SMACv2
收藏资源简介:
SMACv2是由牛津大学开发的一个改进的协作多智能体强化学习基准数据集。该数据集包含14个微观管理场景,通过程序内容生成(PCG)技术,确保了场景的随机性和部分观测性,从而要求智能体使用闭环策略。SMACv2旨在解决原始SMAC数据集的局限性,如缺乏随机性和有意义的部分观测性。数据集的应用领域主要集中在多智能体强化学习算法的测试和评估,特别是在需要复杂协作策略的场景中。
SMACv2 is an improved collaborative multi-agent reinforcement learning benchmark dataset developed by the University of Oxford. This dataset includes 14 micromanagement scenarios, which utilize procedural content generation (PCG) technology to ensure scenario randomness and partial observability, thus requiring agents to adopt closed-loop policies. SMACv2 is designed to address the limitations of the original SMAC dataset, such as the lack of randomness and meaningful partial observability. The primary application domains of this dataset focus on the testing and evaluation of multi-agent reinforcement learning algorithms, especially in scenarios that require complex collaborative strategies.




