ASIMOV-Multimodal-Auto

该数据集由机器人和人类操作者执行日常任务的以自我为中心的视频生成，其中包含与导致安全或不安全结果的指令配对的图像。此外，该数据集还包括一项新颖的想象过程，能够从无害的场景生成危险情境。它是一个大规模的数据集，具有多模态性和多样化的场景，旨在用于视觉问答（VQA）任务，以增强对安全理解的能力。

This dataset is generated from egocentric videos of daily tasks performed by robots and human operators, and includes images paired with instructions that yield either safe or unsafe outcomes. Additionally, it features a novel imaginative process capable of generating dangerous scenarios from benign scenes. As a large-scale, multimodal dataset with diverse scenarios, it is intended for visual question answering (VQA) tasks to improve safety understanding.