基于大语言模型的任务规划智能体框架、技术与挑战

doi:10.11896/jsjkx.260100066

摘要/Abstract

摘要： 任务规划作为人工智能的核心研究内容之一,传统方法难以应对具有长时性、复杂逻辑关系及动态变化任务规划的实际需求。近期大型语言模型(LLM)展现出的强大上下文推理与类人认知能力,为解决上述难题提供了崭新途径。然而,直接将大语言模型应用于任务规划仍面临长期记忆缺失、复杂任务分解能力有限、动态变化响应不及时等瓶颈。针对这些挑战,提出了一种基于大语言模型的任务规划智能体框架。该框架在AI Agent结构基础上,以大语言模型为中心,通过抽象推理、动作驱动、反思、记忆等模块的有机协同,实现了高效的动态任务规划。深入研究了该框架中所涉及的相关关键技术现状,对比分析了不同方法的优势与不足,进行了技术权衡,并分析了框架在面对复杂度、资源消耗以及实时性不同情况下的技术选择。通过总结当前研究面临的挑战和困难,提出了未来研究方向,为基于大语言模型的任务规划提供了理论框架和方法指导。

关键词: 任务规划, 大模型, 微调, 抽象推理, 动作驱动, 反思, 记忆

Abstract: As a core research domain in artificial intelligence,task planning faces challenges with traditional methods struggling to address practical demands involving long-term planning,complex logical relationships,and dynamic task adjustments.Recent breakthroughs in large language models(LLM) have demonstrated remarkable contextual reasoning and human-like cognitive capabilities,offering novel solutions to these challenges.However,direct application of LLM to task planning still encounters bottlenecks including long-term memory deficits,limited task decomposition capacity,and delayed responses to dynamic changes.To tackle these issues,this paper proposes an intelligent agent framework for task planning based on large language models.Building upon the AI Agent architecture,the framework centers on LLM and achieves efficient dynamic task planning through the organic integration of modules such as abstract reasoning,action-driven mechanisms,reflection,and memory systems.This study thoroughly examines the current state of key technologies involved,compares the strengths and limitations of different approaches,and evaluates technical trade-offs.It also analyzes how the framework selects appropriate technologies under varying conditions of complexity,resource consumption,and real-time requirements.By summarizing existing research challenges and difficulties,the paper outlines future research directions,providing a theoretical framework and methodological guidance for task planning based on large language models.

Key words: Task planning, Large language model, Fine-tuning, Abstract reasoning, Action-driven, Reflection, Memory

中图分类号:

TP391

安冉, 程恺, 邵天浩, 李昌原, 陈艳. 基于大语言模型的任务规划智能体框架、技术与挑战[J]. 计算机科学, 2026, 53(9): 24-37. https://doi.org/10.11896/jsjkx.260100066

AN Ran, CHENG Kai, SHAO Tianhao, LI Changyuan, CHEN Yan. Framework,Techniques and Challenges of Task Planning Agent Based on Large Language Model[J]. Computer Science, 2026, 53(9): 24-37. https://doi.org/10.11896/jsjkx.260100066

