Description du projet
Nouveaux algorithmes pour le contrôle des systèmes distribués
Les cadres décisionnels optimaux permettant de contrôler efficacement de grands groupes d’agents autonomes, tels que des essaims de robots, sans contrôle centralisé, font actuellement défaut. Dans ce contexte, le projet DiODe, financé par le Conseil européen de la recherche, créera des algorithmes pour contrôler les systèmes distribués, en mettant l’accent sur le contrôle décentralisé des essaims de robots. La recherche fournira également des outils afin d’étudier les mécanismes naturels de prise de décision en ce qui concerne les organismes vivants, depuis les réseaux intracellulaires jusqu’aux populations d’animaux qui prennent des décisions. Le projet abordera plusieurs objectifs, notamment la conception des mécanismes distribués pour mettre en œuvre des compromis optimaux entre l’échantillonnage d’informations et la prise de décision. Il convertira également la théorie en applications pratiques pour les systèmes artificiels et développera des outils de modélisation accessibles aux scientifiques.
Objectif
This grant will develop and translate a unifying framework for optimal decision-theory, and observations of natural systems, to design distributed algorithms for decentralised decision-making. This will enable a technological step-change in techniques for controlling distributed systems, primarily demonstrated during the grant by decentralised control of robot swarms. These algorithms and associated methodology will also provide hypotheses and tools to change the way scientists think about and interrogate natural decision mechanisms, from intracellular regulatory networks, via neural decision circuits, to decision-making populations of animals. Specific objectives are:
1. Distributed value-sensitive decision-making: undertake optimality analyses of the applicant’s existing decentralised decision-making algorithms based on observations of collective iterated voting-processes in honeybees, and extend these.
2. Distributed sampling and decision-making: design distributed mechanisms that implement optimal compromises between sampling information and making decisions based on that information.
3. Individual-confidence and distributed decision-making: translate machine learning theory to collective behaviour models, designing mechanisms in which weak decision-makers optimally combine their decisions to optimise group performance.
4. Optimal distributed decision-making in collective robotics: translate theory from objective 1 to 3 towards practical applications in artificial systems, demonstrated using collectively-deciding robots.
5. Development of tools for life scientists and validation of theoretical predictions in natural systems: interact with named collaborators to investigate identified decision mechanisms in single cells, in neural circuits, and in social groups. Develop accessible modelling tools to facilitate investigations by life scientists.
Champ scientifique
- natural sciencesbiological sciencesecology
- natural sciencesbiological scienceszoologyentomologyapidology
- natural sciencesbiological sciencesneurobiologycomputational neuroscience
- natural sciencescomputer and information sciencesartificial intelligencemachine learning
- natural sciencescomputer and information sciencessoftwaresoftware applicationssimulation software
Programme(s)
Régime de financement
ERC-COG - Consolidator GrantInstitution d’accueil
S10 2TN Sheffield
Royaume-Uni