图谱赋能AI Agent：探索下一代智能体的发展前景与挑战 原创

摘要

人工智能代理经历了从强化学习（RL）早期主导地位到由大型语言模型（LLMs）驱动的智能代理崛起的范式转变，现在正进一步迈向RL和LLM能力的协同融合。这一进步赋予了人工智能代理越来越强大的能力。尽管取得了这些进展，但要完成复杂的现实世界任务，代理需要有效规划和执行，保持可靠的记忆，并与其他代理顺畅协调。实现这些能力涉及应对始终存在的复杂信息、操作和互动。鉴于这一挑战，数据结构化可以通过将复杂无序的数据转换为结构良好的形式，使代理能够更有效地理解和处理，从而发挥出有希望的作用。在此背景下，图谱在组织、管理和利用复杂数据关系方面具有天然优势，为支持高级人工智能代理所需的能力提供了强大的数据结构化范式。为此，本调查首次系统地回顾了图谱如何赋能人工智能代理。具体来说，我们探讨了图形技术与核心智能体功能的整合，重点介绍了显著的应用案例，并识别了未来研究的潜在途径。通过全面审视这一迅速发展的交叉领域，我们希望激发下一代AI智能体的发展，使其能够应对日益复杂的图谱挑战。相关资源已在GitHub链接中收集并持续更新供社区参考。

Graphs Meet AI Agents: Taxonomy, Progress, and Future Opportunities

​​https://github.com/YuanchenBei/Awesome-Graphs-Meet-Agents​​

核心速览

研究背景

1. 研究问题：这篇文章要解决的问题是如何通过图技术增强AI代理的功能，包括规划、执行、记忆和多智能体协调。具体来说，图技术如何帮助AI代理更有效地处理复杂任务中的信息、操作和交互。
2. 研究难点：该问题的研究难点包括：处理复杂任务中的非结构化数据、在多智能体系统中实现有效的信息传递和协调、以及在动态环境中维护和更新记忆。
3. 相关工作：该问题的研究相关工作包括强化学习（RL）和大语言模型（LLM）在AI代理中的应用，以及图技术在数据组织和知识提取方面的成功应用。

研究方法

这篇论文提出了通过图技术来增强AI代理功能的方法。具体来说，

1. 图在代理规划中的应用：图可以用于组织任务推理形式、安排任务分解过程，并构建高效的任务决策搜索过程。

• 任务推理：使用知识图谱辅助推理和结构组织推理来增强LLM代理的任务理解能力。
• 任务分解：构建任务依赖图（TDG）来表示子任务的依赖关系，并优化执行路径。
• 任务决策搜索：使用状态空间图（SSG）来表示状态之间的转换，并通过搜索算法（如蒙特卡罗树搜索）来优化决策过程。

2.图在代理执行中的应用：图可以帮助组织工具的使用和环境交互。

• 工具使用：构建工具图来清晰地显示工具之间的连接，使代理能够高效地使用和管理大量工具。
• 环境交互：使用场景图来编码视觉场景中的对象及其空间或语义关系，并采用启发式方法和学习方法来建模这些关系。

3.图在代理记忆中的应用：图结构记忆可以有效地揭示各种信息之间的潜在关联。

• 记忆组织：将知识和经验存储为相互连接的表示，使用知识图谱和其他结构化形式来组织长期记忆。
• 记忆检索：使用图检索增强生成（graph-based RAG）来准确高效地检索有用信息。
• 记忆维护：动态更新和细化记忆表示和图拓扑，以响应新的经验和交互。

4.图在多智能体协调中的应用：图可以用于建模智能体之间的通信路径和任务分配。

• 任务特定关系：使用任务依赖图和任务分配图来优化任务执行和信息交换。
• 环境特定关系：根据特定环境的特征动态学习智能体之间的关系权重。
• 协调拓扑优化：通过边重要性测量、图自编码器优化和强化学习来优化多智能体系统的通信拓扑。

