SocialAI School

该数据集是一个包含定制参数化程序生成环境的工具，用于进行关于社会认知能力的实验。它不仅促进了针对强化学习智能体在多代理互动中推断手势意义和能力角色的实验，还提供了多种规模的环境，以适应不同的任务，例如箱子、开关、杠杆、弹珠和发电机等。该数据集的任务包括研究强化学习智能体对指向手势的理解、角色反转模仿以及支架式学习的实验。

This dataset is a tool comprising custom parametric procedurally generated environments for experiments on social cognitive abilities. It not only facilitates experiments in which reinforcement learning (RL) agents infer the meaning of gestures and competence-related roles during multi-agent interactions, but also provides environments of varying scales to accommodate diverse tasks such as boxes, switches, levers, marbles, generators and more. The tasks encompassed by this dataset cover experiments that examine reinforcement learning agents' comprehension of pointing gestures, role-reversal imitation, and scaffolded learning.