Due to the rapid advancement in artificial intelligence technology, intelligent decision-making technology has gradually outperformed humans in many human versus machine contests, particularly in complex multi-agent collaborative task environments. In multi-agent collaboration decision-making, several agents cooperate to accomplish pre-defined tasks and realize certain goals. This technology can be applied in real-life scenarios like autonomous vehicles, drone navigation, disaster relief operations, and military confrontations simulations. This paper starts with a detailed review of the major simulation environments and platforms for multi-agent collaboration decision-making. We make a detailed analysis on the following aspects of these simulation environments: task format, reward distribution, and the technological base. Finally, we make an overall review of the intelligent decision-making methods and algorithms for multi-agent systems (MAS). They can generally be divided into five categories: rule-based (mainly fuzzy logic), game theory-based, evolutionary algorithm-based, deep MARL-based, and LLMs reasoning-based approaches. Considering that the MARL and LLMs-based decision-making approaches have a considerable edge over the conventional approaches like rule, game theory, and evolutionary algorithms, this paper aims to explore the multi-agent approaches based on MARL and LLMs. We offer a comprehensive review of such approaches, along with their methodologies, pros, and cons. Moreover, some future research directions related to multi-agent cooperative decision-making are also discussed.

Keywords:

Intelligent decision-making, Multi-agent systems, Multi-agent cooperative environments, Multi-agent reinforcement learning, Large language models.