What is a Multi-Agent System

A Multi-Agent System (MAS) is a system that consists of multiple agents that communicate and interact with each other to achieve a certain goal(s). In the previous article, we looked at what intelligent agents are and what characteristics they have. In this article, we will go through what Multi-Agent Systems are and describe their characteristics.

What is a Multi-Agent System?

A Multi-Agent System (MAS) is a group of agents or humans and agents that interact with each other and the environment to achieve goals [1]. In such a system, it is assumed that the agents may not have full knowledge of both the environment and the internal state of other agents. 

Interaction between agents is an important feature that enables them to use knowledge of other agents and learn more about the environment in a compressed time period. This type of interaction may be cooperative or competitive. In a cooperative interaction, agents work with each other towards a common goal. The aim of this interaction is to enable agents to distribute and share their knowledge and use the intelligence and capabilities of each other to solve problems. In a competitive interaction, agents may compete to obtain individual resources and achieve individual goals. 

In some domains, interactions did not take place between agents where necessary tasks could be undertaken by a single independent agent. Consider a resource-gathering scenario where agents are required to explore an unknown area to gather resources or find specific targets. This problem could be solved using agents that worked independently or cooperatively. In the former case, each agent was able to search independently and find targets based on a given criterion for the characteristics of a resource or target. In this case, the agents did not need any mutual interaction, while in the latter case, agents could interact to notify other agents of the location of targets/resources and assist each other to perform the task quicker. 

Dudek et al. [2] argued that in some cases using MAS had advantages over using a single agent. For example in a real-world resource-gathering scenario, replacing a single, complex robot/agent with a swarm of simpler robots, had a number of advantages. Since more robots were utilized, the task could be decomposed into simpler tasks. This would then reduce the design complexity of the robot’s development. The system would be more economical, flexible, and scalable. The overall rate of failure would also decrease because if a few robots failed to perform their task(s), the rest of the system would still be able to achieve the interactive goal(s) with the other robots. 

Agents collaborate with each other when they have positive motivations and recognize the need for cooperative actions [3]. The need for cooperation is recognized by an agent when it realizes that its goal cannot be achieved without another agent’s resources. It would search for other agents that could assist in accomplishing its goal and form a group with those that have the required abilities. Agents would then collaborate and negotiate with each other within the team to plan and take necessary actions to solve the problem. 

Multi-agent System Characteristics 

According to Huhn and Singh [1] some of the factors for characterizing MASs include the environment, agents, and inter-agent interactions. Dudek et al. [2] state other factors to define characteristics of MASs including team size, reconfigurability, composition, and communication topology. When MASs are applied to real-world applications other factors are also considered, including communication range, bandwidth, and the agents’ individual capabilities such as sensing and processing abilities. 

Team size can range from a few to a large number of agents. It may also be considered as an infinite number of agents for simulation purposes. The complexity of MASs can increase rapidly with the number of agents, the interactions among them, or the complexity of their behaviour. Team composition depends on the characteristics of individual agents and whether they are ‘homogeneous’ or ‘heterogeneous’ in physical and/or programming structure. Homogeneous agents are similar to each other or they are of the same type. Heterogeneous agents are different and diverse in kind. Reconfigurability refers to the ability of agents to reorganise and form a new subgroup or change their position and configuration. Factors of reconfigurability include the MAS environment and communication topologies. For example, if the environment were small, it would limit agent movement. Depending on communication topology, certain agents might be able to communicate only with agents that are linked to them.

Next, we look at Knowledge Representation and Communication Languages for agents. Continue here to read more about this topic.

References and Credits

• [1] L. Panait and S. Luke. Cooperative multi-agent learning: The state of the art. Autonomous Agents and Multi-Agent Systems, 11(3):387–434, 2005.
• [2] G. Dudek, M. Jenkin, E. Milios, and D. Wilkes. Taxonomy for swarm robots. In Proceedings of the 1993 IEEE/RSJ International Conference  on Intelligent Robots and Systems (IROS ’93), volume 1, pages 441–447. IEEE, Piscataway, NJ, USA, 1993.
• [3] M. Wooldridge. Verifying that agents implement a communication lan- guage. In Proceedings of the sixteenth national conference on Artificial in- telligence and the eleventh Innovative applications of artificial intelligence, AAAI ’99/IAAI ’99, pages 52 – 57, Orlando, Florida, United States, 1999. American Association for Artificial Intelligence.
• Feature Image by marian anbu juwan from Pixabay