{"status":"ok","message-type":"work","message-version":"1.0.0","message":{"indexed":{"date-parts":[[2026,7,7]],"date-time":"2026-07-07T14:32:51Z","timestamp":1783434771014,"version":"3.54.6"},"reference-count":30,"publisher":"MDPI AG","issue":"3","license":[{"start":{"date-parts":[[2026,3,12]],"date-time":"2026-03-12T00:00:00Z","timestamp":1773273600000},"content-version":"vor","delay-in-days":0,"URL":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/"}],"funder":[{"DOI":"10.13039\/501100001809","name":"National Natural Science Foundation of China","doi-asserted-by":"publisher","award":["52571413"],"award-info":[{"award-number":["52571413"]}],"id":[{"id":"10.13039\/501100001809","id-type":"DOI","asserted-by":"publisher"}]},{"name":"National Natural Science Foundation of China Youth Fund Project","award":["52401423"],"award-info":[{"award-number":["52401423"]}]}],"content-domain":{"domain":[],"crossmark-restriction":false},"short-container-title":["Information"],"abstract":"<jats:p>Deploying Large Language Models (LLMs) on Maritime Autonomous Surface Ships (MASS) entails a critical trade-off between reasoning depth, inference latency, and hardware constraints. To fill the existing gap, we introduce MARTIAN (Maritime Agent for Real-time Tactical Inference And Navigation), a 14B-parameter decision support agent engineered for edge deployment on standard vessel hardware (e.g., the NVIDIA Jetson AGX Orin). Central to our approach is the Cognitive Core architecture, which utilizes a verified dataset of 21,800 Chain-of-Thought (CoT) instruction\u2013response pairs to align general linguistic capabilities with maritime procedural logic. Empirical evaluations demonstrate that MARTIAN achieves an overall accuracy of 73.23% (SFT only) and 81.16% (SFT + RAG) on the Bilingual Maritime Multiple-Choice Questionnaire (BM-MCQ), a standardized assessment dataset constructed based on Officer of the Watch (OOW) competencies. Notably, the SFT-only configuration attains 78.53% on pure-logic-intensive COLREG tasks\u2014surpassing the 72B-parameter Qwen-2.5 foundation model in this domain\u2014while maintaining a real-time inference latency of 22.4 ms\/token. Crucially, our ablation studies support a nuanced Interference Hypothesis: while RAG significantly enhances factual recall in knowledge-intensive domains (boosting total accuracy from 73.23% to 81.16%), it concurrently introduces semantic noise that degrades performance in pure logic reasoning tasks (e.g., COLREG maneuvering accuracy decreases from 78.53% to 77.36%). On the basis of this finding, we identify and empirically motivate a decoupled cognitive design principle that separates procedural reflexes (via SFT) from declarative knowledge (via RAG). While the full implementation of an adaptive routing mechanism is deferred to future work, the ablation results presented herein offer a validated, cost-effective reference architecture for deploying transparent and regulation-compliant AI on resource-constrained merchant vessels.<\/jats:p>","DOI":"10.3390\/info17030284","type":"journal-article","created":{"date-parts":[[2026,3,12]],"date-time":"2026-03-12T13:12:24Z","timestamp":1773321144000},"page":"284","update-policy":"https:\/\/doi.org\/10.3390\/mdpi_crossmark_policy","source":"Crossref","is-referenced-by-count":2,"title":["Deploying Efficient LLM Agents on Maritime Autonomous Surface Ships: Fine-Tuning, RAG, and Function Calling in a Mid-Size Model"],"prefix":"10.3390","volume":"17","author":[{"given":"Yiling","family":"Ren","sequence":"first","affiliation":[{"name":"School of Navigation, Wuhan University of Technology, Wuhan 430063, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Mozi","family":"Chen","sequence":"additional","affiliation":[{"name":"School of Navigation, Wuhan University of Technology, Wuhan 430063, China"},{"name":"State Key Laboratory of Maritime Technology and Safety, Wuhan University of Technology, Wuhan 430063, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Junjie","family":"Weng","sequence":"additional","affiliation":[{"name":"School of Navigation, Wuhan University of Technology, Wuhan 430063, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0003-0154-4897","authenticated-orcid":false,"given":"Shengkai","family":"Zhang","sequence":"additional","affiliation":[{"name":"State Key Laboratory of Maritime Technology and Safety, Wuhan University of Technology, Wuhan 430063, China"},{"name":"School of Information Engineering, Wuhan University of Technology, Wuhan 430070, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"given":"Xuedou","family":"Xiao","sequence":"additional","affiliation":[{"name":"School of Navigation, Wuhan University of Technology, Wuhan 