Research On Motion Planning Of Autonomous Underwater Vehicle Based On Multi-constraint

Autonomous Underwater Vehicle(AUV)is an important tool for exploring the ocean.It can perform various tasks such as environmental monitoring,seabed top mapping,environmental assessment,pipeline inspection,target search,and underwater vehicle scientific research.For AUV,the ability of motion planning is an important embodiment of its intelligence.It runs through the AUV and is an important part of AUV Therefore,the research on motion planning technology has important and far-reaching significance.Based on the key research project(41412030201),this paper studies and designs a motion planning system based on multi-constraint.Using the deep reinforcement learning method,considering the sensor limits of AUV and the constraints of its actuators.In a mapless environment,the AUV can avoid obstacles and simultaneously reach the target's motion planning.The main work of this paper is as follows:(1)Based on a small underwater robot developed by Harbin Engineering University,this study refers to its sensor configuration and actuator capabilities,and analyzes and models the motion planning system.(2)The traditional reinforcement learning model is difficult to deal with the continuous action space.In this paper,a policy-based AUV motion planning system is designed and implemented.The deep reinforcement learning method is used to directly approximate the strategy and optimize the strategy to realize the continuous motion space planning of AUV.A more detailed planning effect can be achieved.In addition,for the AUV motion planning task requirements,with reference to the idea of curriculum learning,designed a curriculum for AUV motion planning training.The planning system was tested in a completely unknown environment to verify the feasibility.(3)It is difficult to design the deep reinforcement learning reward function of AUV motion planning system,because of easy to have unexpected solutions and reward sparseness in continuous state action space.This paper design and implement a reward function based on curiosity reward.The improvement method simulates human curiosity and encourages AUV to explore the unknown environmental states.The training process illustrates the advantages of curiosity reward in a larger state space.At the same time,the planning system was tested in a completely unknown environment to verify the feasibility of the system.(4)Considering the effect of the current on AUV,the simulation test of planning system with current interference in unknown environment was carried out.The simulation test verified that the AUV motion planning system proposed in this paper based on curiosity reward training has anti-interference ability to the current environment.