Jérémy Boes

The Self-Adaptive Context Learning Pattern: Overview and Proposal

Over the years, our research group has designed and developed many self-adaptive multi-agent systems to tackle real-world complex problems, such as robot control and heat engine optimization. A recurrent key feature of these systems is the ability to learn how to handle the context they are plunged in, in other words to map the current state of their perceptions to actions and effects. This paper presents the pattern enabling the dynamic and interactive learning of the mapping between context and actions by our multi-agent systems.

Use Cases of Pervasive Artificial Intelligence for Smart Cities Challenges

Software engineering has been historically topdown. From a fully specified problem, a software engineer needs to detail each step of the resolution to get a solution. The resulting program will be functionally adequate as long as its execution environment complies with the original specifications. With their large amount of data, their ever changing multi-level dynamics, smart cities are too complex for a topdown approach. They prompt the need for a paradigm shift in computer science. Programs should be able to self-adapt on the fly, to handle unspecified events,, to efficiently deal with tremendous amount of data. To this end, bottom-up approach should become the norm. Machine learning is a first step,, distributed computing helps. Multi-Agent Systems (MAS) can combine machine learning, distributed computing, may be easily designed with a bottom-up approach. This paper explores how MASs can answer challenges at various levels of smart cities, from sensors networks to ambient intelligence.

neOCampus: A Demonstrator of Connected, Innovative, Intelligent and Sustainable Campus

The progress in communication technologies and data storage capacity has brought tremendous changes in our daily life and also in the city where we live. The city becomes smart and has to integrate innovative applications, and technology to improve the quality of life of its citizens. To experiment and estimate these innovations, we show that a university campus is a ground “in vivo” experiments adequate. The infrastructure deployed in the Toulouse III Paul Sabatier University campus, enabling its transformation in a smart and sustainable campus, is detailed. The approach to answer the main challenges a smart campus has to deal with, is described. Some challenges are illustrated with a case study on the balance between the energy efficiency and the comfort in a class.

Self-organizing multi-agent systems for the control of complex systems

Because of the law of requisite variety, designing a controller for complex systems implies designing a complex system. In software engineering, usual top-down approaches become inadequate to design such systems. The Adaptive Multi-Agent Systems (AMAS) approach relies on the cooperative self-organization of autonomous micro-level agents to tackle macro-level complexity. This bottom-up approach provides adaptive, scalable, and robust systems. This paper presents a complex system controller that has been designed following this approach, and shows results obtained with the automatic tuning of a real internal combustion engine.

Smart Is a Matter of Context

Smart cities involve, in a large scale, a wide array of interconnected components and agents, giving birth to large and heterogeneous data flows. They are inherently cross-disciplinary, provide interesting challenges, and constitute a very promising field for future urban developments, such as smart grids, eco-feedback, intelligent traffic control, and so on. We advocate that the key to these challenges is the proper modelling and exploitation of context. However, said context is highly dynamic and mainly unpredictable. Improved AI and machine learning techniques are required. Starting from some of the main smart cities features, this paper highlights the key challenges, explains why handling context is crucial to them, and gives some insights to address them, notably with multi-agent systems.

Intuitive Method for Pedestrians in Virtual Environments

Recent works about pedestrian simulation can actually be sorted in two categories. The first ones focusing on large crowd simulation aim to solve performance and scalability issues at the expense of behavioral realism of each simulated individual. The second ones aim at individual behavioral realism but the computational cost is too expensive to simulate crowds. In this paper, we propose an alternate approach combining a light reactive behavior with cognitive strategies issued from real life videos. This approach aims at the real time simulation of small crowds of pedestrians (one to two hundred individuals) but with concerns for visual realism regarding heterogeneous behaviors, trajectories and positioning on sidewalks.

Cooperative Multi-agent Approach for Computational Systems of Systems Architecting.

This paper addresses the modeling and design of Systems of Systems (SoS). It presents and illustrates a new generic model to describe formally such systems. This model is used to propose a SoS architecting approach based on adaptive multi-agent systems. In this approach, each component system composing the SoS uses a local cooperative decision process in order to interact with other systems and to collectively give rise to a relevant overall function at the SoS level. The proposed model as well as the proposed approach are instantiated with a simulated unmanned aerial vehicle scenario and compared with another approach dealing with collaboration between systems in a SoS.

Principles and Experiments of a Multi-Agent Approach for Large Co-Simulation Networks Initialization.

Simulating large systems, such as smart grids, often requires to build a network of specific simulators. Making heterogeneous simulators work together is a challenge in itself, but recent advances in the field of co-simulation are providing answers. However, one key problem arises, and has not been sufficiently addressed: the initialization of such networks. Many simulators need to have proper input values to start. But in the network, each input is another simulator’s output. One has to find the initial input values of all simulators such as their computed output is equal to the initial input value of the connected simulators. Given that simulators often contain differential equations, this is hard to solve even with a small number of simulators, and nearly impossible with a large number of them. In this paper, we present a mutli-agent system designed to solve the co-simulation initialization problem, and show preliminary results on large networks.

Lifelong Machine Learning with Adaptive Multi-Agent Systems

Sensors and actuators are progressively invading our everyday life as well as industrial processes. They form complex and pervasive systems usually called ”ambient systems” or ”cyber-physical systems”. These systems are supposed to efficiently perform various and dynamic tasks in an ever-changing environment. They need to be able to learn and to self-adapt throughout their life, because designers cannot specify a priori all the interactions and situations they will face. These are strong requirements that push the need for lifelong machine learning, where devices can learn models and behaviours during their whole lifetime and are able to transfer them to perform other tasks. This paper presents a multi-agent approach for lifelong machine learning