Suixin Shen, Melnikov S.X, Song Gao
As the battlefield environment becomes more and more complex, it is of great significance to study the game process of UAV to understand battlefield behavior. Therefore, the search for a UAV movement strategy has become the focus of research. In addition to the traditional strategy, deep reinforcement learning as a decision algorithm with self-learning ability has attracted much attention. In this regard, for the target tracking task of our UAV to the enemy UAV, deep reinforcement learning is used to train the tracking strategy for our UAV. In order to find the most suitable deep reinforcement learning algorithm for UAV target tracking, a target tracking model was established, and the four algorithms were used for training, and the indexes of online training and the results of offline execution were compared. Finally, Dueling Double Deep q-learning and Proximal Policy Optimization achieved the best training effect and completed the target tracking task.
Deep Reinforcement Learning; Target tracking; Machine Learning;