Efficient 2D Graph SLAM for Sparse Sensing

This repository contains code for Efficient 2D Graph SLAM for Sparse Sensing. The paper can be accessed here.

Installation

We recommend Ubuntu 20.04 with ROS noetic, but it should also work on Ubuntu 18.04 with ROS melodic. Our code requires a C++14 capable compiler.

Prereqs

1. Install dependencies using package manager:

sudo apt install ros-noetic-jsk-rviz-plugins ros-noetic-navigation ros-noetic-joy

2. Follow the instruction on https://google-cartographer.readthedocs.io/en/latest/ to build and install google cartographer

3. Clone the libg2o-release repository (https://github.com/ros-gbp/libg2o-release) and

   We use this repo rather than the latest g2o because g2o comes with its own version of ceres-solver, which may conflict with the version used by cartographer. This makes sure that they will not conflict.

   • checkout release/{your-ros-distro}/libg2o branch
   • in config.h.in, make sure that the macro #define G2O_DELETE_IMPLICITLY_OWNED_OBJECTS 0 is enabled

   we manage memories for edges and vertices ourselves. You will get segfault if you miss this step.

   • in CMakeLists.txt, search for BUILD_WITH_MARCH_NATIVE and make sure it is ON

   You may get segfault if you miss this step

   • follow the readme to build and install it
   • run sudo ldconfig command to update links

4. Clone this repository and build

catkin_make -DCMAKE_BUILD_TYPE=Release

5. Download datasets and evaluation scripts by running

cd src/sparse_gslam/datasets
./download.sh

Running our code

roslaunch sparse_gslam log_runner.launch dataset:={dataset_name}

where {dataset_name} is one of the directory names under src/sparse_gslam/datasets (e.g. intel-lab).

Compute SLAM Metrics

For aces, intel-lab and mit-killian, ground truths are available and SLAM metrics can be calculated. After running our code with the command above and waiting for it to finish, you can use our evaluation script to compute the metrics

cd src/sparse_gslam/datasets
./eval.sh {dataset_name}