Asynchronous Agent Teams for Collaborative Tasks Based on Bottom-Up Alliance Formation and Adaptive Behavioral Strategies
Masashi Hayano, Naoki Iijima, Toshiharu Sugawara · 2017
This paper proposes a method to efficiently form teams for tasks that can be executed by multiple agents with different capabilities in a distributed network environment. Recent growing information and networking technologies have been realizing new types of computerized services that have been achieved by appropriately combining data from networked sensing devices and actuators controlled by intelligent programs in decentralized environments. Because these types of services can be realized by a team of agents acting using their own capabilities, how such teams can be formed effectively and efficiently in a distributed environment in a bottom-up manner is a key issue for autonomic computing. Our proposed method can autonomously recognize the dependable agents based on past successful cooperative behaviors, and they generate a tight alliance structure to execute the given tasks. Such an alliance structure avoids some conflicts by preventing many tasks being allocated to a few capable agents. We experimentally show that the proposed method can stably exhibit good performance and can adapt to environmental changes where task structure varies.