基于路径规划特点的语义目标导航方法  

Object goal navigation based on path planning characteristics

作　　者：高宇[1] 霍静[1] 李文斌 伍静 来煜坤 高阳[1] GAO Yu;HUO Jing;LI Wenbin;WU Jing;LAI Yukun;GAO Yang(State Key Laboratory for Novel Software Technology,Nanjing University,Nanjing 210023,China;School of Computer Science and Informatics,Cardiff University,Cardiff CF103XQ,England)

机构地区：[1]南京大学计算机软件新技术全国重点实验室,江苏南京210023 [2]卡迪夫大学计算机科学与信息学院,英国威尔士卡迪夫CF103XQ

出　　处：《智能系统学报》2024年第1期217-227,共11页CAAI Transactions on Intelligent Systems

基　　金：科技部2030新一代人工智能项目(2021ZD0113303);国家自然科学基金项目(62276128,62192783,62106100);江苏省重点研发项目(BE2022077)。

摘　　要：为了解决语义目标导航任务中存在的探索效率低、深度不精准等问题,本文构建了一个解决语义目标导航任务的框架,在语义地图构建模块中引入了深度图边缘处理以及地图纠错机制;在探索模块中引入了覆盖范围最大化算法;在路径规划模块中引入了替代点机制。本文在一个3D仿真环境下进行了实验。实验结果表明,本文提出的解决方案明显提升了语义目标导航任务的性能。此外,本文所提方法成功应用到了四足机器人上,从而验证了其在现实场景下的泛化性。To solve the problems of low exploration efficiency and imprecise depth in object goal navigation,this article constructs a framework to solve object goal navigation.Depth map edge processing and map error correction mechanisms were introduced in the semantic map construction module;a coverage maximization algorithm was introduced in the exploration module;alternative point mechanisms was introduced in the path planning module.This article conducted experiments in a 3D simulation environment.The experimental results show that the new solution proposed in this article significantly improves the performance of object goal navigation.In addition,the method proposed in this article was successfully applied to quadruped robots,thereby verifying its generalization in real-world scenarios.

关 键 词：人工智能 视觉导航 语义目标导航 语义感知 语义探索 路径规划 机器学习 语义地图 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]