430063, China"}],"role":[{"vocabulary":"crossref","role":"author"}]},{"ORCID":"https:\/\/orcid.org\/0000-0002-1085-7556","authenticated-orcid":false,"given":"Kezhong","family":"Liu","sequence":"additional","affiliation":[{"name":"School of Navigation, Wuhan University of Technology, Wuhan 430063, China"},{"name":"State Key Laboratory of Maritime Technology and Safety, Wuhan University of Technology, Wuhan 430063, China"}],"role":[{"vocabulary":"crossref","role":"author"}]}],"member":"1968","published-online":{"date-parts":[[2026,3,12]]},"reference":[{"key":"ref_1","doi-asserted-by":"crossref","first-page":"119552","DOI":"10.1016\/j.oceaneng.2024.119552","article-title":"COLREG scenario classification and compliance evaluation with temporal and multi-vessel awareness for collision avoidance systems","volume":"313","author":"Gleeson","year":"2024","journal-title":"Ocean Eng."},{"key":"ref_2","doi-asserted-by":"crossref","first-page":"116530","DOI":"10.1016\/j.oceaneng.2023.116530","article-title":"Autonomous collision avoidance method for MASSs based on precise potential field modelling and COLREGs constraints in complex sailing environments","volume":"292","author":"Lyu","year":"2024","journal-title":"Ocean Eng."},{"key":"ref_3","doi-asserted-by":"crossref","first-page":"110875","DOI":"10.1016\/j.ress.2025.110875","article-title":"Investigation of ship collision accident risk factors using BP-DEMATEL method based on HFACS-SCA","volume":"257","author":"Guo","year":"2025","journal-title":"Reliab. Eng. Syst. Saf."},{"key":"ref_4","doi-asserted-by":"crossref","first-page":"110201","DOI":"10.1016\/j.ress.2024.110201","article-title":"Human and organizational factors analysis of collision accidents between merchant ships and fishing vessels based on HFACS-BN model","volume":"249","author":"Wang","year":"2024","journal-title":"Reliab. Eng. Syst. Saf."},{"key":"ref_5","doi-asserted-by":"crossref","first-page":"109166","DOI":"10.1016\/j.ress.2023.109166","article-title":"Quantitative assessment of ship collision risk influencing factors from worldwide accident and fleet data","volume":"234","author":"Sun","year":"2023","journal-title":"Reliab. Eng. Syst. Saf."},{"key":"ref_6","doi-asserted-by":"crossref","first-page":"110489","DOI":"10.1016\/j.ress.2024.110489","article-title":"Systems driven intelligent decision support methods for ship collision and grounding prevention: Present status, possible solutions, and challenges","volume":"253","author":"Zhang","year":"2025","journal-title":"Reliab. Eng. Syst. Saf."},{"key":"ref_7","unstructured":"International Maritime Organization (IMO) (1972). Convention on the International Regulations for Preventing Collisions at Sea, 1972 (COLREG 1972), IMO."},{"key":"ref_8","doi-asserted-by":"crossref","first-page":"119907","DOI":"10.1016\/j.oceaneng.2024.119907","article-title":"A COLREGs compliance reinforcement learning approach for USV manoeuvring in track-following and collision avoidance problems","volume":"316","author":"Sonntag","year":"2025","journal-title":"Ocean Eng."},{"key":"ref_9","doi-asserted-by":"crossref","first-page":"116088","DOI":"10.1016\/j.oceaneng.2023.116088","article-title":"A novel route-plan-guided artificial potential field method for ship collision avoidance: Modeling, integration and test","volume":"288","author":"Zhu","year":"2023","journal-title":"Ocean Eng."},{"key":"ref_10","doi-asserted-by":"crossref","first-page":"119512","DOI":"10.1016\/j.oceaneng.2024.119512","article-title":"A novel collaborative collision avoidance decision method for multi-ship encounters in complex waterways","volume":"313","author":"Dong","year":"2024","journal-title":"Ocean Eng."},{"key":"ref_11","doi-asserted-by":"crossref","first-page":"107450","DOI":"10.1016\/j.ocecoaman.2024.107450","article-title":"A game-based decision-making method for multi-ship collaborative collision avoidance reflecting risk attitudes in open waters","volume":"259","author":"Liu","year":"2024","journal-title":"Ocean Coast. Manag."},{"key":"ref_12","doi-asserted-by":"crossref","first-page":"110810","DOI":"10.1016\/j.ress.2025.110810","article-title":"Investigation into safety acceptance principles for autonomous ships","volume":"257","author":"Bolbot","year":"2025","journal-title":"Reliab. Eng. Syst. Saf."},{"key":"ref_13","unstructured":"Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., and Polosukhin, I. (2017). Attention is All You Need. Advances in Neural Information Processing Systems, Curran Associates, Inc."},{"key":"ref_14","unstructured":"Wen, L., Fu, D., Li, X., Cai, X., Ma, T., Cai, P., Dou, M., Shi, B., He, L., and Qiao, Y. (2024, January 7\u201311). DiLu: A Knowledge-Driven Approach to Autonomous Driving with Large Language Models. Proceedings of the International Conference on Learning Representations (ICLR), Vienna, Austria."