TransGPT: Multi-modal Generative Pre-trained Transformer for Transportation
Peng Wang, Xiang Wei, Fangxu Hu, Wenjuan Han
Abstract
Transformer-based AI systems enhance human capabilities across various domains but face challenges in transportation due to domain-specific knowledge and multimodal data requirements. This paper introduces TransGPT, a large language model tailored for the transportation domain, with two variants: TransGPT-SM for single-modal data and TransGPT-MM for multimodal data. TransGPT-SM is fine-tuned on a textual Transportation dataset (STD), while TransGPT-MM is fine-tuned on a multimodal Transportation dataset (MTD) covering driving tests, traffic signs, and landmarks. Evaluations on benchmark datasets show TransGPT outperforms baseline models in most tasks. TransGPT’s applications include generating synthetic traffic scenarios, explaining traffic phenomena, answering traffic-related questions, providing traffic recommendations, and generating traffic reports, advancing NLP in transportation and benefiting ITS researchers and practitioners.