Marin Lujak

IMT Lille Douai, France

Distributed optimization and decision-making for large and complex systems

Research motivation

Can you imagine a smart environment that perceives your needs and serves everyone seamlessly as needed? Or a fleet of driverless cars that serve the whole city population based on individual needs? Imagine our cities without traffic jams and without air pollution. Imagine our cities without siren sounds, caring seamlessly for the wellbeing of our citizens. Imagine a city where everyone will feel well and at ease and will be helped when needed. Imagine a World that better distributes available resources and provides for prosperity and well-being based on fair, just, and ethical decisions.

That is the version of the World that I envision and dream of; the visions that lead my research in multi-agent systems with distributed optimization and distributed decision-making.

The road towards that vision is anything but easy and fast. There are many actors and technologies necessary to make this vision come true. One of these technologies are distributed and scalable decision-making systems that divide a decision problem into well-sized subproblems solved by relevant decision makers sufficiently fast and with quality of solution guarantees. Among many open issues, there is the one related with the ways to design subproblems for relevant decision makers and the means to orchestrate their decisions and scale them considering a desired system behavior and well-being. There is also the question of incentives: how to incentivize participating actors to create in them the sense of satisfaction when behaving collaboratively and not to exclusively rely on negative reinforcement and punishment measures such as penalties to realize systemís objectives.


Research question, scientific approach, and application

How to efficiently coordinate individual actors of large complex systems in a scalable and robust way with quality of solution guarantees while considering both individual interest and system performance?

To answer this question, I use mostly the paradigm of multi-agent systems (MAS), combinatorial optimization, and artificial intelligence, where each (human, software, infrastructure, or robot) actor is an agent making decisions independently and autonomously based on its local computations and the communication with others.

The challenge in the distribution and decentralization of decision-making lies in the complexity of both the coordination problem at hand and a solution approach that should consider the balance between local computation and communication with others.

I develop mathematical programming models and algorithms that manage the systemís bottlenecks and decompose the coordination problem considering the constraints between individual and shared decisional variables.

The result is a distributed or decentralized decision-making architecture that enables each agent to decide in its best interest considering a momentary context and system constraints.† These constraints are modelled to influence individual decisions such that given fairness and social welfare criteria are optimized.

The quality of solution strongly depends on the quality of available information and is based on sensory and communication technologies, whose developments give rise to new real-time agent coordination technologies applicable in various real-world industrial and business contexts.

My long term objective is to lower the inefficiency of the decision-making paradigm based on Nash equilibrium through plausible MAS coordination solutions that will get closer to the system optimum while increasing fairness and social welfare.

I apply my research to resolving societal challenges including:

 Smart, Green and Integrated Transport and in more specific the development of distributed Route Guidance Systems (RGS) for the assignment of efficient, fair, and envy-free routes to users in a distributed way in real-time and the coordination of commercial fleets without the need for a dispatching center.

 Emergency Management: i) distributed coordination of Emergency Medical Systems with ambulances and ii) distributed coordination of evacuation routes in emergency evacuation of buildings, neighborhoods or cities.

 Multi-Robot Coordination: i) mobile industrial robots within robot teams working on a factory shop-floor and ii) teams of mobile service robots with humans for human assistance and support (newly obtained COMRADES Project).