Elamur Research⌕
☰
home / papers / 2026 / Ro-SLM: Onboard Small Language Models for Robot Task Planning and Operation Code Generation

Ro-SLM: Onboard Small Language Models for Robot Task Planning and Operation Code Generation

Paper · 2026 · Wenhao Wang, Yanyan Li, Long Jiao et al.

Summary

Robots that interpret natural-language commands usually depend on cloud-hosted or very large LLMs. This rules out platforms such as UAVs and small ground vehicles, which may have weak connectivity and limited compute. Ro-SLM addresses this by transferring LLM knowledge into a small model that runs onboard.

The pipeline has two stages. First, LLM agents build a synthetic training set:

  • they create task instructions of graded complexity, guided by robot policies and demonstrations;
  • they produce ground-truth operation code with a corrective generation method, and humans step in only for the hardest tasks;
  • they rewrite instructions as realistic application scenarios.

Code comments are stripped so the model does not latch onto their wording. Second, the small model undergoes SFT and then GRPO, with an LLM judging whether the output matches the reference code.

On simulated UAV tasks, fine-tuned Llama-3.1-8B moves from essentially unusable to near LLM-level results, roughly matching OpenAI o4-mini. A ground-vehicle experiment suggests the approach transfers across platforms. This makes language-driven robot control without network access more realistic.

Key points

  • Using Llama-3.1-8B on simple UAV tasks, Ro-SLM attains a Success Rate of 97.7% along with 98.9% Completeness. On complex tasks it reaches 70% Success Rate and 83.7% Completeness.
  • On Advanced tasks split into four complexity levels, the SLM's success rate at the Very High level exceeds the LLM's. The authors credit the human-verified ground-truth code for this.
  • On eight ROSMASTER X3 ground-vehicle tasks, Ro-SLM scores 75% SR and 81.1% Completeness. The LLM-based Corrective GSCE baseline scores 87.5% SR and 97.2% Completeness.
  • Stated limitations:
    • the 8B model still needs about 16GB of memory;
    • evaluation was done only in simulation;
    • complex tasks still require some human checking of the data;
    • inference is single-pass, with no iterative correction.
  • The code is available at github.com/ai-uavsec/Ro-SLM.

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

Also on the byline: University of Massachusetts Dartmouth · California State University, San Marcos.

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 worksEngineering · 133 worksControl and Systems Engineering · 45 works

🧭 Topics 2Research subjects and languages the work is about.Language models & pretraining · 3,320 worksCode & programming languages · 129 works

Cite

Cite as Wenhao Wang, Yanyan Li, Long Jiao and Jiawei Yuan. 2026. Ro-SLM: Onboard Small Language Models for Robot Task Planning and Operation Code Generation. 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.1720

@inproceedings{wang2026ro,
  title = {{Ro-SLM: Onboard Small Language Models for Robot Task Planning and Operation Code Generation}},
  author = {Wenhao Wang, Yanyan Li, Long Jiao and Jiawei Yuan},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1720},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1720}
}

