Tiltrotors Position Tracking Controller Design Using Deep Reinforcement Learning

摘要

Abstract In this paper, a quad-tiltrotors air-water trans-domain robot is introduced. The nonlinear dynamic behaviours with uncertainties require a robust controller for multi-tasks. For this robot, controllers are designed using deep reinforcement learning method solving position and attitude control when operating as a UAV in the air. A ROS combining Gazebo simulation platform is designed to train the robot. The simulation results show the tiltrotors robot gets capabilities of spots tracking as a quad-rotors, and trajectory tracking as both the quad-rotors and tiltrotors.