| Summary: | 碩士 === 國立中興大學 === 通訊工程研究所 === 107 === In recent years, the application of unmanned aerial vehicles (UAV) is prevalent and they are widely used in various fields. The unmanned aerial vehicles are also known as drones that can be used in disaster rescue, sports events, package delivery, warehouse management, etc. However, we are unable to control the drones precisely indoors due to the lack of global positioning system (GPS) signals. Therefore, we propose an algorithm based on Deep Reinforcement Learning (DRL) to control the drone, improving its accuracy on straight take-off, forwarding, and landing. Besides, we compare the impact of different sets of states and actions adopted by our DRL algorithm.
More specifically, we utilize an ArUco marker as a reference, and calculate the relative position between the drone and the ArUco Marker detected by the camera equipped by the drone. Then, we make the drone fly straightly toward or directly above the marker as much as possible. In particular, we conduct simulation experiments with the Gazebo simulator of Robot Operating System (ROS). The simulation results demonstrate that our proposed method is capable of improving the accuracy of the considered actions of a quadrotor.
|
|---|
