Multiagent System Team - Laboratoire d'Informatique de Grenoble
The MAGMA team is interested in engineering MAS, modelling and simulating complex social systems, and group dynamics. The design and implementation of multi-agent systems poses deep scientific problems concerning collective problem solving, inter-agent communication, planning, cognitive representations and cooperation. For 15 years MAGMA has worked on theoretical and practical issues concerning MAS and has developed agent based applications in many domains, e.g. Energy Management, Smart Homes, Health, Crisis and Emergency Management, Companion agents for everyday life, e.g. education and health. Our work is structured around 4 themes.
Agent based Modelling and Social Simulation
Human societies are a perfect example of a complex adaptive system. Our desire to be able to predict future events, to explain past events, or to have realistic training tools has led to the rise of ABMSS. However, the complexity and richness of human behaviour makes modelling humans an intriguing and challenging problem. We are interested in modelling human communications, cognitive behaviours and socio-technical interactions. The goal is to understand the interpretive and self-adaptive mechanisms that humans use to deal with problems and how collective intelligence emerges and is used in problem- solving.
Autonomous Planning Agents
In the "Automated Planning" theme, the stress is laid on devising theoretical underpinnings and practical tools for agent rational decisions so as to make them work as a team, coordinate their activities and achieve cooperative behaviours of autonomous systems as well as to interact appropriately with human end-users. To that purpose, we develop planning algorithms adapted to different needs: distribution, real-time constraints, adaptation to the context of use, human "usability".
Affective Relational Agents
Extending MAS to include humans we can form "mixed communities of agents and humans". A major objective is to develop agents that are attractive and closer to users and that are believable interlocutors. This has led to an increasing interest in Affective Computing, particularly Embodied Conversational Agents (ECAs). Our aim is to model and implement “Affective Relational Agents” by placing emotion at the heart of the human-agent interaction. Linking emotions and communication we defined an Agent Expressive and Multimodal Conversation Language. We now focus on how ECAs with expressive and affective characteristics can establish and maintain long-term human-agent relationships.
Engineering Multi-agent Systems
The engineering of multi-agent systems (MAS) is evolving towards user-centering, scalability and effective use. The VOWELS method allows user modelling in terms of preferences, capabilities, and activities. Methods and tools for the protection of personal data are investigated at the individual and social levels. Dynamics of trust models and social networks are studied. Evaluation, either microscopic or macroscopic, allows us to check actual behaviour against the expected one. VOWELS supports an iterative design and development cycle of purposive multi-agent systems in close cooperation with the user.