ANR OLYMPIA
Control by neuro-dynamic programming:
stability, robustness and optimality

   

Home
Work Packages
Members
Publications
Events and mobility
Miscellaneous

Home

Presentation

 

Neuro-dynamic programming (N-DP) offers a range of powerful tools to (nearly) optimally control general nonlinear dynamical systems. N-DP is at the heart of the most resounding successes in reinforcement learning and is extremely appealing for control engineering as it may enable the systematic control of complex systems. Nevertheless, several major methodological challenges need to be addressed to fully leverage the potential of N-DP in control engineering, among which the question of robust stability and the computational aspect of these algorithms. In this context, the aim of the OLYMPIA project is to develop analytical and design tools to

  • guarantee the robust stability of nonlinear systems controlled by N-DP
  • tailor the original algorithms to mitigate their computation complexity by exploiting control theoretic properties
  • take into account errors, which inevitably arise when implementing such controllers, and analyze their impact on the closed-loop systems properties
We have identified a number of fascinating research directions where the team's expertise on nonlinear systems, Lyapunov stability as well as hybrid techniques are essential for the success of the OLYMPIA project.

OLYMPIA is a fundamental research project organized in three main technical work-packages. In the first work-package, the focus is on dynamical systems controlled either by dynamic programming algorithms or neural networks. The objective is to certify stability and robustness guarantees for the closed-loop system. In WP2, the idea is to tailor the algorithms at hand, and not to use them off-the-shelf, to ensure the desired stability properties. The computation effort will also be taken into account and efficient algorithms will be devised. Finally, the outcome of WP1 and WP2 will be merged in WP3 to come up with neuro-dynamic programming control strategies endowing the closed-loop system with robust stability guarantees. The implementation of these algorithms will be particularly investigated.

Consortium

 

The project brings together members of the CRAN (Nancy), LAAS (Toulouse) and LAGEPP (Lyon)

News

 

September 2024 - Romain Postoyan will give a keynote lecture entitled "When dynamic programming meets Lyapunov theory: robust stability and improved near-optimality guarantees" at IFAC MICNON in Lyon

June 2024 - Daniele Astolfi will deliver a semi-plenary talk at the French national conference SAGIP in Lyon

April 2024 - The project kick-off meeting was organized in Lyon (LAGEPP)

March 2024 - The project is officially launched!

Acknowledgement

 

The OLYMPIA project is funded by the Agence Nationale de la Recherche (ANR).