Exploration of Drone Trajectory Planning in Unknown Environments Using Reinforcement Learning

Trajectory planning is a very important problem in UAV base station deployment. The traditional path planning technology based on optimization theory needs to know the environmental information in advance, which cannot be accurately mastered in practice and cannot be applied to complex and changeable scenes. The reinforcement learning technology can make UAV interact with the environment without understanding the global information and then train and learn the optimal trajectory. In this paper, the trajectory problem of a single UAV in unknown user location and channel environment is studied, and the trajectory planning is realized by reinforcement learning method to maximize the throughput. In this paper, the Markov decision process is modeled for the trajectory planning problem of location environment information, and two flight stages are proposed. The actor-critic algorithm is used to solve the problem according to the environmental state and continuity of action of UAV. The result shows that UAV can realize trajectory planning in unknown environment.