Hide and Seek Game States

该数据集包含了从捉迷藏游戏中收集的视觉观察和游戏状态，其中代理通过识别自身和寻找者的可见性来学习如何避免被捕捉。该数据集用于分析代理在学习识别寻找者和自身可见性方面的特征，从而洞察代理的弱点如何能促进更丰富的特征学习。数据规模涵盖了从50,000个游戏步骤中收集的信息，这些步骤中代理的能力各不相同。任务包括对寻找者识别和自我可见性意识进行二分类任务。

This dataset comprises visual observations and game states collected from a hide-and-seek game, where agents learn to evade capture by detecting the visibility of both themselves and the seekers. This dataset is designed to analyze the features of agents during their learning of seeker and self-visibility recognition, aiming to provide insights into how agents' weaknesses can foster richer feature learning. The dataset includes information collected from 50,000 game steps, in which agents exhibit diverse capabilities. The downstream tasks involve binary classification for seeker recognition and self-visibility awareness.