Can anyone recommend a reinforcement learning algorithm for a multi-agent environment?

In my simplified example, I'm implementing a Q-Learning system with different 10 agents. The agents compete for resources in stores at different locations by setting a bid price for each item.

All of the agents have different bids and pooled budget of $100. Once the budget is reached the agents cannot buy any more that day.

Each agent will receive a reward if they buy an item. The goal would be to maximize the total amount of items bought between the agents.

Right now the agents don't communicate.

Can someone point me in the right direction for an algorithm that allows agent cooperation?

  • 1
    $\begingroup$ You seem to have elements of both competitive and cooperative behaviour required, making it hard to answer the question. If the goal is to "maximize the total amount of items" (i.e. agents don't win or lose, the group as a whole does), then why do agents "compete for resources"? $\endgroup$ Jun 21, 2018 at 8:55
  • 1
    $\begingroup$ If this is a simulation of something in the real world, cooperation is only limited by constraints in the simulated environment - otherwise you could re-frame the problem as a single agent issue. Those constraints will affect a good answer here, could you elaborate on what limits the agents face? What prevents them simply deferring to a single agent controller that makes the best decision for the group as a whole? $\endgroup$ Jun 21, 2018 at 8:55


You must log in to answer this question.

Browse other questions tagged .