ShipTraj-R1: Reinforcing Ship Trajectory Prediction in Large Language Models via Group Relative Policy Optimization

Yang Zhan; Yunhao Li; Zhang Chao; Yuxu Lu; Yan Li

cs.AIMar 3, 2026

ShipTraj-R1: Reinforcing Ship Trajectory Prediction in Large Language Models via Group Relative Policy Optimization

Authors: Yang Zhan, Yunhao Li, Zhang Chao, Yuxu Lu, Yan Li

Organizations: School of Artificial Intelligence, Optics, and Electronics (iOPEN), Northwestern Polytechnical University, Xi’an 710072, China · Department of Logistics and Maritime Studies, The Hong Kong Polytechnic University, Hong Kong 999077, China · State Key Laboratory of Information Engineering in Surveying, Mapping and Remote Sensing, Wuhan University, Wuhan 430079, China

Abstract

Recent advancements in reinforcement fine-tuning have significantly improved the reasoning ability of large language models (LLMs). In particular, methods such as group relative policy optimization (GRPO) have demonstrated strong capabilities across various fields. However, applying LLMs to ship trajectory prediction remains largely unexplored. In this paper, we propose ShipTraj-R1, a novel LLM-based framework that reformulates ship trajectory prediction as a text-to-text generation problem. (1) We design a dynamic prompt containing trajectory information about conflicting ships to guide the model to achieve adaptive chain-of-thought (CoT) reasoning. (2) We introduce a comprehensive rule-based reward mechanism to incentivize the reasoning format and prediction accuracy of the model. (3) Our ShipTraj-R1 is reinforced through the GRPO mechanism guided by domain-specific prompts and rewards, and utilizes the Qwen3 as the model backbone. Extensive experimental results on two complex and real-world maritime datasets show that the proposed ShipTraj-R1 achieves the least error compared with state-of-the-art deep learning and LLM-based baselines.

Figures & tables

Figure 1: Overview of our proposed ShipTraj-R1 framework.

DL-based Methods
Model	Release Year-month	CSJP		CFDP
Model		FDE	ADE	FDE	ADE
TBENet ${}_{{\color[rgb]{0.5,0.5,0.5}\text{TRE'24}}}$ [ 8 ]		24-11	0.009827	8.1000e-04	0.005993	4.0200e-04
GeoCLSTM ${}_{{\color[rgb]{0.5,0.5,0.5}\text{CAEE'24}}}$ [ 9 ]	24-12	0.070173	5.4530e-03	0.064019	4.1890e-03
DBSCAN-GeoCLSTM ${}_{{\color[rgb]{0.5,0.5,0.5}\text{CAEE'24}}}$ [ 9 ]	24-12	0.033494	2.2640e-03	0.030183	1.4730e-03
GAT-LSTM ${}_{{\color[rgb]{0.5,0.5,0.5}\text{JMSE'24}}}$ [ 11 ]	24-12	0.012828	2.7260e-03	0.009032	1.0520e-03

Table 1: Comparisons with representative DL-based methods, LLM-based methods, and SoTA open-source LLMs.

Figure 2: Qualitative examples of our ShipTraj-R1 on the CSJP dataset. In this case, the ShipTraj-R1 adaptively generates CoT reasoning by itself based on the user prompt.

Figure 3: Illustration of predicted ship trajectories from the CSJP and CFDP datasets.

Model	Type	CoT	CSJP		CFDP
Model	Type	CoT	FDE	ADE	FDE	ADE
Baseline-8B	-	-	0.002293	2.0529e-05	0.000474	6.7403e-07
ShipTraj-R1-8B	SFT	✗	0.001701	1.8153e-05	0.000412	5.1295e-07
ShipTraj-R1-8B	RFT	✗	0.001430	1.6216e-05	0.000341	4.1216e-07
ShipTraj-R1-8B	RFT	$\checkmark$	0.001297	1.1547e-05	0.000311	3.8912e-07

Table 2: Ablation on different fine-tuning methods and chain-of-thought.

Model	Conflicting Ships	CSJP		CFDP
Model	Conflicting Ships	FDE	ADE	FDE	ADE
ShipTraj-R1-8B	✗	0.002756	2.7804e-05	0.000569	6.3881e-07
ShipTraj-R1-8B	$\checkmark$	0.001297	1.1547e-05	0.000311	3.8912e-07

Table 3: Ablation on whether the user prompt includes conflicting ship context.