结果与分析

1. 科学计算：图学习结合代理系统在科学计算领域表现出显著潜力，特别是在自动化科学发现和生物信息学分析方面。
2. 具身AI：图基表示在具身AI中提供了强大的工具，增强了场景理解能力，并支持更明智的决策。
3. 游戏AI：图结构在表示和建模游戏AI中越来越受到关注，特别是在多智能体游戏和文本游戏中。
4. 代理信息检索：图学习可以支持结构化的检索规划和自适应推理，提高复杂任务中的信息检索效率。
5. 工业和自动化系统：图组织和学习可以在工业系统中实现高效的性能，增强系统的可扩展性、动态演化和鲁棒性。
6. 人类社会：LLM代理在分析和模拟人类社会行为方面显示出巨大潜力，特别是在社交网络和信息传播方面。

总体结论

这篇论文全面系统地回顾了图技术与AI代理的交叉领域，探讨了图技术如何增强代理的规划、执行、记忆和多智能体协调功能。此外，还研究了代理范式如何反过来增强图学习。基于详细的回顾，论文总结了有意义的应用、开放问题和未来的研究方向，为下一代面对日益复杂和混乱任务信息的代理提供了新的潜在方法。相关资源已在Github链接中组织并持续更新。

论文评价

优点与创新

1. 系统综述：本文提供了图技术与AI代理交叉领域的第一个全面系统综述，涵盖了从基于强化学习（RL）到基于大型语言模型（LLM）的代理范式。
2. 新颖的分类视角：论文引入了一种新颖的分类视角，探讨了图如何增强代理的核心功能：规划、执行、记忆和多代理协调。
3. 双向创新：论文不仅讨论了图如何增强代理功能，还探讨了代理如何反过来推动图学习的进步，强调了双向创新和整体视角。
4. 应用和未来机会：基于综述，论文进一步讨论了图增强AI代理的有意义应用、关键挑战和未来机会。
5. 资源更新：相关资源在Github链接中收集并持续更新，为研究和工业社区提供了宝贵的参考资料。

不足与反思

1. 基准评估：现有的基准测试在任务定义或评估数据上存在差异，使得统一评估变得具有挑战性。此外，针对多代理推理、大规模动态环境中的记忆和协作等新兴复杂场景，缺乏以图为中心的代理基准。
2. 图基础模型：尽管有许多基于图的图学习方法，但在代理功能中广泛使用的图操作符仍然缺乏。开发有效的图基础模型（GFMs）是一个有前景的方向，特别是从效果、可解释性和可扩展性（EES）的角度设计GFMs。
3. 隐私和安全：图组织和学习在建立实体之间的连接时可能会带来安全问题，包括数据隐私和攻击防御。未来的研究应专注于设计更安全的代理协调和环境交互策略。
4. 多模态代理：尽管LLM代理在语言空间中取得了显著进展，但多模态代理（能够理解和整合文本、视觉和语音等信息流的代理）的研究仍在不断发展。图学习可以在多模态数据的抽象和连接中发挥重要作用。
5. 模型上下文协议（MCP）：MCP协议提供了一种标准化手段，用于无缝集成代理应用和外部数据源、工具和服务。未来的研究可以探索如何利用图学习增强MCP的两个有前景的方面：高效的数据集成和个性化推荐。
6. 开放代理网络（OAN）：一个开放的代理网络（OAN）将是一个公共、去中心化的网络，其中代理被注册和编排。未来的研究可以探索如何在OAN中实现图学习，以提高代理网络的效率和风险控制能力。

关键问题及回答

问题1：图技术在AI代理的规划功能中具体有哪些应用？

1. 任务推理：使用知识图谱辅助推理和结构组织推理来增强LLM代理的任务理解能力。例如，QA-GNN通过结合语言模型和辅助知识图谱来提高问答系统的性能，而ToG和KG-CoT则利用基于知识图谱的检索增强生成（RAG）来提升LLM代理的任务推理能力。
2. 任务分解：构建任务依赖图（TDG）来表示子任务的依赖关系，并优化执行路径。TDG是一种有向无环图（DAG），用于表示任务之间的依赖关系，从而帮助代理识别和执行有效的任务子序列。
3. 任务决策搜索：使用状态空间图（SSG）来表示状态之间的转换，并通过搜索算法（如蒙特卡罗树搜索）来优化决策过程。SSG通过将状态和状态之间的转换表示为图的节点和边，帮助代理在复杂环境中进行有效的决策。

