

need for a better understanding of machine-learning algorithms to control complex CPS.

In this context, new approaches based on analytical proofs using control theory and algorithmic

proofs based on tools from formal methods are urgently needed. In principle, these

new approaches will make it possible to explore all scenarios when verifying or falsifying the

safety of learning-based systems, when used for verification purposes. They will also makes

it possible to learn guaranteed control policies when used for control purposes.

Hence, the goal of this internship is develop tools for guaranteed learning-based control of

dynamical systems. The intern will mainly work on the analysis of control loops including

learning-based components, either at the level of the system [2, 5] or at the level of the

controller [11]. Stability and performance guarantees of the closed-loop system will be analysed

using different tools combining reachability analysis and lyapunov-based approaches.


Duration: 6 months

Supervisors: Adnane SAOUD, Antonio LORIA, Mohamed MAGHENEM.


Contact: The candidate must send to the CV and marks

of all her/his MSc studies.



[1] US National Highway Traffic Safety Administration. [n. d.]. Investigation PE16-007. 2016.

[2] Alex Devonport, Adnane Saoud, and Murat Arcak. Symbolic abstractions from data: A

pac learning approach. arXiv preprint arXiv:2104.13901, 2021.

[3] Tommaso Dreossi, Shromona Ghosh, Alberto Sangiovanni-Vincentelli, and Sanjit A.

Seshia. Systematic testing of convolutional neural networks for autonomous driving,


[4] Daniel A Hashimoto, Guy Rosman, Daniela Rus, and Ozanan R Meireles. Artificial

intelligence in surgery: promises and perils. Annals of surgery, 268(1):70–76, 2018.

[5] Kazumune Hashimoto, Adnane Saoud, Masako Kishida, Toshimitsu Ushio, and Dimos

Dimarogonas. Learning-based symbolic abstractions for nonlinear control systems. arXiv

preprint arXiv:2004.01879, 2020.

[6] Philip Koopman and Michael Wagner. Autonomous vehicle safety: An interdisciplinary

challenge. IEEE Intelligent Transportation Systems Magazine, 9(1):90–96, 2017.

[7] Rosemary Luckin. Machine Learning and Human Intelligence: The Future of Education

for the 21st Century. ERIC, 2018.

[8] Gunasekaran Manogaran and Daphne Lopez. A survey of big data architectures and machine

learning algorithms in healthcare. International Journal of Biomedical Engineering

and Technology, 25(2-4):182–211, 2017.

[9] Amjad Rehman. Machine learning based air traffic control strategy. International Journal

of Machine Learning and Cybernetics, pages 1–11, 2012.

[10] Ahmad EL Sallab, Mohammed Abdou, Etienne Perot, and Senthil Yogamani. Deep reinforcement

learning framework for autonomous driving. Electronic Imaging, 2017(19):70–

76, 2017.

[11] Adnane Saoud and Ricardo G Sanfelice. Computation of controlled invariants for nonlinear

systems: Application to safe neural networks approximation and control. IFACPapersOnLine,

54(5):91–96, 2021.

[12] Mowei Wang, Yong Cui, Xin Wang, Shihan Xiao, and Junchen Jiang. Machine learning

for networking: Workflow, advances and opportunities. IEEE Network, 32(2):92–99,