参考文献

[1] MA J,GAO J Q,JIANG G D,et al.The application of proximal policy optimization algorithm in mission planning[J].Cyber Security And Data Governance,2025,44(S1):446-449.
[2] WANG W,ZHAO K N,YANG L,et al.Hierarchical task planning approach for complex scenarios[J].Systems Engineering and Electronics,2025,47(4):1255-1264.
[3] WANG T,ZHAO M J,XU P,et al.Decision Technology Based on Large Language Model for Wargame[J].Acta Automatica Sinica,2025,51(6):1205-1217.
[4] ZHENG Y Z,CUI X L.Research on the technology of integrating chain of thought with chain of command and decision[J].Control and Decision,2026,41(2):432-444.
[5] WANG L,MA C,FENG X Y,et al.A survey on large language model based autonomous agents[J].Frontiers of Computer Science,2024,18(6):186345.
[6] WU X N,LI R X,WANG L,et al.Coordination Framework for Collaborative Disposal of Multi-intelligent Agents Based on Large Language Models [J].Journal of Data Acquisition and Processing,2024,39(3):559-576.
[7] QIN L,WU W S,LIU D,et al.Autonomous Planning and Pro-cessing Framework for Complex Tasks based on Large Language Models[J].Acta Automatica Sinica,2024,50(4):862-872.
[8] WANG F,CAO J X,JIAO R Y,et al.LLM-based intelligentagent framework for USV path planning in informationscarce environments[J／OL].Chinese Journal of Ship Research,DOI:10.19693/j.issn.1673-3185.04503.
[9] WU X K,CHEN M,LI W,et al.Llm fine-tuning:Concepts,opportunities,and challenges[J].Big Data and Cognitive Computing,2025,9(4):87.
[10] DEVLIN J,CHANG M W,LEE K,et al.Bert:Pre-training of deep bidirectional transformers for language understanding[C] //Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics:Human Language Technologies.Minneapolis:Association for Computational Linguistics,2019:4171-4186.
[11] LI X L,LIANG P.Prefix-tuning:Optimizing continuousprompts for generation [EB/OL].(2021-01-01)[2025-11-16].https://arxiv.org/abs/2101.00190.
[12] HU E J,SHEN Y,WALLIS P,et al.LoRA:Low-Rank Adaptation of Large Language Models[EB/OL].(2021-07-17)[2025-11-16].https://api.semanticscholar.org/CorpusID:235458009.
[13] GULCEHRE C,PAINE T L,SRINIVASAN S,et al.Reinforced self-training(rest) for language modeling[EB/OL].(2023-08-17) [2025-11-16].https://arxiv.org/pdf/2308.08998.
[14] RAMAMURTHY R,AMMANABROLU P,BRANTLEY K,et al.Is reinforcement learning(not) for natural language processing:Benchmarks,baselines,and building blocks for natural language policy optimization[EB/OL].(2022-10-03)[2025-11-16].https://arxiv.org/pdf/2210.01241.
[15] SCHULMAN J,WOLSKI F,DHARIWAL P,et al.Proximalpolicy optimization algorithms[EB/OL].(2017-07-20)[2025-11-16].https://arxiv.org/pdf/1707.06347.
[16] YUAN H,YUAN Z,TAN C,et al.Rrhf:Rank responses toalign language models with human feedback[J].Advances in Neural Information Processing Systems,2023,36:10935-10950.
[17] RAFAILOV R,SHARMA A,MITCHELL E,et al.Direct pre-ference optimization:Your language model is secretly a reward model[J].Advances in Neural Information Processing Systems,2023,36:53728-53741.
[18] MOHAMMADI B,ABBASNEJAD E,QI Y,et al.Parameter-Efficient Action Planning with Large Language Models for Vision-and-Language Navigation[J].Pattern Recognition,2026,172:112462.
[19] LI H,JIAO Z,LIU X,et al.In-situ Value-aligned Human-Robot Interactions with Physical Constraints[EB/OL].(2025-08-11)[2025-11-16].https://arxiv.org/pdf/2508.07606.
[20] WANG R,LI H,HAN X,et al.Learning from failure:Integrating negative examples when fine-tuning large language models as agents[EB/OL].(2024-02-18)[2025-11-16].https://arxiv.org/pdf/2402.11651.
[21] CHEN Z,LIU K,WANG Q,et al.Agent-flan:Designing data and methods of effective agent tuning for large languagemodels[EB/OL].(2024-03-19)[2025-11-16].https://arxiv.org/pdf/2403.12881.
