Another paper accepted at NeurIPS 2021: Agent Modelling under Partial Observability for Deep Reinforcement Learning.