Figure 1.

Learning and generalization phases of our proposed MoP-SAN and baseline methods. (A) Ego agent and the specific partner agent collaborate to complete the Overcooked task in the learning phase. (B) In the generalization phase, the ego agent needs to collaborate with some unseen partner agents to test generalization ability (zero-shot collaboration). (C) This figure shows our proposed MoP model in the generalization phase. By constructing a MoP model, we can first estimate the partner's personality by the personality estimator and predict the actions of the partner by the multi-personality network according to the personality of the partner. Two agents in the same kitchen in all three graphs represent the cooperative relationship between the two agents to complete this cooking task.