2021/03/23 by Ricardo Bedin Grando, Grando, Ricardo Bedin, Junior Costa de Jesus +13
Engineering · Computer Science · #Underwater Vehicles and Communication Systems #Robotic Path Planning Algorithms #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.2103.12883
Since the application of Deep Q-Learning to the continuous action domain in\nAtari-like games, Deep Reinforcement Learning (Deep-RL) techniques for motion\ncontrol have been qualitatively enhanced. Nowadays, modern Deep-RL can be\nsuccessfully applied to solve a wide range of complex decision-making tasks for\nmany types of vehicles. Based on this context, in this paper, we propose the\nuse of Deep-RL to perform autonomous mapless navigation for Hybrid Unmanned\nAerial Underwater Vehicles (HUAUVs), robots that can operate in both, air or\nwater media. We developed two approaches, one deterministic and the other\nstochastic. Our system uses the relative localization of the vehicle and simple\nsparse range data to train the network. We compared our approaches with a\ntraditional geometric tracking controller for mapless navigation. Based on\nexperimental results, we can conclude that Deep-RL-based approaches can be\nsuccessfully used to perform mapless navigation and obstacle avoidance for\nHUAUVs. Our vehicle accomplished the navigation in two scenarios, being capable\nto achieve the desired target through both environments, and even outperforming\nthe geometric-based tracking controller on the obstacle-avoidance capability.\n