Method On Tracking And Shooting AUV Path Planning Based On Deep Reinforcement Learning

The vast sea area is not only rich in resources,but also an important channel for trade and cultural exchanges.Vigorously developing Marine equipment is a necessary means to improve China’s maritime construction and governance capacity.Autonomous Underwater Vehicle(AUV),as a kind of unmanned vehicle for exploring and developing underwater space,can accomplish various scientific investigations and engineering tasks excellingly,and has a wide range of application prospects and important research value.In the field of observation-level AUV,its small size and agility and other characteristics play a difficult role in the tracking and shooting of underwater environment,and how to achieve the ideal shooting effect is the key issue that needs to be paid attention to at present.This subject mainly focuses on the tracking and shooting process of AUV for underwater divers and other objects,and proposes a path planning method of AUV to ensure the shooting effect,which provides a theoretical basis for the feasibility of the path planning task of this type of AUV.Firstly,the research background and significance of this paper are introduced in the initial part,and the comprehensive analysis is carried out on the basis of the research status of common path planning algorithms at home and abroad.In view of the complexity and variability of Marine environment,the deep reinforcement learning algorithm is applied to the path planning problem of tracking shooting AUV.Based on this algorithm,the shooting effect and learning ability of tracking shooting AUV are improved and the environmental adaptability is improved.Secondly,the NAF-DQN algorithm was improved,the DNAF-DQN algorithm was proposed,and the path planning problem during AUV tracking shooting was studied.In view of the particularity of this subject,it is necessary to determine the path in real time according to the shooting image effect,so the deep reinforcement learning algorithm is applied to the path planning field,and the simulation verification is carried out.Then,to solve the problem of low learning efficiency of deep Q learning algorithm based on double normalized advantage function,this paper proposes an improved algorithm based on meta-reinforcement learning and adaptive learning rate.Meta-reinforcement learning is mainly used to speed up learning.By comparing and analyzing the experimental results,the proposed adjustment strategy can effectively improve the convergence speed of the original algorithm,and the convergence stability of the optimized algorithm is also improved.Finally,the concept and significance of the hardware-in-the-loop simulation platform are briefly described,and the hardware and software structure of the developed path planning system is described and analyzed.This paper describes the hardware-in-the-loop simulation test process and the test results of the algorithm in hardware-in-the-loop simulation,thus proving the reliability and stability of the path planning method proposed in this paper for tracking and photographing AUV.