In most of the multi-agent reinforcement learning models I've found, it seems to generate the observations for each of the agents simultaneously and then uses a centralized critic to assess all of the agent's actions together.

However, what if two agents have a finite amount of resources to allocate, and the more one agent spends the less the other agent can spend. So really, the state space of the second agent is conditional on the action of the first agent.

Are there any papers or resources that describe an architecture like this?



You must log in to answer this question.

Browse other questions tagged .