问题2：图技术在AI代理的执行功能中有哪些具体应用？

1. 工具使用：构建工具图来清晰地显示工具之间的连接，使代理能够高效地使用和管理大量工具。例如，GPTSwarm通过将代理抽象为有向图（DAG），每个节点代表一个函数，边表示信息流，从而帮助代理高效地调用和管理工具。
2. 环境交互：使用场景图来编码视觉场景中的对象及其空间或语义关系，并采用启发式方法和学习方法来建模这些关系。场景图通过将对象及其关系表示为图的节点和边，帮助代理更好地理解和交互复杂的环境。

问题3：图技术在AI代理的记忆功能中有哪些具体应用？

1. 记忆组织：将知识和经验存储为相互连接的表示，使用知识图谱和其他结构化形式来组织长期记忆。例如，AriGraph将代理的记忆表示为结构化的知识图谱，包含语义事实和事件，从而帮助代理回忆复杂的结构和关系。
2. 记忆检索：使用图检索增强生成（graph-based RAG）来准确高效地检索有用信息。例如，G-Retriever和GFM-RAG通过结合语义相似性和图度量来设计定制的检索器，从而提高信息检索的准确性和效率。
3. 记忆维护：动态更新和细化记忆表示和图拓扑，以响应新的经验和交互。例如，A-MEM通过动态索引和链接创建互联的知识网络，从而允许代理在不断变化的环境中持续改进其记忆表示。

🚀 Taxonomy

Graph for Agent Planning

Task Reasoning

Knowledge Graph-Auxiliary Reasoning

• (NAACL 2021) QA-GNN: Reasoning with language models and knowledge graphs for question answering [Paper] [Code]
• (ICLR 2024) Think-on-graph: Deep and responsible reasoning of large language model on knowledge graph [Paper] [Code]
• (ICLR 2024) Reasoning on graphs: Faithful and interpretable large language model reasoning [Paper] [Code]
• (IJCAI 2024) Kg-cot: Chain-of-thought prompting of large language models over knowledge graphs for knowledge-aware question answering [Paper]
• (ACL 2024) Mindmap: Knowledge graph prompting sparks graph of thoughts in large language models [Paper] [Code]
• (WWW 2025) Paths-over-graph: Knowledge graph empowered large language model reasoning [Paper]

Structure-Organized Reasoning

• (NeurIPS 2023) Tree of thoughts: Deliberate problem solving with large language models [Paper] [Code]
• (NAACL 2024) GoT: Effective Graph-of-Thought Reasoning in Language Models [Paper] [Code]
• (AAAI 2024) Graph of thoughts: Solving elaborate problems with large language models [Paper] [Code]
• (AAAI 2025) Ratt: A thought structure for coherent and correct llm reasoning [Paper] [Code]
• (Arxiv 2025) Reasoning with Graphs: Structuring Implicit Knowledge to Enhance LLMs Reasoning [Paper]

Task Decomposition

• (CoLM 2024) Agentkit: Structured LLM reasoning with dynamic graphs [Paper] [Code]
• (TMLR 2024) Feudal Graph Reinforcement Learning [Paper]
• (NeurIPS 2024) Can Graph Learning Improve Planning in LLM-based Agents? [Paper] [Code]
• (ACL 2024) Villageragent: A graph-based multi-agent framework for coordinating complex task dependencies in minecraft [Paper] [Code]
• (Arxiv 2024) DAG-Plan: Generating Directed Acyclic Dependency Graphs for Dual-Arm Cooperative Planning [Paper] [Demo]
• (ICRA 2025) Enhancing Multi-Agent Systems via Reinforcement Learning with LLM-based Planner and Graph-based Policy [Paper]
• (Arxiv 2025) DynTaskMAS: A Dynamic Task Graph-driven Framework for Asynchronous and Parallel LLM-based Multi-Agent Systems [Paper]
• (Arxiv 2025) Plan-over-Graph: Towards Parallelable LLM Agent Schedule [Paper] [Code]

