A Q-Network Based Terrain-Following Method for Aircrafts

The ability of terrain following supports aircrafts to break through enemy's defense in low-altitude penetration missions, and many researchers have proposed their own terrain-following algorithms. However, because of the unknown environment and computational complexity, it is difficult to apply these approaches in real world. A Q-Network based self-learning terrain-following method is presented in this paper, which combines reinforcement learning and neural network, enabling aircrafts to make decisions online with raw sensor data. Besides, heuristic information is also used to improve the traditional \epsilon-greedy policy, which makes a better balance between exploration and exploitation. At last, simulation illustrates the effectiveness of this method.