Reinforcement Learning Based Reciprocal Decision‐Making in Multi‐Player Pursuit‐Evasion Differential Games

ABSTRACT This paper investigates reciprocal decision‐making in multi‐player pursuit‐evasion (MPE) differential games by analyzing altruistic decision‐making among players. The irrational decision‐making motivated by altruism is modeled by introducing a distance term between cooperative players and common adversaries into the original performance function. Based on the new performance function, the Nash policy under irrationality is first sought using the maximum principle, based on which reinforcement learning is proposed to approximate the Nash policy. Subsequently, sufficient conditions are proposed to determine whether irrational decision‐making is egoistic or altruistic and whether reciprocity is generated under altruistic decision‐making among players. Finally, the effectiveness of the proposed results is verified by a numerical example.