AVoE

AVoE数据集是由新加坡国家大学机械工程系和A*STAR信息通信研究所合作创建的大规模合成3D VoE数据集，旨在研究物理推理。该数据集包含25,000个试验，涵盖五个事件类别：支持、遮挡、包含、碰撞和障碍。每个事件类别进一步细分为多个子类别，展示基于不同场景刺激的物理变化。数据集通过随机抽样影响交互结果的多个物理刺激，增强了数据集的多样性。AVoE数据集的应用领域包括机器人助手、自动驾驶车辆和安全AI工具，旨在解决物理推理中的预期违反问题，提高AI系统的可解释性和效率。The AVoE dataset is a large-scale synthetic 3D VoE dataset co-developed by the Department of Mechanical Engineering of the National University of Singapore and the A*STAR Institute for Infocomm Research, targeting physical reasoning research. It contains 25,000 trials covering five event categories: support, occlusion, containment, collision, and barrier. Each event category is further subdivided into multiple subcategories, showcasing physical variations based on diverse scene stimuli. The dataset enhances its diversity by randomly sampling multiple physical stimuli that influence interaction outcomes. The AVoE dataset has applications in robotic assistants, autonomous vehicles and safety AI tools, aiming to address expectation violations in physical reasoning and improve the interpretability and efficiency of AI systems.