基于视觉的AUV自主水下管线跟踪方法
2021-05-09王昊颜承昊任俊丽邵思扬
王昊 颜承昊 任俊丽 邵思扬



摘 要:对海下管线的检测是保证其正常运行的必要环节。文章针对海下管道巡检问题,提出一种结合状态表示学习和深度强化学习的方法,使水下机器人能够基于图像进行管线跟踪。利用无监督表征学习方法提取海底管线图像特征,结合任务情况设计了动作、状态空间和奖励函数,通过SAC算法学习一个稳健的跟踪控制策略。最后搭建仿真环境进行试验,证明了所提出方法的有效性和泛化性。
关键词:AUV;海底管线跟踪;状态表示学习;深度强化学习
中图分类号:TP242 文献标识码:A文章编号:2096-4706(2021)20-0016-05
Vision-based AUV Autonomous Underwater Pipeline Tracking Method
WANG Hao, YAN Chenghao, REN Junli, SHAO Siyang
(School of Information Engineering, Dalian Ocean University, Dalian 116023, China)
Abstract: The detection of subsea pipelines is a necessary link to ensure its normal operation. For the subsea pipeline inspection problem, this paper proposes a method combining state representation learning and deep reinforcement learning, so that the underwater robot can track the pipeline based on image. The unsupervised representation learning method is used to extract the image features of submarine pipelines. Combined with the task situation, the action, state space and reward function are designed. A robust tracking control strategy is learned through SAC algorithm. Finally, a simulation environment is built to test the effectiveness and generalization of the proposed method.
Keywords: AUV; subsea pipeline tracking; state representation learning; deep reinforcement learning
0 引 言
海底管道、线缆是各国之间实现石油、天然气等资源运输的重要手段。作为海底基础设施,海底管线在国际信息传输、海上资源开发、海岛开发等领域发挥着重要作用。海底管线一般距离长,影响范围大[1]。近年来,随着海洋资源的不断开发和海底管线的持续建设,海底管线的数量和安装深度也在不断增加[2]。海底管线长期运行于复杂的海底环境中,受海水冲刷、压力等因素影响,容易形成疲劳损伤导致破裂、泄露[3-5],海底管线的稳定性极大地保证了其在整个运营周期内的安全和经济效益。因此,定期对海底管线进行监测和检查是一个必不可少的环节,而监测过程中的层层障碍使其成为一项极具挑战性的任务。
当前,我国海底管线主要铺设于水下200~300米这一区间,导致一般的人工潜水检测无法实现[6],通常采用遥控水下机器人(Remote Operated Vehicle, ROV)和自主水下机器人(Autonomous Underwater Vehicle, AUV)进行作业[7-9]。……
