M. Saitoh and Y. Oyama (Japan)
Cooperative Artificial Intelligence Systems; Autonomous Agents; Reinforcement Learning; Division of Labor
In learning agent team, giving appropriate amount of re ward to each agent is necessary for division of labor. In this paper we present a novel approach for a reward allo cation in which reward payments are conducted between agents. We consider foraging task in small maze as a test problem. Firstly, we investigate an algorithm in which ex changes of (fixed amount of) reward for a food are made between agents. The experimental results show that our approach can produce territorial division of labor and its performance is significantly improved than a global rein forcement approach in which all the agents obtain equally divided reward. Secondly, several extended algorithms in which agents can determine the price of a food on their own are investigated, it is shown that minimal ”negotiations” be tween agents are effective for suitable price determination and good performances of teams.
Important Links:
Go Back
