Overview
An artificial agent society is a collection of software programs, referred to as agents, that are interconnected for the purpose of completing a task. Agents may possess different skills and resources, and tasks require a set of skills and resources for completion. Organizational adaptation in an artificial agent society involves a collection of agents having the ability to change their interconnections for the purposes of forming a team to complete a task. This project aims to develop decentralized methods for individual agents to change their connectivity with other agents in order to form stable teams within a networked agent society.
Active Projects
- Relational Push-Pull Models: Relational Push-Pull Models are used for relational data clustering but we plan to apply the model to multi-agent systems for analyzing team formations.
- Multi-Agent Team Formation Testbed: We are developing a testbed using the Repast toolkit.
Applications
We have identified three interesting application domains to which this research will apply:
- Mobile Robot Exploration: Imagine robots exploring the surface of Mars. Robots may have different mobility, sensors, software routines, and communication capabilities. Different collections of robots are needed to perform varying experiments designed by scientists on Earth. Robots may or may not be able to communicate based on range, bandwidth and power limitations. The network structure can be changed by requesting a particular robot to move to a ` new lodation.
- Distributed Vehicle Monitoring: Distributed vehicle monitoring (DVM) systems consist of a collection of simple sensor devices that relay information to identify moving objects within a bounded area. Sensors in these systems are very small and have substantial resource limitations, most notably power limitations, since most sensors have a single non-rechargable battery. To minimize power consumption, other abilities are limited as well, such as communication bandwidth, sensor range, and memory. The ability to form minimal teams and network structures to accomplish tasks would be useful for sensor systems because of the limited resources. Another use of team formation and network adaptation is fault tolerance, where a sensor that is malfunctioning will be removed from teams or be given a diminished role when the fault is detected.
- Supply Chain Management: Agent systems in this problem area represent different roles in the process of producing and distributing consumable goods, such as manufacturers, distrubutors, wholesalers, retailers and consumers. These different agents are connected together to form a supply chain that runs from manufacturer to consumer. Additionally, agents can connect at the same level (e.g. manufacturers connect to other manufacturers) to represent a cooperative agreement (e.g. two manufacturers working together to fill a large order). Agents have varying rates of production/consumption, and no single agent may be able to fulfill a requirement, hence the need for team formation.
Research Directions
Team Formation: Research in this area involves developing methods for finding a suitable team to complete a set of tasks, given a network structure.
Network Adaptation: Research in this area involves the adaptation of the network to facilitate the formation of teams.
Long Term Coalition Formation: Much research in team formation involves one-shot tasks. Giving the agents the ability to form long-term alliances provides them with the benefits of stability, trust and security.
This project is funded by the NSF CAREER program.