最近开题给的题目是不规则地形slam ,但是现在有的传感器只有一个双目和一个2d雷达和imu,而且都是那种很便宜的传感器,所以先查找论文,找找方案,然后看看后面怎么搞。
1.A SLAM with simultaneous construction of 2D and 3D maps based on Rao-Blackwellized particle filters
传感器:wheel odom,2d laser scan(rplidar),depth camera(kinect v1)
环境: indoor
思路:生成两个地图。利用lidar建立一个grid map,再根据RBPF算法,利用depth camera的数据生成一个octmap。lidar和camera 应该是 松耦合。
2.A Visual SLAM System with Laser Assisted Optimization
(International Conference on Advanced Intelligent Mechatronics)
传感器:2d laser scan(rplidar),kinect 2,
环境:indoor
思路:一篇不咋地的论文,但是论文思路很独特。利用orbslam2的框架,把lidar的位姿作为camera的初始值,然后进行orbslam2的过程。
3.Visual-lidar Odometry and Mapping: Low-drift, Robust, and Fast(vloam)
(icra 2015)
中文解读:https://www.it610.com/article/1188358095743590400.htm
传感器:单目,3d lidar()
思路:cmu zhang ji大神的另一篇文章,单目和3d lidar紧耦合的VLO。
4.Multi-camera visual SLAM for off-road navigation
(Robotics and Autonomous Systems)
构建的大尺度稀疏地图(框架类似于orb)
思路:1)主要针对时间同步问题,提出了两种方案:帧到帧的线性插值。2)高斯牛顿进行优化。
2)针对越野环境阳光直射、遮挡树叶、道路粗糙、传感器故障、稳定可跟踪纹理稀疏等问题,采用多相机协作的方式。
5.IMU-assisted 2D SLAM method for low-texture and dynamic environments
(MDPI applied science)
传感器:2d激光(UTM-30LX-EW),imu(ADIS16365),laptop(core i7-6)
思路:前端imu+lidar,后端loop closure采用SBA。是将雷达的地图和真实值进行对比。并没有ground truth。
6.Terrain estimation for high-speed rough-terrain autonomous vehicle navigation
一篇关于起伏地形的动力学建模的论文。
7.Mapping forests using an unmanned ground vehicle with 3D LiDAR and graph-SLAM
(Computers and Electronics in Agriculture)
思路:一篇利用VLP-16,IMU,GPS,stereo camera对树林进行三维重建的论文。
8.Lidar-IMU Fusion for 2.5D Mapping
(2018 ROBIO)
传感器:2d激光,imu
思路:imu获得高度信息,雷达获取平面激光信息,采用粒子滤波的方法,构建2.5d地图
9.Robot-Centric Elevation Mapping with Uncertainty Estimates(https://core.ac.uk/download/pdf/148004368.pdf)
(2014 Climbing and Walking Robots (CLAWAR) )
代码:https://github.com/ANYbotics/elevation_mapping
思路:ETH的神作,一个四组机器人,以自身为中心,建立实时局部3D地图,在草地和农田 里做的实验,效果很不错。
10.River mapping from a flying robot: state estimation, river detection, and obstacle mapping
思路:cmu的大作,构建河流的3d地图
11.Real-time 3D mapping using a 2D laser scanner and IMU-aided visual SLAM
(Conference: 2017 IEEE International Conference on Real-time Computing and Robotics (IEEE RCAR 2017) )
传感器:2d激光,imu,rgbd camera
思路:国防科技大学的一篇论文,把2d激光竖起来,然后通过相机估计位姿,将每一帧的雷达点云拼起来。思路很强。
12.Visual-LiDAR odometry aided by reduced IMU
(MDPI)
传感器:VLP-16 ,IMU,GPS,
思路:把一些lidar odometry ,例如GICP,NDT等和 visual odometry的方法,例如SBA,RANSAC,bucking等进行了总结。还有sensor fusion的方法进行了总结。bucking算法,在论文Visual Odometry based on Stereo Image Sequences with RANSAC-based Outlier Rejection Scheme 中有详细描述(这篇文章感觉含金量很高),感兴趣的可以去看下。
13.Real-time Depth Enhanced Monocular Odometry
(IROS 2014)
思路:视觉里程计用RGBD或者激光进行深度增强。