IPEC-COMMUNITY/EO-Data1.5M

EO-Data-1.5M是一个大规模的交错视觉-文本-动作数据集，专为具身AI设计。它是EO-Robotics工具链的一部分，用于训练通用机器人策略。该数据集强调视觉、语言和动作模态之间的时间动态和因果依赖关系。它包括交错的操作数据和具身推理问答数据，共17个子集。数据集结合了人工标注和VLM生成的注释，确保了质量和可扩展性。数据来源于多个机器人平台的210万次机器人操作，涵盖了多种任务和场景。

EO-Data-1.5M is a large-scale interleaved vision-text-action dataset designed for embodied AI. It is part of the EO-Robotics toolchain and is used for training generalist robot policies. The dataset emphasizes temporal dynamics and causal dependencies among vision, language, and action modalities. It includes interleaved manipulation data and embodied reasoning QA data, totaling 17 subsets. The dataset combines human-annotated data with VLM-generated annotations, ensuring both quality and scalability. The data is derived from 2.1M robot episodes across multiple platforms, covering a wide range of tasks and scenarios.