A Comprehensive Survey on Multi-Agent Cooperative Decision-Making: Scenarios, Approaches, Challenges and Perspectives

TL;DR

The surveyed work targets the problem of coordinating multiple agents to solve complex, dynamic tasks across real-world domains. It analyzes mainstream approaches, with a focus on MARL and LLMs-based decision-making, and provides a comprehensive taxonomy of methods (CTDE/DTDE/CTCE, rule-based, game-theoretic, evolutionary, and LLM-driven paradigms) alongside major simulation environments and benchmarks. The paper highlights key challenges—such as non-stationarity, scalability, multi-agent credit assignment, and hallucinations in language-model reasoning—while outlining future directions that include theoretical integration of LLMs with MARL, multi-modal and multi-task optimization, and ethical considerations. By detailing simulation platforms, practical implementations, and diverse applications (autonomous driving, UAVs, disaster response, and robotics), the survey aims to bridge theory and practice and stimulate reproducible progress in multi-agent cooperative decision-making.

Abstract

With the rapid development of artificial intelligence, intelligent decision-making techniques have gradually surpassed human levels in various human-machine competitions, especially in complex multi-agent cooperative task scenarios. Multi-agent cooperative decision-making involves multiple agents working together to complete established tasks and achieve specific objectives. These techniques are widely applicable in real-world scenarios such as autonomous driving, drone navigation, disaster rescue, and simulated military confrontations. This paper begins with a comprehensive survey of the leading simulation environments and platforms used for multi-agent cooperative decision-making. Specifically, we provide an in-depth analysis for these simulation environments from various perspectives, including task formats, reward allocation, and the underlying technologies employed. Subsequently, we provide a comprehensive overview of the mainstream intelligent decision-making approaches, algorithms and models for multi-agent systems (MAS). Theseapproaches can be broadly categorized into five types: rule-based (primarily fuzzy logic), game theory-based, evolutionary algorithms-based, deep multi-agent reinforcement learning (MARL)-based, and large language models(LLMs)reasoning-based. Given the significant advantages of MARL andLLMs-baseddecision-making methods over the traditional rule, game theory, and evolutionary algorithms, this paper focuses on these multi-agent methods utilizing MARL and LLMs-based techniques. We provide an in-depth discussion of these approaches, highlighting their methodology taxonomies, advantages, and drawbacks. Further, several prominent research directions in the future and potential challenges of multi-agent cooperative decision-making are also detailed.

Figures (25)

• Figure 1: An overview of the evolution of scenarios and methods in decision-making from single-agent to multi-agent systems.
• Figure 2: Illustration of our systematic review of multi-agent intelligent decision-making research. Compared to previous reviews, we have incorporated comprehensive introduction and analysis, with each segment corresponding to a specific chapter in the survey.
• Figure 3: The paradigms visualization of CTDE (left), DTDE (centre), and CTCE (right), consisting of three crucial elements: agent (i.e., algorithm or model), environment, central controller (Optional).
• Figure 4: A schematic representation of three distinct communication methods among agents, with arrows indicating the direction of message transmission. (a) Broadcasting communication: The activated agent transmits messages to all other agents within the communication network. (b) Targeted communication: Agents selectively communicate with specific target agents based on a supervisory mechanism that regulates the timing, content, and recipients of the messages. (c) Networked communication: Agents engage in localized interactions with their neighboring agents within the network.
• Figure 5: Typical Scenarios in Multi-Agent Particle Environment (MPE).