Research on Multi-Agent Systems

Distributed arti cial intelligence is the sub field of arti cial intelligence that focuses on complex systems that are inhabited by multiple agents. The main goal of research in this area is to provide principles for the construction and application of multi-agent systems as well as means for coordinating the behavior of multiple independent agents.

There is a variety of application fi elds for multi-agent systems and, what is of special interest, multi-agent reinforcement learning systems. The following list gives a brief and, for sure, incomplete overview of practical applications for learning approaches used in the scope of cooperative multi-agent systems. Reinforcement learning and related approaches have been applied to optimize agent behavior in the scope of

  • mobile telephony (e.g. for channel allocation or ad-hoc networks)
  • network technology (e.g. for data packet routing)
  • elevator control (e.g. for adaptive elevator dispatching)
  • energy and oil distribution (e.g. for electric power generation networks or for optimizing pipeline operations)
  • computing power management (e.g. for load balancing across servers), for distributed computing or for constrained job dispatchingin mainframe systems
  • autonomous robots (e.g. for robotic soccer or for exploration of unknown territory) and computer games (e.g. for rst-person shooter games
  • rescue operations (e.g. for enabling units to decide independently which sites to search and secure
  • surveillance and security tasks (e.g. for distributed sensor networks)  or for patrolling task or for military application
  • the field of production planning and factory optimization, where machines may act independently with the goal of achieving maximal joint productivity

The sub-pages to this page highlight a selection of projects with focus on multi-agent systems, on implementing cooperative decision strategies and on learning in cooperative multi-agent environments.