A Proportional Navigation Based Reinforcement Learning Approach: Efficiency Enhancements and Practical Applications

Vast Proportional Navigation Guidance (PNG) Artificial Intelligence (AI)-based approaches were presented last decade. The approaches are sorted to Supervised and Reinforcement learning (RL). Q-learning has been utilized excessively in the pursuer-target interception scenarios due to its relevancy. That is because Q-learning maps the proportional navigation guidance states Line of Sight (LOS) rate to actions (acceleration) via the learned policy. However, optimal policy and energy efficiency cannot be achieved due to the Q-learning discretization constraint. This study proposed a novel approach for mitigating that constraint by shaping the reward function. The proposed function incorporates a smoothing reward for actions in addition to the distance-based component. The conducted experiments reveal the optimal policy improvement which minimizes the flight energy.