The invention relates to an automatic driving decision-making method and system, a medium, an automatic driving system and a vehicle, and relates to the technical field of automatic driving. The method comprises the following steps: acquiring a scene category, a target and an environment detection result; generating a first prompt text in a natural language form based on the scene category, the target and the environment detection result; querying the large language model by using the first prompt text to obtain a first query result in a natural language form; determining an expected target according to a plurality of presets and generating a formatted first prompt template; combining the first query result with a formatted first prompt template to generate a second prompt text in a natural language form; and querying the large language model by using the second prompt text to obtain a second query result in a structured natural language form. Through the method, the driving decision can be made by using the common sense of a large language model, the method has interpretability, and the method can be integrated to a traditional automatic driving software protocol stack.
本公开涉及一种自动驾驶决策方法、系统、介质、自动驾驶系统及车辆,涉及自动驾驶技术领域。其中,该方法包括:获取场景类别、目标和环境检测结果;基于场景类别、目标和环境检测结果生成自然语言形式的第一提示文本;利用第一提示文本查询大型语言模型,得到自然语言形式的第一查询结果;根据多个预设确认预期目标且生成格式化的第一提示模板;将第一查询结果与格式化的第一提示模板组合,生成自然语言形式的第二提示文本;利用第二提示文本查询大型语言模型,得到结构化的自然语言形式的第二查询结果。通过以上方法,能利用大型语言模型的常识进行驾驶决策且具备可解释性,且可集成到传统自动驾驶软件协议栈。
Automatic driving decision-making method and system, medium, automatic driving system and vehicle
自动驾驶决策方法、系统、介质、自动驾驶系统及车辆
2024-03-22
Patent
Electronic Resource
Chinese
Automatic driving decision-making method and system and vehicle
European Patent Office | 2024
|Automatic driving vehicle behavior decision-making system and method
European Patent Office | 2023
|European Patent Office | 2023
|European Patent Office | 2023
|