|InterJournal Complex Systems, 1082
|Manuscript Number: |
Submission Date: 2004
|Social Foraging Theory for Multiagent Decision-Making System Design|
Foraging theory is used to model animal decision making from an evolutionary perspective. Optimality models have considered a wide range of factors including animal objectives, social context, and effects of cognitive and environmental constraints. Some models have been tested experimentally while others have provided general insights and unifying principles. Here, we develop an analogy where we view an agent (e.g., vehicle or software module) as an animal, a communication network as implementing inter-agent sociality, and the domain of operation as a foraging environment. We explain how to view tasks and objectives of agents as individual and group objectives. We then apply social foraging theory to a particular system, where autonomous vehicles connected via a communication network perform cooperative search. We first define the engineering design process for multiagent systems within an evolutionary game-theoretic framework. We explain why sociality between vehicular agents may emerge due to enhanced searching capabilities, group-environment interactions, or increased protection from predation or failure. Then we use an "evolutionarily stable strategy" approach to make predictions about optimal group size for vehicles solving the cooperative search problem. Our predictions are evaluated in an autonomous vehicle simulation testbed. Our overall goal is to gain better insights into how to design controllers for groups of agents that are optimized for a given set of objectives and domain of operation, and yet have properties of “robustness” often seen in nature. Ultimately, we hope engineering design principles for networked multiagent systems will have a reciprocal impact on evolutionary studies of cooperation in animals.
|Submit referee report/comment|