Abstract: In view of the problem that the unmanned aerial vehicle （UAV）is unable to take reasonable countermeasures in the stochastic external environment during executing the mission in the air, an approach of UAV route following based on deep reinforcement learning is proposed. The quadrotor UAV dynamics model is found by force analysis of UAV and transformation of Euler angle. Under the framework of deep reinforcement learning, such relative factors of UAV as the coordinates, the Euler angles, the flight velocity, etc. are analyzed, the state space is fuzzified as the state input of deep reinforcement learning. Compared with the traditional method, the buile non-linear flight dynamics and dynemic model of the quadrotor UAV is more realistic. The simulation results show that the quadrotor UAV can perform the task of randomly generated route following with high efficiency and low error after continuous training.