A deep reinforcement learning approach incorporating genetic algorithm for missile path planning

The flight path planning of the missile is important in long-range air-to-ground strike missions. Constraints about missile guidance and guidance handover are considered, and path planning is required to conform to the missile motion model. Therefore, the missile’s allowable flight space and flight mode are further restricted, and the decision-making scale and difficulty of the path planning problem are significantly increased. A genetic algorithm incorporated twin delayed deep deterministic policy gradient (GA-TD3) algorithm is proposed for missile path planning, which uses high-quality data generated by GA to improve the TD3 training effect. Firstly, a missile path planning model is established based on the missile’s motion equations, and the missile guidance and guidance handover constraints are stated in detail. Then a fast path generation method is proposed, which uses several leading points to generate a leading path based on the optimal control theory, and the genetic algorithm is used to improve the leading path quality. Finally, the deep reinforcement learning model for the missile path planning problem is established based on the TD3 framework, and the leading paths participate in the leading training to improve the training effect. Simulation cases of 4 threat areas and 3 guidance platforms demonstrate the efficiency of the GA-TD3. Furthermore, the influence of three factors on the algorithm’s performance is tested, including the leading path quality, leading path number, and leading training cycle.