TransitLM: A Large-Scale Dataset and Benchmark for Map-Free Transit Route Generation

Hanyu Guo, Jiedong Yang, Chao Chen, Longfei Xu, Kaikui Liu, Xiangxiang Chu

arXiv:2605.22355 · 2026-05-22 공개 · arXiv · PDF

llm-training large-scale-dataset transit-route-generation map-free gps-coordinate-mapping public-transit route-planning china-cities

Abstract

Public transit route planning traditionally depends on structured map infrastructure and complex routing engines, and no existing dataset supports training models to bypass this dependency. We present TransitLM, a large-scale dataset of over 13 million transit route planning records from four Chinese cities covering 120,845 stations and 13,666 lines, released as a continual pre-training corpus and benchmark data for three evaluation tasks with complementary metrics. Experiments show that an LLM trained on TransitLM produces structurally valid routes at high accuracy and implicitly grounds arbitrary GPS coordinates to appropriate stations without any explicit mapping. These results demonstrate that transit route planning can be learned entirely from data, enabling end-to-end, map-free route generation directly from origin-destination information. The dataset and benchmark are available at https://huggingface.co/datasets/GD-ML/TransitLM, with evaluation code at https://github.com/HotTricker/TransitLM.

한국어 요약

📋 한 줄 요약

**[Transit Routing / LLM Pretraining]** TransitLM이 4 중국 도시 1,300만+ 대중교통 경로 기록(12만 역·1.4만 노선)을 continual pre-training corpus로 공개, LLM이 GPS 좌표→역 그라운딩·구조적 유효 경로 생성을 map-free로 학습.

🎯 핵심 기여도

💡 핵심 아이디어

대중교통 경로 계획은 map infrastructure·routing engine 없이도 end-to-end·map-free로 데이터만으로 학습 가능하며, 충분 규모(13M+)의 다도시 route planning corpus와 적절 benchmark가 있으면 LLM이 GPS 좌표→station을 implicit grounding하고 structurally valid route를 생성할 수 있다.

🔬 기술적 접근법

📊 주요 결과

💭 의의 및 한계

**의의**: Map infrastructure 의존 제거의 paradigm 검증, GPS 좌표→station grounding이라는 implicit spatial reasoning 능력 노출, 도시 routing 데이터의 LLM corpus화 첫 대규모 시도. **한계**: 4 중국 도시 한정으로 다른 국가·도시 일반화 추가 검증 필요, dynamic schedule·delay 등 실시간 정보 반영 미커버, 13M record가 transit 외 다른 modality·도시 routing에는 부족.

🚀 실용적 활용