This is a ground aware version of cartographer(3D) project!
In this project, ground points are detected and used to construct ground-alignment constraints, which help to get better roll&pitch estimate.
To be more specific, an efficient online ground segmentation method is adopted to detect ground points for each scan, and a local map of ground points is maintained with ikd-Tree (中文解析) structure. During the correlative scan match, a scan-to-submap alignment is first conducted to adjust x, y, z and yaw, then a scan-to-local map manner ground plane alignment is conducted to optimize roll and pitch.
The performance of ground_aware_cartographer has been verified, the gif below shows the result of a 200*100 square meters dataset.
It is important to realize that taking ground constraint into consideration do help to reduce Z drift! See the picture below: the original Google cartographer suffers from severe z drift when passing through a valley-like area, while ground_aware_cartographer does not have that problem.
Picture below shows how local map works in low playback speed: it only contain ground pionts of nearby key frames, and always move with robot. With the help of ikd-Tree, maintaining such a local map is not expensive at all.
Please follow the official guidelines (Compiling Cartographer ROS) to install dependencies, since this project was developed totally based on the official Google Cartographer project.
Note That this project was developed and tested on ubuntu 16.04 only, so you are highly recommended to use the same ubuntu distribution.
Compared with original Google cartographer, compiling ground_aware_cartographer is a little bit different. The source code of cartographer has not been released in this repository, so you need to tell cmake to use pre-compiled cartographer package by appending "-DUSE_THIRDPARTY_CARTOGRAPHER=true".
catkin_make_isolated --install --use-ninja -DUSE_THIRDPARTY_CARTOGRAPHER=true
Before running the roslaunch command below, you should configure wgh_demo_backpack_3d.launch by yourself.
roslaunch cartographer_ros wgh_demo_backpack_3d.launch
The ground detection algorithm requires lidar elevation angle info and static transform info, however I haven't provided a way to configure these info yet, so currently ground_aware_cartographer cannot work with any public datasets.
To successfully run ground_aware_cartographer, you may have to used this home-made dataset:
2022-03-31-04-GateRoadDync.bag. Note that this dataset is just a simple sample to run ground_aware_cartographer, while it cannot fully show the performance and advantage of ground_aware_cartographer.
- Support public datasets.