Elamur Research⌕
☰
home / papers / 2026 / OntoGuard: Enforcing Action Admissibility for LLM Agents in Complex Interactive Environments

OntoGuard: Enforcing Action Admissibility for LLM Agents in Complex Interactive Environments

Paper · 2026 · Pengxiang Liu, Tao Ren, Wei Xiong et al.

Summary

LLM agents in interactive simulators often hallucinate. They attempt actions that are impossible in the current state or that break the required order of steps. Methods based on imitation or sparse-reward learning only fit the statistics of the data and never represent these rules explicitly. OntoGuard is a neuro-symbolic framework that separates world logic from planning and works in two stages:

  • Offline: it builds an environmental ontology and a behavioral ontology from oracle demonstrations, then completes them with LLM commonsense and ConceptNet. The result is two constraint graphs, G env and G beh.
  • Online: a verification loop checks each candidate action against both graphs before execution. If the action is invalid, the agent receives an interpretable explanation and is re-prompted to correct itself.

On ScienceWorld and VirtualHome, the approach matches or surpasses ReAct, Reflexion, SwiftSage and DGAP, especially in task success. It also stays robust when little expert data is available. This shows that explicit admissibility checking is a practical way to make agents more reliable in settings where mistakes cannot be undone.

Key points

  • Two rule types. Environmental constraints cover physical validity, such as whether an object exists or what it affords. Behavioral constraints cover task-specific temporal dependencies, such as not opening the bee hive before plants flower in the "Mendelian Genetics" task.
  • Expert-data ablation on ScienceWorld.
    • EXEC stays between 91.00 and 92.33 at every data proportion.
    • Removing expert data drops SR from 85.83 down to 57.33. Even the 0% setting still edges out zero-shot ReAct (56.75).
    • Average final score falls from 96.20 to 82.38, versus 36.17 for ReAct.
    • Long-horizon tasks lose the most when demonstrations are scarce.
  • Guard accuracy. On 400 balanced state-action pairs, the guard shows higher precision than an LLM Self-Check baseline, at the cost of slightly lower recall. In other words, it occasionally blocks valid actions.
  • VirtualHome.
    • In-Distribution: OntoGuard reaches 94.83 EXEC and 89.50 SR, versus 93.33 and 88.00 for DGAP.
    • The two novel settings: it has the best SR in both (64.50 and 74.50), though SwiftSage has higher EXEC in one of them.
  • Cost and limitation. Token cost per action is comparable to ReAct. The rules are static and limited by the diversity of the expert traces, and the authors leave online rule updating to future work.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🏛️Chinese Academy of SciencesBeijing · China194 works · 23,906 citations🎓University of Chinese Academy of SciencesBeijing · China147 works · 8,480 citations🔬Institute of SoftwareBeijing · China21 works · 4,980 citations

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

Cite

Cite as Pengxiang Liu, Tao Ren, Wei Xiong, Tingrui Yang, Junjie Wang and Jun HU. 2026. OntoGuard: Enforcing Action Admissibility for LLM Agents in Complex Interactive Environments. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1051

@inproceedings{liu2026ontoguard,
  title = {{OntoGuard: Enforcing Action Admissibility for LLM Agents in Complex Interactive Environments}},
  author = {Pengxiang Liu, Tao Ren, Wei Xiong, Tingrui Yang, Junjie Wang and Jun HU},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1051},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1051}
}

