Unmanned aerial vehicle end-to-end control method based on deep reinforcement learning

Abstract

The invention belongs to the field of unmanned aerial vehicle autonomous control, and relates to an unmanned aerial vehicle end-to-end control method based on deep reinforcement learning. The invention discloses an unmanned aerial vehicle end-to-end control method based on deep reinforcement learning. The method comprises the steps of designing a deep reinforcement learning neural network; designing an end-to-end reward and punishment function rland for training unmanned aerial vehicle landing trajectory control of the designed deep reinforcement learning neural network; building an unmanned aerial vehicle landing scene in the simulation environment; and training the designed deep reinforcement learning neural network based on the simulation environment with the built unmanned aerial vehicle landing scene. According to the invention, the processed image obtained by the airborne camera of the unmanned aerial vehicle is used as input, and the unmanned aerial vehicle control instruction can be directly obtained after deep reinforcement learning network processing, so that autonomous landing of the unmanned aerial vehicle is realized.

Description

technical field [0001] The invention belongs to the field of autonomous control of unmanned aerial vehicles, and relates to an end-to-end control method for unmanned aerial vehicles based on deep reinforcement learning. Background technique [0002] The current problem of UAV autonomous landing control mainly uses hand-crafted geometric features and sensor data fusion to identify fiducial markers and guide the UAV to move in its direction. For example, the method disclosed in the document "Carlos Sampedro, Alejandro Rodriguez-Ramos, Ignacio Gil, Luis Mejias, Pascual Campoy. Image-Based Visual Servoing Controller for Multirotor Aerial Robots Using Deep Reinforcement Learning[J].IEEE.2018", which will firstly learn from The image obtained in Gazebo is converted into the state deviation between the UAV and the landing point, and then the control amount required for the movement of the UAV is obtained through the DDPG method, but this method is cumbersome; Continuous direction ...

AI Technical Summary

Problems solved by technology

For example, the method disclosed in the document "Carlos Sampedro, Alejandro Rodriguez-Ramos, Ignacio Gil, Luis Mejias, Pascual Campoy. Image-Based Visual Servoing Controller for Multirotor Aerial Robots Using Deep Reinforcement Learning[J].IEEE.2018", which will firstly learn from The image obtained in Gazebo is converted into the state deviation between the UAV and the landing point, and then the control amount required for the movement of the UAV is obtained through the DDPG method, but this method is relatively cumbersome; or, it is directly output through the image information. Continuous direction instructions, such as the literature "Riccardo Polvara, Massimiliano Patacchiola, Sanjay Sharma, Jian Wan, Andrew Manning, Robert Sutton, Angelo Cangelosi. Toward End-to-End Control for UAV Autonomous Landing via Deep Reinforcement Learning[J].IEEE.2018" A deep reinforcement learning based approach to UAV landing control is proposed based on a hierarchical structure of Deep Q-Networks (DQN), which are used as high-end control strategies for navigation in different phases, including front-back, left-right Control commands such as descent, but this method is less accurate

Images