ACE-MAPPO: A Hybrid Framework for Cooperative Air Combat

To tackle the difficulties in multi-agent reinforcement learning for cooperative air combat beyond visual range, researchers have introduced the Adversarial Curriculum and Evolutionary-enhanced Multi-agent Proximal Policy Optimization (ACE-MAPPO). This innovative framework combines evolutionary algorithms with MAPPO, enhancing exploration efficiency, sample usage, and policy generalization. It features a genetic soft update mechanism that promotes population diversity and prevents local optima, alongside an evolutionary-augmented prioritized trajectory replay strategy that increases sample efficiency. The focus of this research is on enabling autonomous decision-making for unmanned combat aerial vehicles operating in complex, adversarial settings.