Curriculum-RL Based Air Combat Decision-Making

Deep reinforcement learning has been able to solve a wide range of complex decision-making tasks that were previously out of reach for a machine. But there are many challenges in using reinforcement learning to solve air combat decision-making problems. One of them is that rewards are sparse in the air combat and most of today's DRL algorithms struggle with such sparsity. Therefore, we propose a new method called Curriculum-RL which guide agent to learn a task from the simple to the complex. While solving the sparse problem, Curriculum-RL helps the agent to learn the air combat strategy better than the general weighted reward structure. In the end, we test our approach in the two-plane air combat and analyze the result.