Multi-Agent Environment

The dataset used in the paper is a multi-agent environment where agents learn to coordinate their actions to achieve a common goal. The dataset is used to evaluate the proposed method of using a generative model to infer missing observations during decentralized execution.