Memory-Based Reinforcement Learning for Trans-Domain Tiltrotor Robot Control

Abstract

Aiming at the problems of motion control with high precision for a new type of air-water trans-domain tiltrotors, a deep reinforcement learning controller is applied to these conditions. Reinforcement learning algorithm with memory capability allows the robot to learn from dynamic information collected in the past. In this paper, the trans-domain tiltrotors are supposed operating as a quad-rotors with fixed-wing in the air. Moreover, simulation is based on ROS and Gazebo platform for training the reinforcement learning repeatedly and the results demonstrate this algorithm gets better accuracy and effectiveness compared with other non-current methods in the conditions of the tiltrotors control task.