AI Agent 前沿论文精选综述(2023-2025)
更新时间:2026年5月9日
主题:LLM-based Autonomous Agents / 大语言模型自主智能体
📖 一、研究背景与概述
随着大语言模型(LLM)在广泛知识获取上的突破,基于 LLM 的自主智能体(Autonomous Agents)已成为 AI 领域最活跃的研究方向之一。[1] 与传统强化学习训练出的窄域智能体不同,LLM 智能体能够通过自然语言理解、推理和规划来完成复杂任务,体现了向人类智能靠拢的潜力。
🔬 二、核心研究方向
2.1 工具使用(Tool Use)
让 LLM 学会调用外部工具是扩展其能力的关键。经典工作包括:
- ReAct[2]:将推理(Reasoning)与行动(Acting)协同,ICLR 2023
- Toolformer[3]:让 LLM 自学使用工具,NeurIPS 2023
- HuggingGPT[4]:连接 ChatGPT 与 Hugging Face 生态,NeurIPS 2023
- ToolLLM[5]:掌握 16000+ 真实 API,EMNLP 2023
2.2 规划与推理(Planning)
规划能力是智能体的核心认知能力,主要方法包括:
| 方法 | 论文 | 会议/年份 | 核心思路 |
|---|---|---|---|
| 思维树 (ToT) | Tree of Thoughts[6] | NeurIPS 2023 | BFS/DFS 搜索 |
| MCTS | LATS[7] | ICML 2024 | 蒙特卡洛树搜索 |
| Least-to-Most | Least-to-Most[8] | NeurIPS 2023 | 复杂问题分解 |
| PDDL | LLM+P[9] | arXiv 2023 | 符号规划结合 |
2.3 反馈学习(Feedback Learning)
通过自我反思与外部反馈提升推理质量:
- Reflexion[10]:语言智能体的口头强化学习,NeurIPS 2023
- Self-Refine[11]:迭代式自我改进,NeurIPS 2023
- CRITIC[12]:工具交互式批判,ICLR 2024
2.4 多智能体系统(Multi-Agent)
多个 LLM 智能体协作解决问题:
- ChatDev:虚拟软件公司,多智能体协作开发
- Multi-Agent Debate:多智能体辩论提升推理
- AgentVerse:多智能体协作框架
📊 三、主流综述论文
| 论文标题 | 作者 | 发表 | 链接 |
|---|---|---|---|
| A Survey on Large Language Model based Autonomous Agents | Wang et al. | Frontiers of Computer Science 2024 | [arXiv] [DOI] |
| The Rise and Potential of LLMs as Agent: A Survey | Xi et al. | arXiv 2023 | [arXiv] |
| A Review of Prominent Paradigms for LLM-Based Agents | Li X. | CoLing 2025 | [arXiv] |
| Understanding the Planning of LLM Agents: A Survey | Huang et al. | arXiv 2024 | [arXiv] |
| AgentAI: A Comprehensive Survey on Autonomous Agents | Chen et al. | Expert Systems 2025 | [ScienceDirect] |
| A Survey on the Memory Mechanism of LLM Agents | - | arXiv 2024 | [arXiv] |
| LLM and AI Agents for Autonomous Systems: A Survey | Ferrag et al. | Semantic Scholar | [Semantic Scholar] |
🛠️ 四、关键框架与工具
- LangChain/LangGraph:主流 LLM 应用开发框架
- AutoGPT:自主任务执行智能体
- AutoGen:微软多智能体协作框架
- CrewAI:角色扮演多智能体框架
- MetaGPT:软件开发的 SOTA 多智能体框架
📈 五、市场与技术趋势
根据最新市场分析[13]:
- 2024 年全球 AI Agent 市场规模约 54.3 亿美元
- 预计 2030 年将达到 471 亿美元
- 年复合增长率(CAGR)超过 43%
🚀 六、未来研究方向
- 自主演化智能体(Self-Evolving Agents):智能体能够在线学习、自我改进
- 世界模型(World Modeling):让 LLM 建立环境模拟能力
- 长期记忆机制:解决智能体上下文遗忘问题
- 多模态融合:结合视觉、语言、动作的全身智能体
- 具身智能(Embodied AI):机器人与智能体的结合
📚 参考文献
- Wang L., Ma C., Feng X. et al. "A Survey on Large Language Model based Autonomous Agents." Frontiers of Computer Science, 2024. https://arxiv.org/abs/2308.11432 | https://doi.org/10.1007/s11704-024-40231-1
- Yao S., Zhao J., Yu D. et al. "ReAct: Synergizing Reasoning and Acting in Language Models." ICLR, 2023. https://arxiv.org/abs/2210.03629
- Schick T., Dwivedi-Zippert S., Tsatsulis E. et al. "Toolformer: Language Models Can Teach Themselves to Use Tools." NeurIPS, 2023. https://arxiv.org/abs/2302.04761
- Shen Y., Song K., Tan X. et al. "HuggingGPT: Solving AI Tasks with ChatGPT and its Friends in Hugging Face." NeurIPS, 2023. https://arxiv.org/abs/2303.17580
- Qin Y., Liang S., Ye Y. et al. "ToolLLM: Facilitating Large Language Models to Master 16000+ Real-world APIs." EMNLP, 2023. https://arxiv.org/abs/2307.16789
- Yao S., Yu D., Zhao J. et al. "Tree of Thoughts: Deliberate Problem Solving with Large Language Models." NeurIPS, 2023. https://arxiv.org/abs/2305.10601
- Zhou A., Wang K., Ma Y. et al. "Language Agent Tree Search Unifies Reasoning, Acting, and Planning." ICML, 2024. https://arxiv.org/abs/2311.04463
- Zhou J., Zhou A., Zhao Z. et al. "Least-to-Most Prompting Enables Complex Reasoning in Large Language Models." NeurIPS, 2023. https://arxiv.org/abs/2305.10601
- Liu X., Yu H., Zhang X. et al. "LLM+P: Empowering Large Language Models with Optimal Planning Proficiency." arXiv, 2023. https://arxiv.org/abs/2304.11477
- Shinn N., Labash B., Gopinath A. "Reflexion: Language Agents with Verbal Reinforcement Learning." NeurIPS, 2023. https://arxiv.org/abs/2303.11366
- Madaan A., Tandon N., Gupta P. et al. "Self-Refine: Iterative Refinement with Self-Feedback." NeurIPS, 2023. https://arxiv.org/abs/2303.17651
- Gou Z., Shao L., Feng G. et al. "CRITIC: Large Language Models Can Self-Correct with Tool-Interactive Critiquing." ICLR, 2024. https://arxiv.org/abs/2303.08758
- Ma T., Chen X. "Emerging Trends in LLM Agent Development." SSRN, 2024. https://papers.ssrn.com/sol3/papers.cfm?abstract_id=6119146
本文整理自 GitHub: xinzhel/LLM-Agent-Survey 及 arXiv 公开论文,更多论文列表请访问:LLM-Agent-Survey