The accuracy of autonomous navigation and obstacle avoidance of unmanned aerial vehicles (UAVs) in complex environments has become one challenging task. In this paper, an autonomous navigation and obstacle avoidance of the UAV (ANOAU) algorithm based on deep reinforcement learning (DRL) has been proposed to achieve accurate path planning in complex environments. In our work, we use an actor–critic-based DRL framework to achieve autonomous UAV control from sensor input to the output of the UAV’s action and design a set of reward functions that can be adapted to autonomous navigation and obstacle avoidance for the UAV in the complex environment. Meanwhile, to alleviate the decision-making bias caused by the incomplete observables of the UAV, we use a gate recurrent unit network to enhance the ability to perceive the uncertain environment, enhance the perception representation and improve the accuracy of UAV real-time decision-making. Experimental simulation results verify that the ANOAU algorithm achieves good UAV flight attitude adaptive adjustment in navigation and obstacle avoidance tasks and significantly improves the generalization ability and training efficiency of the UAV navigation controller in a complex environment.
Loading....