Task Decision Searching

• (NeurIPS 2014) Deep Learning for Real-Time Atari Game Play Using Offline Monte-Carlo Tree Search Planning [Paper]
• (AAAI 2018, Best Paper) Memory-augmented monte carlo tree search [Paper]
• (ACML 2020) Monte-Carlo Graph Search: the Value of Merging Similar States [Paper]
• (ICAPS 2021) Improving alphazero using monte-carlo graph search [Paper]
• (AI 2024) Evolving interpretable decision trees for reinforcement learning [Paper]
• (AAMAS 2024) Continuous monte carlo graph search [Paper] [Code]
• (ICLR 2024) Promptagent: Strategic planning with language models enables expert-level prompt optimization [Paper] [Code]
• (ICML 2024) Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models [Paper] [Code]

Graph for Agent Execution

Tool Usage

• (ACL 2024) Graph Chain-of-Thought: Augmenting Large Language Models by Reasoning on Graphs [Paper] [Code]
• (ECCV 2024) ControlLLM: Augment Language Models with Tools by Searching on Graphs [Paper] [Code]
• (ICML 2024) GPTSwarm: Language Agents as Optimizable Graphs [Paper] [Code]
• (Arxiv 2024) ToolNet: Connecting Large Language Models with Massive Tools via Tool Graph [Paper]
• (Arxiv 2025) Graph RAG-Tool Fusion [Paper] [Code]

Environment Interaction

Heuristic-Based Relationship

• (ICRA 2023) Multiagent Reinforcement Learning for Autonomous Routing and Pickup Problem with Adaptation to Variable Demand [Paper]
• (COR 2024) A Graph Reinforcement Learning Framework for Neural Adaptive Large Neighbourhood Search [Paper]
• (AAMAS 2024) Towards Generalizability of Multi-Agent Reinforcement Learning in Graphs with Recurrent Message Passing [Paper] [Code]
• (RAL 2024) Language-Grounded Dynamic Scene Graphs for Interactive Object Search with Mobile Manipulation [Paper] [Code]
• (Arxiv 2024) PlanAgent: A Multi-modal Large Language Agent for Closed-loop Vehicle Motion Planning [Paper]
• (CVPR 2025) GUI-Xplore: Empowering Generalizable GUI Agents with One Exploration [Paper] [Code]
• (Arxiv 2025) Multi-agent Auto-Bidding with Latent Graph Diffusion Models [Paper]
• (Arxiv 2025) A Schema-Guided Reason-while-Retrieve framework for Reasoning on Scene Graphs with Large-Language-Models (LLMs) [Paper]

Learning-Based Relationship

• (NeurIPS Workshop 2018) Deep Multi-Agent Reinforcement Learning with Relevance Graphs [Paper] [Code]
• (CoRL 2023) Learning Control Admissibility Models with Graph Neural Networks for Multi-Agent Navigation [Paper]
• (AAMAS 2023) TransfQMix: Transformers for Leveraging the Graph Structure of Multi-Agent Reinforcement Learning Problems [Paper] [Code]
• (TAI 2024) Reinforcement Learned Multi–Agent Cooperative Navigation in Hybrid Environment with Relational Graph Learning [Paper]
• (NCA 2024) Graph network-based human movement prediction for socially-aware robot navigation in shared workspaces [Paper]

Graph for Agent Memory

Memory Organization

• (JMS 2021) Towards Self-X Cognitive Manufacturing Network: An Industrial Knowledge Graph-Based Multi-Agent Reinforcement Learning Approach [Paper]
• (Arxiv 2024) AriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM Agents [Paper]
• (Arxiv 2024) On the Structural Memory of LLM Agents [Paper] [Code]
• (Arxiv 2024) From Local to Global: A GraphRAG Approach to Query-Focused Summarization [Paper] [Code]
• (Arxiv 2024) KG-Retriever: Efficient Knowledge Indexing for Retrieval-Augmented Large Language Models [Paper] [Code]
• (SIGIR 2025) Enhancing the Patent Matching Capability of Large Language Models via the Memory Graph [Paper] [Code]
• (AAAI 2025) LLM-Powered Decentralized Generative Agents with Adaptive Hierarchical Knowledge Graph for Cooperative Planning [Paper] [Code]
• (WWW 2025) Graphusion: A RAG Framework for Scientific Knowledge Graph Construction with a Global Perspective [Paper] [Code]
• (Arxiv 2025) Agentic Reasoning: Reasoning LLMs with Tools for the Deep Research [Paper] [Code]

