融合激光雷达构图与卫星地图的USV导航系统设计

Design of USV Navigation System Based on Lidar Image-Satellite Map Fusing

为解决水面无人舰(Unmanned Surface Vehicle,USV)定位与导航的问题,使用舰载激光雷达扫描实时数据点云图像,结合卫星地图,共同构建含有岸基线的环境水域图像,为USV提供可行区域或路径,设计控制系统,利用深度强化学习网络(Deep Q-Learning Network,DQN)在固定水域环境对系统进行学习训练,不断调整网络参数并优化经验池中的相关数据集,训练后的USV在陌生水域具有自主航行能力和避障能力。设计的USV系统经过训练学习后在不同水域环境中进行测试并与其他控制算法进行对比,测试结果表明:在宽阔水域的误差仅为3.05 m,在水流相对比较湍急的狭长区域则为3.81 m,均好于传统的航位推算法,对于大范围远距离的区域探索,尤其是实时性要求较高的任务,完全可满足精度要求。

The real-time point-cloud Image from shipborne Lidar scanning in conjunction with satellite map is used to construct the environment water area with shore baseline,which provides USV(Unmanned Surface Vehicle)with indication of feasible area or path,supports the design of control system.The USV navigation system with DQN(Deep Q-Learning Network)is trained in a set water environment to tune its parameters and optimize its experience pool.The trained USV is able to autonomously navigate in unfamiliar waters and avoid obstacles.Tests are carried out and good results achieved:error of 3.05 m in open water area and 3.81 m in narrow water area with swift current.