AGENT
收藏资源简介:
AGENT是一个大规模的3D动画数据集,由麻省理工学院创建,旨在评估机器代理的核心心理学推理能力。数据集围绕四个核心场景构建,包括目标偏好、行动效率、未观察到的约束和成本-奖励权衡,这些场景旨在探究人类直觉心理学的关键概念。AGENT数据集通过程序生成的3D动画,模拟了代理在各种物理约束和对象交互中的移动,旨在测试机器学习模型对这些关键情境的理解。该数据集的应用领域包括评估和改进机器代理在理解人类心理状态和行为方面的能力,特别是在社交互动和协作任务中。
AGENT is a large-scale 3D animation dataset developed by the Massachusetts Institute of Technology (MIT) for evaluating the core psychological reasoning capabilities of machine agents. The dataset is built upon four core scenarios: goal preference, action efficiency, unobserved constraints, and cost-reward trade-offs, which target the exploration of key concepts in human intuitive psychology. The AGENT dataset employs procedurally generated 3D animations to simulate agents' movements amid diverse physical constraints and object interactions, with the objective of testing machine learning models' comprehension of these critical scenarios. Potential applications of this dataset involve evaluating and improving the capacity of machine agents to comprehend human mental states and behaviors, especially within social interaction and collaborative task settings.