References

40 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Josh Achiam, Steven Adler, Sandhini Agarwal, Lama Ahmad, Ilge Akkaya, Florencia Leoni Aleman, Diogo Almeida, Janko Altenschmidt, Sam Altman, Shyamal Anadkat, and 1 others. 2023. [gpt]-4 technical report. arXiv preprint arXiv:2303.08774.
  2. Montserrat Gonzalez Arenas, Ted Xiao, Sumeet Singh, Vidhi Jain, Allen Ren, Quan Vuong, Jake Varley, Alexander Herzog, Isabel Leal, Sean Kirmani, Mario Prats, Dorsa Sadigh, Vikas Sindhwani, Kanishka Rao, Jacky Liang, and Andy Zeng. 2024. How to prompt your robot: A promptbook for manipulation skills… find
  3. Berk Atil, Alexa Chittams, Liseng Fu, Ferhan Ture, Lixinyu Xu, and Breck Baldwin. 2024. Llm stability: A detailed analysis with some surprises. arXiv preprint arXiv:2408.04667. — [arXiv:2408.04667](https://arxiv.org/abs/2408.04667) find
  4. Guojun Chen, Xiaojing Yu, Neiwen Ling, and Lin Zhong. 2025a. Typefly: Low-latency drone planning with large language models. IEEE Transactions on Mobile Computing, 24(9):9068–9079. find
  5. Yongchao Chen, Jacob Arkin, Charles Dawson, Yang Zhang, Nicholas Roy, and Chuchu Fan. 2024. Autotamp: Autoregressive task and motion planning with llms as translators and checkers. In 2024 IEEE International Conference on Robotics and Automation (ICRA), pages 6695–6702. find
  6. Yuxuan Chen, Yixin Han, and Xiao Li. 2025b. Fastnav: Fine-tuned adaptive small-language- models trained for multi-point robot navigation. IEEE Robotics and Automation Letters, 10(1):390–397. find
  7. Zihan Chen, Song Wang, Zhen Tan, Xingbo Fu, Zhenyu Lei, Peng Wang, Huan Liu, Cong Shen, and Jundong Li. 2025c. A survey of scaling in large language model reasoning. arXiv preprint arXiv:2504.02181. find
  8. Abraham George and Amir Barati Farimani. 2025. Llm trainer: Automated robotic data generating via demonstration augmentation using llms. arXiv preprint arXiv:2509.20070. — [arXiv:2509.20070](https://arxiv.org/abs/2509.20070) find
  9. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  10. Edward J Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. Lora: Low-rank adaptation of large language models. ICLR, 1(2):3.
  11. Zichao Hu, Junyi Jessy Li, Arjun Guha, and Joydeep Biswas. 2025. Robo-instruct: Simulator-augmented instruction alignment for finetuning code LLMs. In Second Conference on Language Modeling. find
  12. Shashank Mohan Jain. 2022. Hugging face. In Introduction to transformers for NLP: With the hugging face library and models to solve problems, pages 51–67. Springer. find
  13. Frank Joublin, Antonello Ceravola, Pavel Smirnov, Felix Ocker, Joerg Deigmoeller, Anna Belardinelli, Chao Wang, Stephan Hasler, Daniel Tanneberg, and Michael Gienger. 2024. Copal: Corrective planning of robot actions with large language models. In 2024 IEEE International Conference on Robotics and A… find
  14. Jacky Liang, Wenlong Huang, Fei Xia, Peng Xu, Karol Hausman, Brian Ichter, Pete Florence, and Andy Zeng. 2023. Code as policies: Language model programs for embodied control. In 2023 IEEE International Conference on Robotics and Automation (ICRA), pages 9493–9500. find
  15. Zhenyan Lu, Xiang Li, Dongqi Cai, Rongjie Yi, Fangming Liu, Xiwen Zhang, Nicholas D Lane, and Mengwei Xu. 2024. Small language models: Survey, measurements, and insights. arXiv preprint arXiv:2409.15790.
  16. Ggaliwango Marvin, Nakayiza Hellen, Daudi Jjingo, and Joyce Nakatumba-Nabende. 2023. Prompt engineering in large language models. In International conference on data intelligence and cognitive informatics, pages 387–402. Springer. find
  17. Shervin Minaee, Tomas Mikolov, Narjes Nikzad, Meysam Chenaghlu, Richard Socher, Xavier Amatriain, and Jianfeng Gao. 2024. Large language models: A survey. arXiv preprint arXiv:2402.06196.
  18. MG Mohanan and Ambuja Salgoankar. 2018. A survey of robotic motion planning in dynamic environments. Robotics and Autonomous Systems, 100:171–185. find
  19. OpenAI Models. 2025a. GPT-5.1. Accessed: 25-Dec-2025. find
  20. OpenAI Models. 2025b. o4-mini. Accessed: 25-Dec-2025. find
  21. Long Ouyang, Jeffrey Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, Chong Zhang, Sandhini Agarwal, Katarina Slama, Alex Ray, and 1 others. 2022. Training language models to follow instructions with human feedback. Advances in neural information processing systems, 35:27730–27744.
