Meta Reinforcement Learning Based Underwater Manipulator Control

Abstract

Robots have garnered significant attention owing to their advantages in terms of replacing human labor under hazardous environments. In particular, because underwater construction robots can perform various tasks that are highly dangerous under deep sea environments, the development of manipulator control technology for these underwater robots is crucial. In this study, we therefore introduce an underwater manipulator control method based on meta reinforcement learning. Specifically, we construct a real-world underwater robot manipulator environment using ROS Gazebo and conduct simulations for the testing and verification of the proposed method.