Model	KL loss coef	CSJP		CFDP
Model	KL loss coef	FDE	ADE	FDE	ADE
ShipTraj-R1-8B	1e-4	0.001297	1.1547e-05	0.000311	3.8912e-07
ShipTraj-R1-8B	5e-4	0.001394	1.4580e-05	0.000336	4.4628e-07
ShipTraj-R1-8B	1e-3	0.002287	2.0466e-05	0.000519	7.3491e-07
ShipTraj-R1-8B	5e-3	0.003650	3.6814e-05	0.000842	2.1277e-06

Table 4: Ablation on the the KL loss coefficient.

Explore similar work

Towards Long-Horizon Vessel Trajectory and Destination Forecasting with Reasoning Large Language Models
Jun 7, 2026Hongwei Wang, Miao Zhou, Fengde Wang +10RL for Language Models Vessel Trajectory Prediction
Institute of High Performance Computing (IHPC), A*STAR, Singapore 138632, Singapore. · The Key Laboratory of Road and Traffic Engineering, Ministry of Education, Tongji University, Shanghai 201804, China. · Meituan Inc., Shenzhen, China. +3
M3-Former: Multimodal Transformer with Mixture-of-Experts for Long-Term Vessel Trajectory Prediction
Sep 11, 2026Wenzhe Jin, Haina TangMixture of Experts Vessel Trajectory Prediction
School of Artificial Intelligence, University of Chinese Academy of Sciences, Beijing, China
EnvShip-Bench: An Environment-Enhanced Benchmark for Short-Term Vessel Trajectory Prediction
Jun 13, 2026Kun Ma, Qilong Han, Chengjing Song +3Forecasting Benchmarks Vessel Trajectory Prediction
Harbin Engineering University Harbin, China · Politecnico di Torino Turin, Italy · Institute of Software, Chinese Academy of Sciences Beijing, China

Jun 7, 2026J/K move · Enter open · S save

Towards Long-Horizon Vessel Trajectory and Destination Forecasting with Reasoning Large Language Models

Hongwei Wang, Miao Zhou, Fengde Wang, Yuting Wang+9

Institute of High Performance Computing (IHPC), A*STAR, Singapore 138632, Singapore. · The Key Laboratory of Road and Traffic Engineering, Ministry of Education, Tongji University, Shanghai 201804, China. · Meituan Inc., Shenzhen, China.+3

RL for Language Models Vessel Trajectory Prediction Reinforcement Learning with Verifiable Rewards Trajectory Prediction

Open paper AskPDF arXiv

Long-horizon maritime trajectory prediction is important for shipping management, logistics planning, and maritime risk analysis, yet month-level forecasting remains insufficiently studied. Existing deep learning methods mainly focus on short- and mid-term coordinate extrapolation and often struggle to preserve route feasibility and destination correctness over extended horizons. This paper investigates joint long-horizon vessel trajectory and destination forecasting with reasoning-capable large language models, and develops a Maritime LLM post-training framework based on Reinforcement Learning with Verifiable Reward (RLVR). An AIS-based benchmark is constructed with 60-day historical trajectories and 30-day forecasting horizons, where trajectories are converted into semantic textual representations for RL prompt construction. RLVR aligns LLMs with maritime forecasting objectives by enforcing physical validity, providing early-weighted trajectory supervision, and evaluating destination correctness through hierarchical matching and curriculum learning. Experimental results show that RLVR-trained LLMs substantially improve over zero-shot LLMs and representative deep learning baselines, especially on destination-related metrics. Among the evaluated RLVR-trained variants, 4B LLMs achieve the best overall performance, suggesting that reward-compatible optimization and task-specific capacity matching are more important than simply using larger 8B or 14B LLMs. The results also show that LSTM remains a strong deep learning baseline under limited fine-tuning data, while Transformer-style spatio-temporal models typically require larger datasets and richer structured inputs. Overall, this work advances semantic, verifier-aligned maritime forecasting for operational decision support.

Abstract

Explore similar work

Towards Long-Horizon Vessel Trajectory and Destination Forecasting with Reasoning Large Language Models

M3-Former: Multimodal Transformer with Mixture-of-Experts for Long-Term Vessel Trajectory Prediction

EnvShip-Bench: An Environment-Enhanced Benchmark for Short-Term Vessel Trajectory Prediction