Agent Environment Cycle Games
This addresses a conceptual mismatch in MARL modeling for software environments, potentially improving implementation accuracy and flexibility.
The authors argue that Partially Observable Stochastic Games (POSGs) are ill-suited for software multi-agent reinforcement learning (MARL) environments, leading to unexpected behaviors, and introduce Agent Environment Cycle Games (AEC Games) as a more representative model, proving its equivalence to POSGs and demonstrating its ability to elegantly represent all MARL environments, including turn-based games like chess.
Partially Observable Stochastic Games (POSGs) are the most general and common model of games used in Multi-Agent Reinforcement Learning (MARL). We argue that the POSG model is conceptually ill suited to software MARL environments, and offer case studies from the literature where this mismatch has led to severely unexpected behavior. In response to this, we introduce the Agent Environment Cycle Games (AEC Games) model, which is more representative of software implementation. We then prove it's as an equivalent model to POSGs. The AEC games model is also uniquely useful in that it can elegantly represent both all forms of MARL environments, whereas for example POSGs cannot elegantly represent strictly turn based games like chess.