References

30 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Jacob Abernethy, Alekh Agarwal, Teodor Vanislavov Marinov, and Manfred K. Warmuth. 2024. A mechanism for sample-efficient in-context learning for sparse retrieval tasks. In Proceedings of The 35th International Conference on Algorithmic Learning Theory, volume 237 of Proceedings of Machine Learning… find
  2. Prithviraj Ammanabrolu and Matthew Hausknecht. 2020. Graph constrained reinforcement learning for natural language action spaces. In International Conference on Learning Representations. find
  3. Lawrence W Barsalou. 2008. Grounded cognition. Annu. Rev. Psychol., 59(1):617–645. brian ichter, Anthony Brohan, Yevgen Chebotar, Chelsea Finn, Karol Hausman, Alexander Herzog, Daniel Ho, Julian Ibarz, Alex Irpan, Eric Jang, Ryan Julian, Dmitry Kalashnikov, Sergey Levine, Yao Lu, Carolina Parada, Ka… find
  4. Lili Chen, Kevin Lu, Aravind Rajeswaran, Kimin Lee, Aditya Grover, Misha Laskin, Pieter Abbeel, Aravind Srinivas, and Igor Mordatch. 2021. Decision transformer: Reinforcement learning via sequence modeling. Advances in neural information processing systems, 34:15084–15097. find
  5. Shibo Hao, Yi Gu, Haodi Ma, Joshua Jiahua Hong, Zhen Wang, Daisy Zhe Wang, and Zhiting Hu. 2023. Reasoning with language model is planning with world model. In The 2023 Conference on Empirical Methods in Natural Language Processing. find
  6. Ji He, Jianshu Chen, Xiaodong He, Jianfeng Gao, Lihong Li, Li Deng, and Mari Ostendorf. 2016. Deep reinforcement learning with a natural language action space. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1621–1630, Berlin,… find
  7. Wenlong Huang, Fei Xia, Ted Xiao, Harris Chan, Jacky Liang, Pete Florence, Andy Zeng, Jonathan Tompson, Igor Mordatch, Yevgen Chebotar, Pierre Sermanet, Tomas Jackson, Noah Brown, Linda Luu, Sergey Levine, Karol Hausman, and brian ichter. 2022. Inner monologue: Embodied reasoning through planning wi… find
  8. Nathan Lambert, Valentina Pyatkin, Jacob Morrison, LJ Miranda, Bill Yuchen Lin, Khyathi Chandu, Nouha Dziri, Sachin Kumar, Tom Zick, Yejin Choi, Noah A. Smith, and Hannaneh Hajishirzi. 2025. RewardBench: Evaluating reward models for language modeling. In Findings of the Association for Computational… find
  9. Shuang Li, Xavier Puig, Chris Paxton, Yilun Du, Clinton Wang, Linxi Fan, Tao Chen, De-An Huang, Ekin Akyürek, Anima Anandkumar, Jacob Andreas, Igor Mordatch, Antonio Torralba, and Yuke Zhu. 2022. Pre-trained language models for interactive decision-making. In Advances in Neural Information Processin… find
  10. Bill Yuchen Lin, Yicheng Fu, Karina Yang, Faeze Brahman, Shiyu Huang, Chandra Bhagavatula, Prithviraj Ammanabrolu, Yejin Choi, and Xiang Ren. 2023. Swiftsage: A generative agent with fast and slow thinking for complex interactive tasks. In Thirty-seventh Conference on Neural Information Processing S…
  11. Jianlan Luo, Perry Dong, Yuexiang Zhai, Yi Ma, and Sergey Levine. 2024. RLIF: Interactive imitation learning as reinforcement learning. In The Twelfth International Conference on Learning Representations. find
  12. Aman Madaan, Niket Tandon, Prakhar Gupta, Skyler Hallinan, Luyu Gao, Sarah Wiegreffe, Uri Alon, Nouha Dziri, Shrimai Prabhumoye, Yiming Yang, Shashank Gupta, Bodhisattwa Prasad Majumder, Katherine Hermann, Sean Welleck, Amir Yazdanbakhsh, and Peter Clark. 2023. Self-refine: Iterative refinement with… find
  13. Yao Mu, Qinglong Zhang, Mengkang Hu, Wenhai Wang, Mingyu Ding, Jun Jin, Bin Wang, Jifeng Dai, Yu Qiao, and Ping Luo. 2023. EmbodiedGPT: Vision-language pre-training via embodied chain of thought. In Thirty-seventh Conference on Neural Information Processing Systems. find
  14. Xavier Puig, Kevin Kyunghwan Ra, Marko Boben, Jiaman Li, Tingwu Wang, Sanja Fidler, and Antonio Torralba. 2018. Virtualhome: Simulating household activities via programs. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 8494–8502. find