Memory Retrieval

• (NeurIPS 2024) G-Retriever: Retrieval-Augmented Generation for Textual Graph Understanding and Question Answering [Paper] [Code]
• (Arxiv 2024) LightRAG: Simple and Fast Retrieval-Augmented Generation [Paper] [Code]
• (ICLR 2025) Simple is Effective: The Roles of Graphs and Large Language Models in Knowledge-Graph-Based Retrieval-Augmented Generation [Paper] [Code]
• (NAACL 2025) GRAG: Graph Retrieval-Augmented Generation [Paper] [Code]
• (Arxiv 2025) GFM-RAG: Graph Foundation Model for Retrieval Augmented Generation [Paper] [Code]
• (Arxiv 2025) PathRAG: Pruning Graph-based Retrieval Augmented Generation with Relational Paths [Paper] [Code]

Memory Maintenance

• (NeurIPS 2024) HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models [Paper] [Code]
• (Arxiv 2024) LightRAG: Simple and Fast Retrieval-Augmented Generation [Paper] [Code]
• (Arxiv 2024) KG-Agent: An Efficient Autonomous Agent Framework for Complex Reasoning over Knowledge Graph [Paper]
• (Arxiv 2024) AriGraph: Learning Knowledge Graph World Models with Episodic Memory for LLM Agents [Paper] [Code]
• (AAAI 2025) LLM-Powered Decentralized Generative Agents with Adaptive Hierarchical Knowledge Graph for Cooperative Planning [Paper] [Code]
• (Arxiv 2025) Zep: A Temporal Knowledge Graph Architecture for Agent Memory [Paper] [Code]
• (Arxiv 2025) A-Mem: Agentic Memory for LLM Agents [Paper] [Code]
• (Arxiv 2025) InstructRAG: Leveraging Retrieval-Augmented Generation on Instruction Graphs for LLM-Based Task Planning [Paper]

Graphs for Multi-Agent Coordination

Coordination Message Passing

Task-Specific Relationship

• (NeurIPS 2022) Learning NP-Hard Multi-Agent Assignment Planning using GNN: Inference on a Random Graph and Provable Auction-Fitted Q-learning [Paper]
• (ICRA 2025) Enhancing Multi-Agent Systems via Reinforcement Learning with LLM-based Planner and Graph-based Policy [Paper]
• (ICLR 2025) Scaling Large Language Model-based Multi-Agent Collaboration [Paper] [Code]
• (Arxiv 2025) DynTaskMAS: A Dynamic Task Graph-driven Framework for Asynchronous and Parallel LLM-based Multi-Agent Systems [Paper]
• (Arxiv 2025) MAGNNET: Multi-Agent Graph Neural Network-based Efficient Task Allocation for Autonomous Vehicles with Deep Reinforcement Learning [Paper]
• (Arxiv 2025) GNNs as Predictors of Agentic Workflow Performances [Paper] [Code]

Environment-Specific Relationship

• (ICASSP 2021) Graphcomm: A Graph Neural Network Based Method for Multi-Agent Reinforcement Learning [Paper]
• (ITSC 2022) Graph Convolution-Based Deep Reinforcement Learning for Multi-Agent Decision-Making in Interactive Traffic Scenarios [Paper]
• (TITS 2022) Multi-Agent Trajectory Prediction with Heterogeneous Edge-Enhanced Graph Attention Network [Paper]
• (TCCN 2023) MAGNNETO: A Graph Neural Network-based Multi-Agent system for Traffic Engineering [Paper] [Code]
• (TPAMI 2023) Robust Multi-Agent Communication With Graph Information Bottleneck Optimization [Paper]
• (IROS 2024) Transformer-based Multi-Agent Reinforcement Learning for Generalization of Heterogeneous Multi-Robot Cooperation [Paper]
• (ICLR 2025) Exponential Topology-enabled Scalable Communication in Multi-agent Reinforcement Learning [Paper] [Code]

