Multi-Agent Reinforcement Learning is an extension of single-agent RL to problems with multiple interacting agents. It is challenging due to non-stationary environments and credit assignment across agents. Baseline methods like Independent Q-Learning treat other agents as part of the environment. Cooperation methods use centralized critics and decentralized actors. Zero-sum methods were applied to StarCraft. General-sum methods like LOLA learn opponent models to optimize strategies. Experience replay and communication protocols help agents learn cooperative behaviors.