EgoTeam
收藏资源简介:
EgoTeam数据集是由卡尔斯鲁厄理工学院、湖南大学等多机构联合创建的首个多机器人协作视觉问答数据集,旨在评估模型通过整合多个移动机器人的同步自我中心视频进行协作空间推理的能力。该数据集包含超过11.4万条问答对,覆盖19种问题类型、四个难度层级和三种团队规模,数据来源于Habitat和iGibson模拟器以及真实世界的四足机器人实验,总计约385.7小时的视频素材。数据创建过程涉及在模拟环境中部署多机器人团队生成探索轨迹与交互事件,并利用场景元数据和GPT-4o自动生成问答对,同时通过人工标注确保质量。该数据集主要应用于推动具身人工智能领域的发展,专门解决多机器人协同感知、跨视角关联、团队级场景理解以及动态空间推理等核心问题,为开发能够理解共享动态环境的智能系统提供了关键基准。
The EgoTeam dataset is the first multi-robot collaborative visual question answering (VQA) dataset jointly created by multiple institutions including Karlsruhe Institute of Technology (KIT) and Hunan University. It is designed to evaluate the capability of models to perform collaborative spatial reasoning by integrating synchronized egocentric videos from multiple mobile robots. Boasting over 114,000 question-answer pairs, the dataset covers 19 question types, four difficulty levels, and three team sizes. The data is sourced from Habitat and iGibson simulators as well as real-world quadruped robot experiments, with a total of approximately 385.7 hours of video footage. The dataset construction process involves deploying multi-robot teams in simulated environments to generate exploration trajectories and interaction events, automatically generating question-answer pairs using scene metadata and GPT-4o, and ensuring data quality through manual annotation. This dataset primarily serves to advance the field of embodied artificial intelligence (Embodied AI), specifically addressing core challenges including multi-robot collaborative perception, cross-view alignment, team-level scene understanding, and dynamic spatial reasoning, providing a critical benchmark for developing intelligent systems capable of understanding shared dynamic environments.