},{"key":"ref_15","doi-asserted-by":"crossref","first-page":"100116","DOI":"10.1016\/j.commtr.2023.100116","article-title":"GPT-4 enhanced multimodal grounding for autonomous driving: Leveraging cross-modal attention with large language models","volume":"4","author":"Liao","year":"2024","journal-title":"Commun. Transp. Res."},{"key":"ref_16","unstructured":"Chen, Y., Cui, W., Chen, Y., Tan, M., Zhang, X., Zhao, D., and Wang, H. (2023). RoboGPT: An intelligent agent of making embodied long-term decisions for daily instruction tasks. arXiv."},{"key":"ref_17","first-page":"1","article-title":"Emergent Abilities of Large Language Models","volume":"2022","author":"Wei","year":"2022","journal-title":"Trans. Mach. Learn. Res."},{"key":"ref_18","unstructured":"Zhou, G., Hong, Y., Wang, Z., Wang, X.E., and Wu, Q. (October, January 29). NavGPT-2: Unleashing Navigational Reasoning Capability for Large Vision-Language Models. Proceedings of the European Conference on Computer Vision (ECCV), Milan, Italy."},{"key":"ref_19","doi-asserted-by":"crossref","first-page":"110382","DOI":"10.1016\/j.clinimag.2024.110382","article-title":"Comparative diagnostic accuracy of GPT-4o and LLaMA 3-70b: Proprietary vs. open-source large language models in radiology","volume":"118","author":"Li","year":"2025","journal-title":"Clin. Imaging"},{"key":"ref_20","unstructured":"Hu, E.J., Shen, Y., Wallis, P., Allen-Zhu, Z., Li, Y., Wang, S., Wang, L., and Chen, W. (2022, January 25\u201329). LoRA: Low-Rank Adaptation of Large Language Models. Proceedings of the International Conference on Learning Representations (ICLR), Virtual."},{"key":"ref_21","doi-asserted-by":"crossref","first-page":"10088","DOI":"10.52202\/075280-0441","article-title":"QLoRA: Efficient Finetuning of Quantized LLMs","volume":"Volume 36","author":"Dettmers","year":"2023","journal-title":"Advances in Neural Information Processing Systems (NeurIPS)"},{"key":"ref_22","doi-asserted-by":"crossref","unstructured":"Lin, J., Tang, J., Tang, H., Yang, S., Xiao, G., and Han, S. (2024). AWQ: Activation-aware Weight Quantization for On-Device LLM Compression and Acceleration. GetMobile: Mobile Computing and Communications, Association for Computing Machinery.","DOI":"10.1145\/3714983.3714987"},{"key":"ref_23","first-page":"27730","article-title":"Training language models to follow instructions with human feedback","volume":"Volume 35","author":"Ouyang","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref_24","first-page":"9459","article-title":"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks","volume":"Volume 33","author":"Lewis","year":"2020","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref_25","unstructured":"Xiao, S., Liu, Z., Zhang, P., and Muennighoff, N. (2023). C-Pack: Packaged Resources To Advance General Chinese Embedding. arXiv."},{"key":"ref_26","first-page":"68539","article-title":"Toolformer: Language Models Can Teach Themselves to Use Tools","volume":"Volume 36","author":"Schick","year":"2023","journal-title":"Advances in Neural Information Processing Systems (NeurIPS)"},{"key":"ref_27","unstructured":"DeepSeek-AI, Guo, D., Yang, D., and Zhang, H. (2025). DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. arXiv."},{"key":"ref_28","doi-asserted-by":"crossref","first-page":"16344","DOI":"10.52202\/068431-1189","article-title":"FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness","volume":"Volume 35","author":"Dao","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref_29","first-page":"24824","article-title":"Chain-of-Thought Prompting Elicits Reasoning in Large Language Models","volume":"Volume 35","author":"Wei","year":"2022","journal-title":"Advances in Neural Information Processing Systems"},{"key":"ref_30","unstructured":"Hendrycks, D., Burns, C., Basart, S., Zou, A., Mazeika, M., Song, D., and Steinhardt, J. (2021). Measuring Massive Multitask Language Understanding. Proceedings of the International Conference on Learning Representations (ICLR), Vienna, Austria, Curran Associates, Inc."}],"container-title":["Information"],"original-title":[],"language":"en","link":[{"URL":"https:\/\/www.mdpi.com\/2078-2489\/17\/3\/284\/pdf","content-type":"unspecified","content-version":"vor","intended-application":"similarity-checking"}],"deposited":{"date-parts":[[2026,3,12]],"date-time":"2026-03-12T13:51:27Z","timestamp":1773323487000},"score":1,"resource":{"primary":{"URL":"https:\/\/www.mdpi.com\/2078-2489\/17\/3\/284"}},"subtitle":[],"short-title":[],"issued":{"date-parts":[[2026,3,12]]},"references-count":30,"journal-issue":{"issue":"3","published-online":{"date-parts":[[2026,3]]}},"alternative-id":["info17030284"],"URL":"https:\/\/doi.org\/10.3390\/info17030284","relation":{},"ISSN":["2078-2489"],"issn-type":[{"value":"2078-2489","type":"electronic"}],"subject":[],"published":{"date-parts":[[2026,3,12]]}}}