[22] ZHANG J,LAN T,MURTHY R,et al.Agentohana:Designunified data and training pipeline for effective agentlearning [EB/OL].(2024-02-23) [2025-11-16].https://arxiv.org/pdf/2402.15506.
[23] LEHNERT L,SUKHBAATAR S,SU D J,et al.Beyond A^*:Better planning with transformers via search dynamics bootstrapping[EB/OL].(2024-02-21)[2025-11-16].https://arxiv.org/pdf/2402.14083.
[24] WEI J,WANG X,SCHUURMANS D,et al.Chain-of-thoughtprompting elicits reasoning in large language models[J].Advances in Neural Information Processing Systems,2022,35:24824-24837.
[25] KOJIMA T,GU S S,REID M,et al.Large language models are zero-shot reasoners[J].Advances in Neural Information Processing Systems,2022,35:22199-22213.
[26] SHEN Y,SONG K,TAN X,et al.Hugginggpt:Solving ai tasks with chatgpt and its friends in hugging face[J].Advances in Neural Information Processing Systems,2023,36:38154-38180.
[27] TUSHAR K,HARSH T,MATTHEW F,et al.Decomposedprompting:A modular approach for solving complex tasks[C] //Eleventh International Conference on Learning Representations(ICLR).Kigali:OpenReview.net,2023.
[28] YAO S,YU D,ZHAO J,et al.Tree of thoughts:Deliberate problem solving with large language models[EB/OL].(2023-05-17) [2025-11-16].https://arxiv.org/abs/2305.10601.
[29] BESTA M,BLACH N,KUBICEK A,et al.Graph of thoughts:Solving elaborate problems with large language models[C] //Proceedings of the AAAI Conference on Artificial Intelligence.Washington,DC:AAAI Press,2024:17682-17690.
[30] WANG L,XU W,LAN Y,et al.Plan-and-solve prompting:Improving zero-shot chain-of-thought reasoning by large language models[EB/OL].(2023-05-06)[2025-11-16].https://arxiv.org/pdf/ 2305.04091.
[31] WANG X,WEI J,SCHUURMANS D,et al.Self-consistencyimproves chain of thought reasoning in language models[EB/OL].(2022-03-21)[2025-11-16].https://arxiv.org/pdf/2203.11171.
[32] PRASAD A,KOLLER A,HARTMANN M,et al.Adapt:As-needed decomposition and planning with language models[C] //Findings of the Association for Computational Linguistics(NAACL 2024).Association for Computational Linguistics,2024:4226-4252.
[33] GUI R,WANG Z,WANG J,et al.HyperTree Planning:Enhancing LLM Reasoning via Hierarchical Thinking[C] //Procee-dings of the Forty-second International Conference on Machine Learning.Vienna:PMLR,2025:46506.
[34] WU J,ANTONOVA R,KAN A,et al.Tidybot:Personalized robot assistance with large language models[J].Autonomous Robots,2023,47(8):1087-1102.
[35] SUN S,LIU Y,WANG S,et al.Pearl:Prompting large language models to plan and execute actions over long documents[C] //Proceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics.Association for Computational Linguistics,2024:469-486.
[36] SINGH I,BLUKIS V,MOUSAVIAN A,et al.ProgPrompt:Generating Situated Robot Task Plans using Large Language Models[C] //2023 IEEE International Conference on Robotics and Automation.IEEE Press,2023:11523-11530.
[37] HAZRA R,DOS MARTIRES P Z,DE RAEDT L.Saycanpay:Heuristic planning with large language models using learnable domain knowledge[C] //Proceedings of the AAAI Conference on Artificial Intelligence.Washington,DC:AAAI Press,2024:20123-20133.
[38] LIU B,JIANG Y,ZHANG X,et al.Llm+ p:Empowering large language models with optimal planning proficiency[EB/OL].(2023-04-22)[2025-11-16].https://arxiv.org/pdf/2304.11477.
[39] DAGAN G,KELLER F,LASCARIDES A.Dynamic Planningwith a LLM[C] //Proceedings of the Language Gamification Workshop 2024 at NeurIPS.Vancouver:NeurIPS,2024:1-14.
[40] CHALVATZAKI G,YOUNES A,NANDHA D,et al.Learning to reason over scene graphs:a case study of finetuning gpt-2 into a robot language model for grounded task planning[J].Frontiers in Robotics and AI,2023,10:1221739.
