Automatic Ground Collision Avoidance Control and Decision-Making of Fighter Base on Deep Reinforcement Learning

Control flight into terrain (CFIT) refers to the fighter crash or serious damage and casualties caused by unknown terrain obstacles, fog and other bad weather or other operational errors, and the flight crew control the aircraft crashing into the mountains, crashing into the ground or flying into the water. In order to eliminate the control flight into terrain accident of fighter, an automatic ground collision avoidance system (Auto GCAS) is proposed. By matching the predicted flight trajectory with the current terrain, warning and maneuvering evasion instructions are issued to control the fighter to evade terrain. In order to adapt to various large attitude states and complex terrain environment, a method of automatic ground collision avoidance control and decision-making for fighter based on deep reinforcement learning (DRL) is proposed. Considering the limitation of parameter optimization method of deep reinforcement learning network with stochastic gradient descent, especially when it often falls into local optimal solution, it can not reach global optimal solution. Therefore, the genetic algorithm (GA), which conforms to the principle of survival of the fittest and survival of the fittest, is introduced into the optimization method of the parameters of the deep reinforcement learning network to obtain the global optimal solution.