Learning Transformer-based World Models with Contrastive Predictive Coding