Conference paper · 2025 · Open access

ATLAS: Agent Tuning via Learning Critical Steps

Zijing Chen, Ming Li, Yuxuan Huang, Yali Du, Meng Fang, Tianyi Zhou

Findings of the Association for Computational Linguistics: ACL (Findings of ACL), 25334–25349 · Published 1 January 2025 · DOI 10.18653/v1/2025.findings-acl.1299

Free PDFView at the publisherCite this paper

In brief

“ATLAS: Agent Tuning via Learning Critical Steps” is a 2025 conference paper by Zijing Chen and 5 co-authors, published in Findings of the Association for Computational Linguistics: ACL (Findings of ACL). OpenAlex counts 30 citations as of 16 November 2026, which puts it in the top 1% of works published in 2025 by citations. It is open access under a CC BY licence: the full text is free to read at Findings of the Association for Computational Linguistics: ACL (Findings of ACL).

Citations
30
Most cited in
2026 (17)
Field-weighted impact
10.56
References
32
Authors
6

Abstract

Large Language Model (LLM) agents have demonstrated remarkable generalization capabilities across multi-domain tasks.Existing agent tuning approaches typically employ supervised finetuning on entire expert trajectories.However, behavior-cloning of full trajectories can introduce expert bias and weaken generalization to states not covered by the expert data.Additionally, critical steps-such as planning, complex reasoning for intermediate subtasks, and strategic decision-making-are essential to success in agent tasks, so learning these steps is the key to improving LLM agents.For more effective and efficient agent tuning, we propose ATLAS that identifies the critical steps in expert trajectories and finetunes LLMs solely on these steps with reduced costs.By steering the training's focus to a few critical steps, our method mitigates the risk of overfitting entire trajectories and promotes generalization across different environments and tasks.In extensive experiments, an LLM finetuned on only ∼30% critical steps selected by ATLAS outperforms the LLM finetuned on all steps and recent open-source LLM agents.ATLAS maintains and improves base LLM skills as generalist agents interacting with diverse environments.

Cite this paper

Formatted from the catalogue record. Check names, capitals and page numbers against the paper before you submit.

APA 7

Chen, Z., Li, M., Huang, Y., Du, Y., Fang, M., & Zhou, T. (2025). ATLAS: Agent Tuning via Learning Critical Steps. Findings of the Association for Computational Linguistics: ACL (Findings of ACL), 25334–25349. https://doi.org/10.18653/v1/2025.findings-acl.1299

MLA 9

Chen, Zijing, et al. “ATLAS: Agent Tuning via Learning Critical Steps.” Findings of the Association for Computational Linguistics: ACL (Findings of ACL), 2025, pp. 25334–25349. https://doi.org/10.18653/v1/2025.findings-acl.1299.

Chicago

Chen, Zijing, Ming Li, Yuxuan Huang, Yali Du, Meng Fang, and Tianyi Zhou. 2025. “ATLAS: Agent Tuning via Learning Critical Steps.” Findings of the Association for Computational Linguistics: ACL (Findings of ACL): 25334–25349. https://doi.org/10.18653/v1/2025.findings-acl.1299.

Harvard

Chen, Z. et al. (2025) ‘ATLAS: Agent Tuning via Learning Critical Steps’, Findings of the Association for Computational Linguistics: ACL (Findings of ACL), pp. 25334–25349. doi:10.18653/v1/2025.findings-acl.1299.

IEEE

Z. Chen, M. Li, Y. Huang, Y. Du, M. Fang, and T. Zhou, “ATLAS: Agent Tuning via Learning Critical Steps,” Findings of the Association for Computational Linguistics: ACL (Findings of ACL), pp. 25334–25349, 2025, doi: 10.18653/v1/2025.findings-acl.1299.

BibTeX

@inproceedings{chen2025atlas,
  title     = {{ATLAS: Agent Tuning via Learning Critical Steps}},
  author    = {Chen, Zijing and Li, Ming and Huang, Yuxuan and Du, Yali and Fang, Meng and Zhou, Tianyi},
  booktitle = {Findings of the Association for Computational Linguistics: ACL (Findings of ACL)},
  year      = {2025},
  pages     = {25334--25349},
  doi       = {10.18653/v1/2025.findings-acl.1299},
  url       = {https://doi.org/10.18653/v1/2025.findings-acl.1299}
}

Download for a reference manager: BibTeX (.bib) · RIS for Zotero, Mendeley and EndNote (.ris)

Topic and keywords

Physical Sciences › Computer Science › Artificial Intelligence › AI-based Problem Solving and Planning

  • LLM agents
  • expert trajectories
  • behavior cloning
  • overfitting
  • generalization
  • planning
  • strategic decision-making

Work with this paper

Most-cited papers in Computer Science

All highly cited Computer Science papers

Questions about this paper

How do I cite this paper in APA style?

Chen, Z., Li, M., Huang, Y., Du, Y., Fang, M., & Zhou, T. (2025). ATLAS: Agent Tuning via Learning Critical Steps. Findings of the Association for Computational Linguistics: ACL (Findings of ACL), 25334–25349. https://doi.org/10.18653/v1/2025.findings-acl.1299

How many times has this paper been cited?

30 times, according to OpenAlex on 16 November 2026. Google Scholar and Scopus count different sets of sources, so their numbers differ.

Is this paper free to read?

Yes. It is open access: read the free full text at Findings of the Association for Computational Linguistics: ACL (Findings of ACL).

Who wrote this paper?

Zijing Chen, Ming Li, Yuxuan Huang, Yali Du, Meng Fang, Tianyi Zhou.

What is the DOI of this paper?

10.18653/v1/2025.findings-acl.1299. The permanent link is https://doi.org/10.18653/v1/2025.findings-acl.1299.

Bibliographic data from OpenAlex (CC0), last checked 16 November 2026. ResearchGuru does not host this paper; the links go to the publisher or the repository that does.