A Comparison of Grouping Behaviors on Rule-Based and Learning-Based Multi-agent Systems
Grouping behavior, such as bird flocking, terrestrial animal herding, and fish schooling, is one of well-known emergent phenomena. Several models have been proposed for describing grouping behaviors, and two types of models can be defined: rule-based model and learning-based model. In rule-based models, each agent in a group has fixed interaction rules with respect to other agents. On the other hand, agents in learning-based model acquire their rules by the interactions of other agents with a learning scheme such as Q-learning. In this paper, we adopt quantities obtained from trails of agents, in order to investigate the properties for grouping behaviors of agents. We also evaluate rule-based and learning-based models by using these quantities under the environments with and without predatory agents.
KeywordsMulti-agent system Grouping behavior Q-learning Anisotropy
- Ballerini, M., Cabibbo, N., Candelier, R., Cavagna, A., Cisbani, E., Giardina, I., Lecomte, V., Orlandi, A., Parisi, G., Procaccini, A., Viale, M., Zdravkovic, V.: Interaction ruling animal collective behavior depends on topological rather than metric distance: evidence from a field study. Proc. National Acad. Sci. USA 105(4), 1232–1237 (2008)CrossRefGoogle Scholar
- Morihiro, K., Isokawa, T., Nishimura, H., Tomimasu, M., Kamiura, N., Matsui, N.: Reinforcement learning scheme for flocking behavior emergence. J. Adv. Comput. Intell. Intell. Inf. 11(2), 155–161 (2007)Google Scholar
- Morihiro, K., Nishimura, H., Isokawa, T., Matsui, N.: Grouping and anti-predator behaviors for multi-agent systems based on reinforcement learning scheme. In: Innovations in Multi-Agent Systems and Application-1, Chap. 6, pp. 149–183. Springer-Verlag, Berlin, Heidelberg (2010)Google Scholar