Models as a Key Factor of Environments Design in Multi-Agent Reinforcement Learning

This paper presents a generalized graphical outline of the current state of the art in the development of Markov decision processes (MDPs). This systematization opens ways to further improve the field of reinforcement learning (RL), by creating RL algorithms based on new or modified existing MDP models. The paper contains an overview of RL environments and an experimental result of using different environments for multi-agent reinforcement learning. The experimental research of MDP models was carried out in standard digital environments of machine learning: MuJoCo and SMAC.