[3P4-J-7] Agents: social multiagents

[3P4-J-7-02] An allocation strategy with deep reinforcement learning for efficient task processing in multi agent system

〇Genki Matsuno¹, Sho Tanaka², Hiroki Hara², Syunyo Kawamoto², Syo Shimoyama², Takashi Kawashima², Daisuke Tsumita², Yasushi Kido¹, Osamu Hashimoto¹, Tomohiro Takagi² (1. Skydisc, Inc., 2. Meiji University)

Keywords:multi-agent, task allocation, reinforcement learning

In this study, it was considered how to make matching between task required resource and member capability that will be applied into multi-agent systems.

Supported by Reinforcement Learning strategy with deep Neural Network technique, a modern solution was conducted accompanied with standard baseline methods and evaluated from several suggestive viewpoints.

According to the numerical experiments, it is elucidated that RL strategy has some advantages when targeting on both execution time duration and accuracy of combination matching.