Modelling the Dynamic Joint Policy of Teammates with Attention Multi-agent DDPG https://arxiv.org/abs/1811.07029