[41] DZIRI N,LU X,SCLAR M,et al.Faith and fate:Limits oftransformers on compositionality[J].Advances in Neural Information Processing Systems,2023,36:70293-70332.
[42] HAO Y,ZHANG Y,FAN C.Planning Anything with Rigor:General-Purpose Zero-Shot Planning with LLM-based Forma-lized Programming[C] //Proceedings of the Thirteenth International Conference on Learning Representations.ICLR,2025:731.
[43] JU D,JIANG S,COHEN A,et al.To the Globe(TTG):Towards Language-Driven Guaranteed Travel Planning[C] //Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing:System Demonstrations.Association for Computational Linguistics,2024:240-249.
[44] CHEN W,MA X,WANG X,et al.Program of ThoughtsPrompting:Disentangling Computation from Reasoning for Numerical Reasoning Tasks[J].Transactions on Machine Lear-ning Research,2023,10.
[45] ZHUANG Y,CHEN X,YU T,et al.ToolChain*:Efficient Action Space Navigation in Large Language Models with A^* Search[C] //Proceedings of the Twelfth International Confe-rence on Learning Representations.Vienna:OpenReview.net,2024:49-57.
[46] HUANG W,XIA F,XIAO T,et al.Inner Monologue:Embodied Reasoning through Planning with Language Models[C] //Conference on Robot Learning.Atlanta:PMLR,2023:1769-1782.
[47] YAO S Y,ZHAO J,YU D,et al.React:Synergizing reasoning and acting in language models[C] //Proceedings of the Eleventh International Conference on Learning Representations.Kigali:OpenReview.net,2023:10-17.
[48] SONG C H,WU J,WASHINGTON C,et al.Llm-planner:Few-shot grounded planning for embodied agents with large language models[C] //Proceedings of the IEEE/CVF International Conference on Computer Vision.IEEE Press,2023:2986-2997.
[49] SUN H,ZHUANG Y,KONG L,et al.Adaplanner:Adaptiveplanning from feedback with language models[J].Advances in Neural Information Processing Systems,2023,36:58202-58245.
[50] GUAN L,VALMEEKAM K,SREEDHARAN S,et al.Leveraging pre-trained large language models to construct and utilize world models for model-based task planning[J].Advances in Neural Information Processing Systems,2023,36:79081-79094.
[51] MA Y J,LIANG W,WANG G,et al.Eureka:Human-Level Reward Design via Coding Large Language Models[C] //Procee-dings of the Twelfth International Conference on Learning Representations.Vienna:OpenReview.net,2024:47-57.
[52] RAMAN S S,COHEN V,ROSEN E,et al.Planning with large language models via corrective re-prompting[C] //Proceedings of the NeurIPS 2022 Foundation Models for Decision Making Workshop.New Orleans:NeurIPS Foundation,2022:1-21.
[53] RAMAN S S,COHEN V,IDREES I,et al.Cape:Corrective actions from precondition errors using large language models[C] //Proceedings of the 2024 IEEE International Conference on Robotics and Automation(ICRA).IEEE Press,2024:14070-14077.
[54] WANG Z,CAI S,CHEN G,et al.Describe,explain,plan and select:interactive planning with large language models enables open-world multi-task agents[C] //Proceedings of the 37th International Conference on Neural Information Processing Systems.Red Hook:Curran Associates,Inc.,2023:34153-34189.
[55] ZHA L,CUI Y,LIN L H,et al.Distilling and retrieving genera-lizable knowledge for robot manipulation via language corrections[C] //2024 IEEE International Conference on Robotics and Automation(ICRA).IEEE Press,2024:15172-15179.
[56] SONG Y,YIN D,YUE X,et al.Trial and Error:Exploration-Based Trajectory Optimization of LLM Agents[C] //Procee-dings of the 62nd Annual Meeting of the Association for Computational Linguistics.Bangkok:Association for Computational Linguistics,2024:7584-7600.
[57] YANG J,PRABHAKAR A,NARASIMHAN K,et al.Inter-code:Standardizing and benchmarking interactive coding with execution feedback[J].Advances in Neural Information Proces-sing Systems,2023,36:23826-23854.
