NCUT-AI/Carla-Road-X1

Carla-Road-X1是一个用于训练语言模型从自然语言描述生成OpenDRIVE（.xodr）道路网络文件的文本到xodr数据集。该数据集包含约428,811个样本，其中训练样本385,929个，验证样本42,882个，以及40个模板样本。数据来源于OpenStreetMap区域转换的xodr子网络和手动设计的CARLA道路网络模板。数据格式为JSONL，使用Gemma聊天模板（系统/用户/助理消息）。数据集旨在微调大型语言模型，以支持自动驾驶仿真场景生成和鸟瞰图道路网络合成。Carla-Road-X1 is a text-to-xodr dataset for training language models to generate OpenDRIVE (.xodr) road network files from natural language descriptions. The dataset includes approximately 428,811 samples, with 385,929 training samples, 42,882 validation samples, and 40 template samples. Data sources consist of OSM-converted xodr sub-networks and hand-crafted CARLA road network templates. The format is JSONL with Gemma chat template (system/user/assistant messages). It is intended for fine-tuning LLMs for OpenDRIVE xodr generation, BEV road network synthesis, and autonomous driving simulation scene generation.