A research-oriented multi-agent reinforcement learning framework for autonomous air combat decision-making using value approximation, neural policy iteration, and curriculum learning.