Architecture and Interface for Collaborating with a Group of Agents in an Adversarial Game
Natsuki Matsunami, Masayuki Itô, Sotaro Karakama · 2020
Despite the impressive advances in artificial intelligence (AI), close collaboration between people and AI systems is still hard to achieve. To overcome this problem, we have designed an architecture consisting of the Contract Net Protocol, the Behavior Tree, and a human–multiple agent interface that we call "human-swarm interaction." Its effectiveness and feasibility were evaluated using team play in a simulated Tail Tag game. Matches with up to 19 AI agents and 1 person on one team and 20 people on the other team were played. The results indicate that our system is scalable and that there is room for improvement. We identified two key features for AI agents to achieve collaboration with people in a serious game as represented by Tail Tag game: the ability to at least remain alive in any situation, and active declaration of roles being attempted as well as sharing of roles.