  22. Zachary Ravichandran, Ignacio Hounie, Fernando Cladera, Alejandro Ribeiro, George J Pappas, and Vijay Kumar. 2025. Distilling on-device language models for robot planning with minimal human intervention. arXiv preprint arXiv:2506.17486. — [arXiv:2506.17486](https://arxiv.org/abs/2506.17486) find
  23. Microsoft Research. 2025. Setup blocks environment for airsim. https://microsoft.github.io/ AirSim/unreal_blocks/. Accessed: 2-Feb-2025. ↗
  24. Nicy Scaria, Silvester John Joseph Kennedy, and Deepak Subramani. 2025. Sensitivity of small language models to fine-tuning data contamination. arXiv preprint arXiv:2511.06763. find
  25. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. arXiv preprint arXiv:1707.06347. find
  26. Shital Shah, Debadeepta Dey, Chris Lovett, and Ashish Kapoor. 2018. Airsim: High-fidelity visual and physical simulation for autonomous vehicles. In Field and Service Robotics: Results of the 11th International Conference, pages 621–635. Springer. find
  27. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, and 1 others. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  28. Junjie Wen, Yichen Zhu, Jinming Li, Minjie Zhu, Zhibin Tang, Kun Wu, Zhiyuan Xu, Ning Liu, Ran Cheng, Chaomin Shen, Yaxin Peng, Feifei Feng, and Jian Tang. 2025. Tinyvla: Toward fast, data-efficient vision-language-action models for robotic manipulation. IEEE Robotics and Automation Letters, 10(4):3… find
  29. Bangguo Yu, Hamidreza Kasaei, and Ming Cao. 2023. L3mvn: Leveraging large language models for visual target navigation. In 2023 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pages 3554–3560. find
  30. Xianyang Zhan, Agam Goyal, Yilun Chen, Eshwar Chandrasekharan, and Koustuv Saha. 2025. Slm-mod: Small language models surpass llms at content moderation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Tech… find
  31. Andrea Tagliabue, Kota Kondo, Tong Zhao, Mason Peterson, Claudius T Tewari, and Jonathan P How. 2024. Real: Resilience and adaptation using large language models on autonomous aerial robots. In 2024 IEEE 63rd Conference on Decision and Control (CDC), pages 1539–1546. IEEE. find
  32. Yusen Zhang, Sarkar Snigdha Sarathi Das, and Rui Zhang. 2025. Demystify verbosity compensation behavior of large language models. In Proceedings of the 2nd Workshop on Uncertainty-Aware NLP (UncertaiNLP 2025), pages 160–178. find
  33. Gemini Team, Rohan Anil, Sebastian Borgeaud, Jean-Baptiste Alayrac, Jiahui Yu, Radu Soricut, Johan Schalkwyk, Andrew M Dai, Anja Hauth, Katie Millican, and 1 others. 2023. Gemini: a family of highly capable multimodal models. arXiv preprint arXiv:2312.11805.
  34. Jinbiao Zhu, Lishan Wang, Yi Guc, and Fei Wang. 2024. Ofslms: Offline fine-tuned small language models based on hybrid synthetic knowledge for robot introspective decision-making. In 2024 IEEE International Conference on Robotics and Biomimetics (ROBIO), pages 1839–1844. find
  35. Gemini Robotics Team, Saminda Abeyruwan, Joshua Ainslie, Jean-Baptiste Alayrac, Montserrat Gonzalez Arenas, Travis Armstrong, Ashwin Balakrishna, Robert Baruch, Maria Bauza, Michiel Blokzijl, and 1 others. 2025. Gemini robotics: Bringing ai into the physical world. arXiv preprint arXiv:2503.20020. find
  36. A Sai H. Vemprala, Rogerio Bonatti, Arthur Bucker, and Ashish Kapoor. 2024. Chatgpt for robotics: Design principles and model abilities. IEEE Access, 12:55682–55696. find
  37. Wenhao Wang, Yanyan Li, Long Jiao, and Jiawei Yuan. 2025a. Gsce: a prompt framework with enhanced reasoning for reliable llm-driven drone control. In 2025 International Conference on Unmanned Aircraft Systems (ICUAS), pages 441–448. find
  38. Wenhao Wang, Yanyan Li, Long Jiao, and Jiawei Yuan. 2025b. Llm-driven corrective robot operation code generation with static text-based simulation. arXiv preprint arXiv:2512.02002. — [arXiv:2512.02002](https://arxiv.org/abs/2512.02002) find
  39. Wenhao Wang, Yanyan Li, Long Jiao, and Jiawei Yuan. 2026. Large language model-driven closed-loop uav operation with semantic observations. IEEE Internet of Things Journal, 13(7):14465–14476. find
  40. Task Instruction Generation and System Prompts find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-ro-slm-onboard-small-language/