Figure 4.

Score comparison between the baseline and MoP-SAN models in the generalization and learning phases. The (left) figure shows that our MoP-SAN outperforms other baselines in terms of generalization performance, with a strong generalization ability to complete cooperative tasks with unseen partners. The (right) figure shows that our MoP-SAN improves the poor performance and large variance of SAN in the learning phase. Agents A-E denote different agents with different random seeds.