[58] XI Z,CHEN W,GUO X,et al.The Rise and Potential of Large Language Model Based Agents:A Survey[J].Science China Information Sciences,2025,68(2):121101.
[59] PACKER C,FANG V,PATIL S G,et al.MemGPT:Towards LLMs as Operating Systems[EB／OL].(2023-11-15)[2025-11-16].https://arxiv.org/abs/2311.08719.
[60] LIU L,YANG X,SHEN Y,et al.Think-in-memory:Recalling and post-thinking enable llms with long-term memory[EB／OL].(2023-11-15)[2025-11-16].https://arxiv.org/abs/2311.08719.
[61] SCHUURMANS D.Memory augmented large language models are computationally universal[EB／OL].(2023-01-10)[2025-11-16].https://arxiv.org/abs/2301.04589.
[62] LU J,AN S,LIN M,et al.Memochat:Tuning llms to use memos for consistent long-range open-domainconversation[EB/OL].(2023-08-16)[2025-11-16].https://arxiv.org/abs/2308.08239.
[63] PARK J S,O’BRIEN J,CAI C J,et al.Generative agents:Interactive simulacra of human behavior[C] //Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology.New York:Association for Computing Machinery,2023:1-22.
[64] SHINN N,CASSANO F,GOPINATH A,et al.Reflexion:Language agents with verbal reinforcement learning[J].Advances in Neural Information Processing Systems,2023,36:8634-8652.
[65] ZHANG D,CHEN L,ZHANG S,et al.Large language models are semi-parametric reinforcement learning agents[J].Advances in Neural Information Processing Systems,2023,36:78227-78239.
[66] WANG Y,JIANG Z,CHEN Z,et al.Recmind:Large language model powered agent for recommendation[C] //Findings of the Association for Computational Linguistics:NAACL 2025.Albuquerque:Association for Computational Linguistics.2025:4351-4364.
[67] WANG G,XIE Y,JIANG Y,et al.Voyager:An Open-Ended Embodied Agent with Large Language Models[EB/OL].(2023-05-25)[2026-05-30].https://arxiv.org/abs/2305.16291.
[68] SHI H,XIE B,LIU Y,et al.Memoryvla:Perceptual-cognitive memory in vision-language-action models for robotic manipulation[EB/OL].(2025-08-26)[2025-11-16].https://arxiv.org/abs/2508.19236.
[69] JANNY S,POIRIER H,ANTSFELD L,et al.Reasoning in vi-sual navigation of end-to-end trained agents:a dynamical systems approach [C] //Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition(CVPR).IEEE Press,2025:12111-12121.
[70] NAVIGLI R,CONIA S,ROSS B.Biases in large language mo-dels:Origins,inventory,and discussion[J].Journal of Data and Information Quality,2023,15(2):10.
[71] GALLEGOS I O,ROSSI R A,BARROW J,et al.Biasand fairness in large language models:A survey[J].Computational Linguistics,2024,50(3):1097-1179.
[72] ZHANG Y,LI Y,CUI L,et al.Siren’s Song in the AI Ocean:A Survey on Hallucination in Large Language Models[J].Computational Linguistics,2025,51:1-46.
[73] YUAN L,LIU S H,WANG Y,et al.A Comprehensive Review of Optimization Techniques in Retrieval-Augmented Generation for Large Language Models[J].Chinese Journal of Computers,2026,49(2):383-422.
[74] WEI T,ZHONG Z Y,LIU Y,et al.High-order Program Driven By Large Language Model[J].Journal of Information Security Research,2025,11(11):34-45.
[75] XU B,PENG Z,LEI B,et al.ReWOO:Decoupling Reasoningfrom Observations for Efficient Augmented Language Models[EB/OL].(2023-05-23)[2026-05-29].https://arxiv.org/abs/2305.18323.
[76] YUAN G,XIAO J,HE J,et al.Multi-agent cooperative areacoverage:a two-stage planning approach based on reinforcement learning[J].Information Sciences,2024,678:121025.
[77] LI A,XIE Y,LI S,et al.Agent-Oriented Planning in Multi-Agent Systems[EB/OL].(2024-10-03)[2025-11-16].https://arxiv.org/abs/2410.02189.
[78] LIU J,ZHOU P,DU Y,et al.CaPo:Cooperative Plan Optimization for Efficient Embodied Multi-Agent Cooperation[EB/OL].(2024-11-07)[2025-11-16].https://arxiv.org/abs/2411.04679.
