Current PhD Students
Collaborative decision making for autonomous intersection crossing,
Subject : Improving the management of unsignalised intersections remains a critical issue, due to the multitude of driver manoeuvres and inherent conflicts that affect traffic safety and efficiency.
The introduction of autonomous driving technology may therefore be a solution to improve intersection crossing. However, the cohabitation of autonomous and human-driven vehicles could create uncertain and complex interactions and could jeopardise the coexistence of autonomous vehicles and other road users in a mixed environment.
As a strategic part of the autonomous car, the decision process ensures the link between the observations of the environment and the reactions of the vehicle. This component is responsible for the actions chosen. These actions must be understandable to other road users, in order to ensure safe and efficient interactions. To this end, it is relevant to design algorithms that could model human decision processes. Dynamic game theory is an effective paradigm for dealing with interactions between different decision agents.
In this work, we propose a decision making approach based on a game theory framework to study intersection crossing in the urban scenario of a congested intersection. The research question driving this work is to address the problem of decision making and path generation in a highly interactive environment, considering path optimisation problems, input constraints and the general non-linear state.
Past PhD students
Optimisation pour la prise en compte de cahiers des charges génériques en automatique, démarré en octobre 2003.
* Optimisation to take into account generic specifications in control
|Director of the thesis. Doctoral thesis in science Paris-Sud Orsay. The problem: for a better consideration of a specification, it is possible not to start from a particular criterion but to use a general criterion gathering all the specifications. In this case, the requirements are formulated more precisely but the resulting optimisation problems are non-convex and have local minima. This problem is studied as an extension of the work on convex formulations (H-infinite optimisation, LMI, …) using different methods of general optimisation: use of a priori information (in the form of an intelligent initial corrector); use of parameter changes for conditioning (taking into account the expected behaviour of the solution), hierarchisation of the global problem into sub-problems with superior resolution qualities. The classification and study of the most used criteria show a geometrical kinship between the criteria: they are not derivable and in general not Lipchitz, on the other hand they are almost everywhere. For the global resolution, new methods of exploration of the parameter space are developed and used, in particular to exploit the particular geometry of the criteria, we will be interested in subgradient type approaches.
Optimisation de production transport et stockage d’un réseau multi énergie, démarré en octobre 2002.
* Optimisation of generation, transmission and storage in a multi-energy network
|Director of the thesis. PhD thesis in science Paris-Sud Orsay. Co-financing with the Control Department of the EDF Research Department Context: on the scale of a large industrial site or an urban area, energy management represents a real economic challenge. Although the diversification of possible sources (electricity, gas, fuel oil, heat, cold, pressure, etc.) makes it possible to envisage greater technical and economic efficiency of a heterogeneous energy mix, it also presents risks and difficulties. Subject: The aim of this subject is to design optimisation methods for a heterogeneous fleet, over a time horizon of a few days, allowing optimal management of production. The robustness of the proposed methods should ensure their efficiency even in case of errors in the load forecast. This work requires the creation of control and simulation models for heterogeneous fleets (multi-energy), the development of optimisation methods adapted to the planning and control of the process.
Robustification de lois de commande prédictive par la paramétrisation de Youla. septembre 2000-septembre 2003.
* Robustification of predictive control laws by Youla parametrization
|Occasional collaboration in the supervision of first year students. Doctoral thesis from the University of Paris-Sud. Development of a method of retuning a corrector, using Youla’s parametrisation, adapted to the “predictive method” context. Development of a parametrization method adapted to two degrees of freedom correctors.
Machines tournante sur paliers magnétiques actifs : modélisation et contrôle anti-vibratoire numérique, 1999-2002.
* Active magnetic bearing machines: modelling and digital anti-vibration control
|Director of the thesis. Doctoral thesis in science Paris-Sud Orsay, Cifre agreement with S2M. Prize for the best student thesis CCA 2002. Obtained a national patent then a world patent. Context: on the scientific level, S2M is known as the industrial reference in the design of machines of all kinds (large compressors, turbomachines, machine-tool spindles, turbopumps, etc.). ); this leadership can only be maintained if it is accompanied by a certain scientific influence and a community of interest with a university partner. Internally, S2M wishes to modify and modernise its design and analysis tools, in order to harmonise and integrate the entire design chain into a single environment. Topic: Implementation of modern robust control methods, multivariate identification in state space. Specific study of new anti-vibration approaches. Integration of these different approaches.
Analyse de cahier des charges en automatique par optimisation convexe, 1999-2002
*Analysis of specifications in automation by convex optimisation
|Director of the thesis. PhD thesis in science Paris-Sud Orsay, Collaboration with EDF on industrial applications validating the developed methods. MENRT grant. Topic: For the feasibility analysis of specifications, two approaches have been developed: a trajectory approach which allows the study of the intrinsic performance limits attainable in the input-output of a system, and a transfer approach which allows the study of the compatibility of temporal and frequency specifications expressed in the input-output of a loop system. This second approach uses the de Youla parameterisation. A physical interpretation of the de Youla parameter based on an original presentation is presented. In addition, the techniques developed have been applied to an industrial case: the regulation of the water level in a steam generator for a pressurised water reactor.
Méthodologie de prise en compte de la robustesse dans les techniques de linéarisation : application au pilotage d’un missile fortement manœuvrant. 1996-1999.
* Methodology for taking robustness into account in linearisation techniques: application to the control of a highly manoeuvrable missile
|Thesis director. Doctoral thesis in science Paris-Sud Orsay, Cifre agreement with Aérospatial-Matra. Topic: Several methodological approaches for the control of highly manoeuvring machines are developed and compared: non-linear PID, non-linearity compensation method, linearisation and linearisation on several time scales. Different correctors are proposed. The interest and weakness of each one are highlighted and compared in terms of closed loop behaviour, synthesis complexity and implementation complexity. All the methodologies developed aim to take robustness into account: via the notion of local margin for the non-linear PID, via the choice of a good synthesis model for the non-linear controls, via the distribution of slow and fast dynamics when taking into account the time scale. All results were fully validated at Aerospatial-Matra and compared to reference controls.