I'm trying to define the q-learning multiple agent reinforcement learning models but I'm not sure if it is correct or not. N the number of agents and $i \in N$ and the k is the time episode.

$Q^{i}_{k+1}(s_{k}, a^{i}_{k}) =(1-\alpha_k)Q^{i}_k(s_{k}, a^{i}_{k})+ \alpha_k[{r^i}_k(s_{k},\\ a^{1}_{k},...,a^{N}_{k}) + \beta\max_{b^i}Q^{i}_k(s_{k+1},b^i)]$

Any help will be appreciated.

  • 1
    $\begingroup$ What makes you think that Q-learning can be "naturally" generalised to the context of multiple agents? Have you read about it somewhere? If yes, it may be opportune to cite it in your question. I know there are multi-agent algorithms which are based on the "temporal difference" idea. You may want to look at these algorithms. Some of them may be the generalization you're looking for. $\endgroup$ – nbro Feb 11 at 11:23
  • $\begingroup$ @nbro see this mdpi.com/1424-8220/18/2/375/htm $\endgroup$ – i_th Feb 12 at 11:13

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Browse other questions tagged or ask your own question.