The McLEOD group (Modelling and Control of Complex systems in Large Environments requiring Optimized Decision) relates to the modelling, running and observing of complex systems. These issues are mainly addressed by two scientific communities: the “computer” community which deals with modelling and high-level decision-making; the “automation” community which works on the control and application of low-level setpoints while offering tools for the observation of dynamic systems. Here we propose a convergence of their points of view and approaches by considering complex systems in the form of intelligent entities (agents), potentially in interaction and whose behaviours follow the life cycle: perception & modelling -> decision -> action; in other words, a closed loop control.

McLEOD research

While the“modelling”, “decision” and “action” steps are at the heart of the group’s research, the “perception” part will form the basis of collaborations with other CERI SN groups. They may involve teacher-researchers specialising in data vision and processing and the group’s automation engineers (particularly through the notion of observation).

At the “decision” level 

The group’s work developed at “decision” level will be clearly positioned in artificial intelligence. In particular, they make scientific contributions in three sub-fields of AI:

  • optimisation under constraints with the development of algorithms for solving COP (Constraint Optimisation Problems) / DCOP (Distributed COP) type problems;
  • decision-making under uncertainties based on the use of stochastic models and reasoning such as Markovian models (MDP, DEC-MDP, etc.);
  • modelling of complex systems in the form of multi-agent systems for decision and/or simulation purposes;
  • distributed and decentralised decision-making via coordination mechanisms between agents making it possible to converge towards a collective decision.

 

At the “action” level

The work developed at the “action” level will result from automation and will focus more particularly on:

  • the control/running of complex and large-scale systems, presenting non-linearities, variable delays, and hybrid dynamics;
  • coordination of actions in real time taking into account unforeseen changes in the environment;
  • systems modelling (model reduction in order to keep an exploitable tool despite the complexity and the a priori dimension of the system, data-driven modelling for systems for which a “white box” type knowledge model is unattainable, etc.);
  • observation of systems (in the sense of automation) allowing the optimisation of the use of sensors (in numbers, in location, in a “scrutiny” body, etc.);
  • the stability of the control laws of the systems studied.

At the crossroads of automation and AI

The work developed at the crossroads of automation and AI could lead to the exploration of original issues such as:

  • the implementation of a hybrid control architecture allowing decision and control models on different time scales with a functional level (rapid and safe response) supervised by a deciding level;
  • supervision of switching systems: adapting setpoints to the constraints of physical systems and integrating switching possibilities in deliberative models;
  • or the use of optimisation methods for optimal sampling of network systems.