PhysicalRisk

该数据集是一个包含来自多种环境如厨房、浴室、实验室和工厂的广泛样本集，其中不仅包含安全提示，还涉及场景信息和可能违反安全提示的任务指导。此外，该数据集还包括307条安全提示、921个文本风险场景以及3,684个多模态场景，用于评估不同大型语言模型的风险意识。规模上，该数据集共有4,605个样本，分布在15个不同的环境之中。其任务是对在实体人工智能代理中进行任务规划时的物理风险意识进行评估。

This dataset is a comprehensive sample collection spanning diverse environments including kitchens, bathrooms, laboratories and factories. The samples within it include not only safety guidelines, but also scenario information and task instructions that may violate these safety guidelines. Furthermore, this dataset specifically contains 307 safety guidelines, 921 textual risk scenarios and 3,684 multimodal scenarios, which are designed to evaluate the risk awareness of various large language models (LLMs). In terms of scale, the dataset has a total of 4,605 samples distributed across 15 distinct environments. Its core task is to assess physical risk awareness during task planning implemented in physical AI Agents.