Coordination Topology Optimization

• (AAAI 2020) Multi-Agent Game Abstraction via Graph Attention Neural Network [Paper]
• (AAMAS 2021) Deep Implicit Coordination Graphs for Multi-agent Reinforcement Learning [Paper] [Code]
• (AAMAS 2021) Multi-Agent Graph-Attention Communication and Teaming [Paper]
• (NeurIPS 2021) Learning Distilled Collaboration Graph for Multi-Agent Perception [Paper] [Code]
• (TNNLS 2022) Online Multi-Agent Forecasting with Interpretable Collaborative Graph Neural Networks [Paper]
• (AAMAS 2023) Learning Graph-Enhanced Commander-Executor for Multi-Agent Navigation [Paper] [Code]
• (ICLR 2024) Learning Multi-Agent Communication from Graph Modeling Perspective [Paper] [Code]
• (Arxiv 2024) G-Designer: Architecting Multi-agent Communication Topologies via Graph Neural Networks [Paper] [Code]
• (COLM 2024) A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration [Paper] [Code]
• (ICML 2024) GPTSwarm: Language Agents as Optimizable Graphs [Paper] [Code]
• (JCISE 2025) Adaptive Network Intervention for Complex Systems: A Hierarchical Graph Reinforcement Learning Approach [Paper]
• (ICRA 2025) Reliable and Efficient Multi-Agent Coordination via Graph Neural Network Variational Autoencoders [Paper] [Code]
• (ICLR 2025) Cut the Crap: An Economical Communication Pipeline for LLM-based Multi-Agent Systems [Paper] [Code]
• (Arxiv 2025) Deep Meta Coordination Graphs for Multi-agent Reinforcement Learning [Paper] [Code]
• (Arxiv 2025) Adaptive Graph Pruning for Multi-Agent Communication [Paper]

Agents for Graph Learning

Graph Annotation and Synthesis

• (NeurIPS 2020) Graph Policy Network for Transferable Active Learning on Graphs [Paper] [Code]
• (AAAI 2022) Batch Active Learning with Graph Neural Networks via Multi-Agent Deep Reinforcement Learning [Paper]
• (Arxiv 2024) Exploring the Potential of Large Language Models in Graph Generation [Paper]
• (Arxiv 2024) LLM-Based Multi-Agent Systems are Scalable Graph Generative Models [Paper] [Code]
• (ICLR Workshop 2025) IGDA: Interactive Graph Discovery through Large Language Model Agents [Paper]
• (Arxiv 2025) Plan-over-Graph: Towards Parallelable LLM Agent Schedule [Paper] [Code]
• (Arxiv 2025) GraphMaster: Automated Graph Synthesis via LLM Agents in Data-Limited Environments [Paper]

Graph Understanding

• (KDD 2020) Policy-GNN: Aggregation Optimization for Graph Neural Networks [Paper]
• (WWW 2021) SUGAR: Subgraph Neural Network with Reinforcement Pooling and Self-Supervised Mutual Information Mechanism [Paper] [Code]
• (NeurIPS 2023) MAG-GNN: Reinforcement Learning Boosted Graph Neural Network [Paper] [Code]
• (ICLR 2023) Agent-based Graph Neural Networks [Paper] [Code]
• (Arxiv 2023) Graph Agent: Explicit Reasoning Agent for Graphs [Paper]
• (Arxiv 2023) A Versatile Graph Learning Approach through LLM-based Agent [Paper]
• (KDD 2024) GraphWiz: An Instruction-Following Language Model for Graph Computational Problems [Paper] [Code]
• (SIGIR 2024) GraphGPT: Graph Instruction Tuning for Large Language Models [Paper] [Code]
• (ICLR 2024) One For All: Towards Training One Graph Model For All Classification Tasks [Paper] [Code]
• (ICML 2024) LLaGA: Large Language and Graph Assistant [Paper] [Code]
• (KDD 2024) ZeroG: Investigating Cross-dataset Zero-shot Transferability in Graphs [Paper] [Code]
• (Arxiv 2024) GraphAgent: Agentic Graph Language Assistant [Paper] [Code]
• (Arxiv 2024) Scalable and Accurate Graph Reasoning with LLM-based Multi-Agents [Paper]
• (Arxiv 2024) GraphTeam: Facilitating Large Language Model-based Graph Reasoning via Multi-Agent Collaboration [Paper]
• (Arxiv 2024) GraphInstruct: Empowering Large Language Models with Graph Understanding and Reasoning Capability [Paper] [Code]
• (AAAI 2025) Graph Agent Network: Empowering Nodes with Inference Capabilities for Adversarial Resilience [Paper]

