基于起伏地形的果园仿真环境搭建与SLAM算法测试  

Orchard simulation environment construction based on undulating terrain and SLAM algorithm testing

作　　者：朱泓逸 李秋洁[1] ZHU Hongyi;LI Qiujie(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037,China)

机构地区：[1]南京林业大学机械电子工程学院,南京210037

出　　处：《林业工程学报》2024年第2期133-142,共10页Journal of Forestry Engineering

基　　金：国家重点研发计划(2022YFD2202105)。

摘　　要：针对目前仿真果园环境研究不考虑地形起伏的问题,基于二维正态分布函数生成具有起伏特征的地面模型,并搭建果园仿真环境测试基于2D激光雷达的同步定位与建图(simultaneous localization and mapping, SLAM)算法的性能。通过控制土地模型的起伏变化程度、最大高低落差以及突起分布密度等参数生成不同地形特征的土地;通过2D激光雷达、里程计等传感器获取移动机器人在果园仿真模型中的观测数据;通过可视化SLAM定位偏差及SLAM建图效果评价4种经典2D激光SLAM算法(Hector SLAM,GMapping, Karto SLAM,Cartographer)在具有不同地形的果园环境中的性能。实验结果表明:在果园环境中,随着土地起伏变化程度及最大高低落差的增加,2D激光SLAM算法的定位性能与建图性能将会降低;Hector SLAM能够提供更精准的定位结果,但是建图鲁棒性较差;GMapping能够获得更准确的环境地图,但是定位鲁棒性较差;Cartographer的定位及建图的鲁棒性均较为良好,但会出现少量偏差;Karto SLAM相较于其他算法,在果园环境中不具备优势。Terrain relief is an essential aspect of environment perception for agriculture mobile robots to operate in orchard environments.For those robots obtaining surrounding environment through 2D LiDAR(Light Detection and Ranging),terrain relief could cause the pitching and folding motions of 2D LiDAR in three-dimensional space,which would bring false information into further calculations of agriculture robots poison in orchard world and result in an unreliable localization and map building result.Since the terrain model is hardly constructed in any existing orchard simulation environment either in the research of SLAM algorithms or navigation field,this study constructed an orchard simulation environment utilizing the terrain model with undulating features based on the 2D normal distribution function and tested it with 2D laser-based SLAM algorithms.This specific model could be generated by different fluctuation degrees,maximum altitude intercept,and density of protrusion distribution to simulate the random fluctuation characteristics of real orchard terrain.During experiments,2D LiDAR,odometry,and other sensors were used to obtain the mobile robot s movement data.Four classical 2D laser-based SLAM algorithms(Hector SLAM,GMapping,Karto SLAM,and Cartographer)were tested in orchard simulation environments with different terrain models which contracted with different terrain parameters.Localization offsets and mapping outcomes were two key aspects of evaluating SLAM results.Offset data,localization trail and global map were provided to evaluate the performance of SLAM algorithms.The experiment results which conducted in 13 different orchard simulation environments with different terrain models showed that the localization and mapping performance of the 2D laser-based SLAM algorithm decreased with the increase of fluctuation degree and maximum altitude intercept.The projection distribution density did not affect the localization accuracy or the map building ability as large as fluctuation degree and maximum altitude inte

关 键 词：果园环境 2D SLAM 激光雷达 Gazebo仿真 

分 类 号：S24[农业科学—农业电气化与自动化] TP23[农业科学—农业工程]