A direct-indirect reward sharing model in multiagent reinforcement learning
Yoshinobu Bochi, Tadachika Ozono, Toramatsu Shintani · 2003
We propose a new multiagent model called direct-indirect reward sharing model. An evaluator in our model determines contributions of agents in a goal state, and shares a reward according to them. We have examined two methods for determining the ratios of contributions. One method uses a constant ratio that is derived the theory for suppressing irrational rules we prove in this paper. The other method evaluates them based on the evaluator's past memory.