[79] WEI H,ZHANG Z,HE S,et al.PlanGenLLMs:A Modern Survey of LLM Planning Capabilities[C] //Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics(ACL 2025).Vienna:Association for Computational Linguistics,2025:19497-19521.
[80] YANG D S,LI Q,LIU Y C,et al.Macro-scale C2 Process Mecha-nism:Revision and Application of Theory Model of PREA Loop[J].Journal of Command and Control,2022,8(4):389-401.
[81] JI X,JIANG K,CHENG H D,et al.Architecture and Core Pillars of All-domain Operation Command Information System[J].Journal of Command and Control,2023,9(2):225-232.
[82] MCCARTHY S M.Hierarchical Planning in Security Games:A Game Theoretic Approach to Strategic, Tactical and Operatio-nal Decision Making[D].Los Angeles:Universityof Southern California,2018.
[83] LI Z,WANG K,WANG B,et al.Human-machine fusion intelligent decision-making:concept,framework and application[J].Journal of Electronics & Information Technology,2025,47(10):3439-3464.
[84] ZOU H P,HUANG W C,WU Y,et al.LLM-Based Human-Agent Collaboration and Interaction Systems:A Survey[EB/OL].(2025-05-01)[2025-11-16].https://arxiv.org/abs/2505.00753.
[85] REN M L,CHEN N Y,QIU H.Human-machine collaborative decision-making:An evolutionary roadmap based on cognitive intelligence[J].International Journal of Social Robotics,2023,15(7):1101 -1114.
[86] LIU Q D,HE W X,YAO E G,et al.Reinforcement Learningfrom X-Feedback:Recent Advances and FutureProspects[J].Chinese Journal of Computers,2026,49(3):497-519.
[87] YIN H B,TIAN D X.World models in autonomous driving:a review and outlook [J].Journal of Harbin Institute of Techno-logy,2025,57(12):165-178.
[88] XIONG L,FENG H J,ZHANG P Z, et al.Research Progress of Learning Generation Method for Autonomous Vehicle Interaction Scene[J].Journal of South China University of Technology(Natural Science Edition),2026,54(3):31-51.
[89] TIAN D X,XIAO X,ZHOU J S,et al.A Review of AI-driven Trajectory Prediction Methodsfor Autonomous Vehicles [J].Journal of Transportation Systems Engineering and Information Technology,2025,25(5):1-24.
[90] LU Y,WANG J,GUO L,et al.R-Horizon:How Far Can Your Large Reasoning Model Really Go in Breadth andDepth?[EB/OL].(2025-10-09)[2025-11-16].https://arxiv.org/pdf/2510.08189.
[91] CHENG Z,FAN R,HAO S,et al.K2-think:A parameter-efficient reasoning system [EB/OL].(2025-09-09)[2025-11-16].https://arxiv.org/abs/2509.07604.
[92] CHO K,VAN MERRINBOER B,BAHDANAU D,et al.On the Properties of Neural Machine Translation:Encoder-Decoder Approaches[C] //Proceedings of SSST-8,Eighth Workshop on Syntax,Semantics and Structure in Statistical Translation.Doha:Association for Computational Linguistics,2014:103-111.
[93] VASWANI A,SHAZEER N,PARMAR N,et al.Attention isall you need[J].Advances in Neural Information Processing Systems,2017,30:5998-6008.
[94] GU A,DAO T.Mamba:Linear-time sequence modeling with selective state spaces[EB/OL].(2023-12-01)[2026-05-30].https://arxiv.org/abs/2312.00752.
[95] MA C,NAJARIAN K.Rethinking the long-range dependency in Mamba/SSM and transformer models[EB/OL].(2025-09-04)[2025-11-16].https://arxiv.org/pdf/2509.04226.
[96] GUO T H,LI X H,YIN Q H,et al.Survey of the application of large language model in robot task planning[J].Journal of Tianjin University of Technology and Education,2025,35(1):12-18.
[97] WANG J X.Research and application of multiagent collaborative framework for complex task reasoning[D].Beijing:Beijing University of Chemical Technology,2025.
[98] SUN Z B,HU S,ZHANG F,et al.Improved YOLOv7-tiny with long short range dependency feature pyramid network[J].Journal of Image and Graphics,2025,30(8):2775-2789.

Metrics

Viewed

Full text

Abstract

Cited

Shared

Discussed