💻 Benchmarks and Open-Source Toolkits

General

• (NeurIPS 2021, RL Agent, Multi-Agent Coordination) Benchmarking Multi-Agent Deep Reinforcement Learning Algorithms in Cooperative Tasks [Paper] [Code]
• (JMLR 2024, RL Agent, Multi-Agent Coordination) BenchMARL: Benchmarking Multi-Agent Reinforcement Learning [Paper] [Code]
• (JMLR 2025, RL Agent, Agent Memory) Memory Gym: Towards Endless Tasks to Benchmark Memory Capabilities of Agents [Paper] [Code]
• (EMNLP 2023, LLM Agent, Tool Usage) API-Bank: A Comprehensive Benchmark for Tool-Augmented LLMs [Paper] [Code]
• (NeurIPS 2023, LLM Agent, Task Reasoning, Task Decomposition) PlanBench: An Extensible Benchmark for Evaluating Large Language Models on Planning and Reasoning about Change [Paper] [Code]
• (ICLR 2024, LLM Agent, General) AgentBench: Evaluating LLMs as Agents [Paper] [Code]
• (ICLR 2024, LLM Agent, Tool Usage) ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs [Paper] [Code]
• (NeurIPS 2024, LLM Agent, Tool Usage) GTA: A Benchmark for General Tool Agents [Paper] [Code]
• (ICML 2024, LLM Agent, Task Reasoning, Task Decomposition, Tool Usage) TravelPlanner: A Benchmark for Real-World Planning with Language Agents [Paper] [Code]
• (ICLR 2025, LLM Agent, Tool Usage, Agent-Environment Interaction) τ-bench: A Benchmark for Tool-Agent-User Interaction in Real-World Domains [Paper] [Code]
• (NAACL 2025, LLM Agent, Tool Usage, Agent-Environment Interaction) ToolSandbox: A Stateful, Conversational, Interactive Evaluation Benchmark for LLM Tool Use Capabilities [Paper] [Code]
• (Arxiv 2025, LLM Agent, Task Reasoning, Task Decomposition, Multi-Agent Coordination) REALM-Bench: A Real-World Planning Benchmark for LLMs and Multi-Agent Systems [Paper] [Code]

Graph-Related

• (LLM Agent, Tool Usage, Multi-Agent Coordination) LangGraph [Docs] [Code]
• (NeurIPS 2024, LLM Agent, Graph Modeling) GLBench: A Comprehensive Benchmark for Graph with Large Language Models [Paper] [Code]
• (NeurIPS 2024, LLM Agent, Graph Modeling) Can Large Language Models Analyze Graphs like Professionals? A Benchmark, Datasets and Models [Paper] [Code]
• (ICLR 2025, LLM Agent, Graph Modeling) GraphArena: Evaluating and Exploring Large Language Models on Graph Computation [Paper] [Code]
• (ICML 2024, LLM Agent, Task Reasoning, Tool Usage, Multi-Agent Coordination) GPTSwarm: Language Agents as Optimizable Graphs [Paper] [Code]
• (ICLR 2025, LLM Agent, Multi-Agent Coordinatio) Scaling Large Language Model-based Multi-Agent Collaboration [Paper] [Code]
• (Arxiv 2025, LLM Agent, Tool Usage) Graph RAG-Tool Fusion [Paper] [Code]
• (Arxiv 2025, LLM Agent, Task Reasoning, Task Decomposition) GNNs as Predictors of Agentic Workflow Performances [Paper] [Code]

本文转载自​​知识图谱科技​​，作者：Wolfgang