  15. Haofu Qian, Chenjia Bai, Jiatao Zhang, Fei Wu, Wei Song, and Xuelong Li. 2025. Discriminator-guided embodied planning for LLM agent. In The Thirteenth International Conference on Learning Representations. find
  16. Stephane Ross, Geoffrey Gordon, and Drew Bagnell. 2011. A reduction of imitation learning and structured prediction to no-regret online learning. In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics, volume 15 of Proceedings of Machine Learning Research… find
  17. Yu-Zhe Shi, Haofei Hou, Zhangqian Bi, Fanxu Meng, Xiang Wei, Lecheng Ruan, and Qining Wang. 2024. AutoDSL: Automated domain-specific language design for structural representation of procedures with constraints. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistic… find
  18. Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik R Narasimhan, and Shunyu Yao. 2023. Reflexion: language agents with verbal reinforcement learning. In Thirty-seventh Conference on Neural Information Processing Systems.
  19. Chan Hee Song, Jiaman Wu, Clay Washington, Brian M. Sadler, Wei-Lun Chao, and Yu Su. 2022. Llm-planner: Few-shot grounded planning for embodied agents with large language models. 2023 IEEE/CVF International Conference on Computer Vision (ICCV), pages 2986–2997.
  20. Robyn Speer, Joshua Chin, and Catherine Havasi. 2017. Conceptnet 5.5: An open multilingual graph of general knowledge. In Proceedings of the AAAI conference on artificial intelligence, volume 31. find
  21. Shunyu Yao, Rohan Rao, Matthew Hausknecht, and Karthik Narasimhan. 2020. Keep CALM and explore: Language models for action generation in text-based games. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 8736–8754, Online. Association for Compu…
  22. Shunyu Yao, Dian Yu, Jeffrey Zhao, Izhak Shafran, Thomas L. Griffiths, Yuan Cao, and Karthik R Narasimhan. 2023. Tree of thoughts: Deliberate problem solving with large language models. In Thirty-seventh Conference on Neural Information Processing Systems. find
  23. Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik R Narasimhan, and Yuan Cao. 2022. React: Synergizing reasoning and acting in language models. In The eleventh international conference on learning representations.
  24. Weiran Yao, Shelby Heinecke, Juan Carlos Niebles, Zhiwei Liu, Yihao Feng, Le Xue, Rithesh R N, Zeyuan Chen, Jianguo Zhang, Devansh Arpit, Ran Xu, Phil L Mui, Huan Wang, Caiming Xiong, and Silvio Savarese. 2024. Retroformer: Retrospective large language agents with policy gradient optimization. In Th… find
  25. Zirui Zhao, Wee Sun Lee, and David Hsu. 2023. Large language models as commonsense knowledge for large-scale task planning. In Thirty-seventh Conference on Neural Information Processing Systems. find
  26. Zhehua Zhou, Jiayang Song, Kunpeng Yao, Zhan Shu, and Lei Ma. 2024. Isr-llm: Iterative self-refined large language model for long-horizon sequential task planning. In ICRA, pages 2081–2088. find
  27. Ruoyao Wang, Peter Jansen, Marc-Alexandre Côté, and Prithviraj Ammanabrolu. 2022. Scienceworld: Is your agent smarter than a 5th grader? pages 11279– 11298. find
  28. Zihao Wang, Shaofei Cai, Guanzhou Chen, Anji Liu, Xiaojian Ma, and Yitao Liang. 2023. Describe, explain, plan and select: Interactive planning with LLMs enables open-world multi-task agents. In Thirty-seventh Conference on Neural Information Processing Systems. find
  29. Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, brian ichter, Fei Xia, Ed H. Chi, Quoc V Le, and Denny Zhou. 2022. Chain of thought prompting elicits reasoning in large language models. In Advances in Neural Information Processing Systems.
  30. Haoran Xu, Xianyuan Zhan, Honglei Yin, and Huiling Qin. 2022. Discriminator-weighted offline imitation learning from suboptimal demonstrations. In International Conference on Machine Learning, pages 24725–24742. PMLR. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/liu-2026-ontoguard-enforcing-action-admissibility-llm/