Multi-Agent Model-Based Deep Reinforcement Learning: Testing the Method in Grid-World Environments

This paper investigates the identification and resolution of social dilemmas in multi-agent reinforcement learning. The model of multi-agent environment was developed to simulate agent movement within grid-world scenarios. The study focuses on the MiniGrid and MultiGrid environments and proposes a method for multi-agent model-based reinforcement learning in these settings. The method was tested in comparison with value-based Q-learning, Deep Q-Network (DQN), Independent Q-Network (IQN), and Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithms. Experimental results demonstrate the effectiveness of the proposed approach in multi-agent grid-world environments.