Abstract: Deep reinforcement learning (DRL) can significantly improve the autonomy and effectiveness of air combat maneuver decision (ACMD). The design of reward functions faces significant challenges ...