Collection of multi-robot SLAM algorithms for ROS
Simulation enviroment:
- Robot - Turtlebot 3
- Navigation stack - move_base
- Simulator - Gazebo
- World model - willowgarage.world
All algorithms use LIDAR as sensor (with or without odometry) and grid-maps.
Main package mrslam_sim contains all required launch files. Look inside those files for additional info. Some simulations require modified versions of packages. So, it's recommended to use my forks of this repositories.
Package range_ctrl simulates connection interruptions between robots with specified range, if simulation runned inside one master-process. Default connection range is 10 meters.
Package msg_collector collects messages from specific for every algorithm ROS topics, when connection between two robots lost. When connection established again, it sends all saved messages. It's very useful addition for those algorithms, for which data lose is critical.
You can control robots inside RViz by selecting green arrows on top panel and pointing on the map.
Install GCC 7 as default compiler, older versions aren't supported by some packages. Also install some dependences from sources:
Then run:
git clone https://github.com/MaxGsomGsom/Multi-robot-SLAM-experiment
cd Multi-robot-SLAM-experiment
rosdep install --from-paths . --ignore-src --rosdistro=kinetic
catkin_make
1.1. Simple map merging algoritm map_merging. Just superimposes all maps. G-mapping used as base SLAM (Rao-Blackwellised particle filter).
roslaunch mrslam_sim m_gmapping_mapmerging.launch
1.2. The same as 1.1. but can be run on different hosts with separate copies of gazebo and ROS.
Robots connections provided by multimaster_fkie package.
Connections between hosts can established with ad-hoc WIFI network.
Run src/mrslam_sim/scripts/adhoc_net.sh on every host to setup WIFI adapters in Ad-hoc mode.
Setup SSH keys from main computer to secondary (src/mrslam_sim/scripts/install_ssh_keys.sh) to allow connection without password.
Don't forget to enter correct IP's in scripts.
Launch of simulation separated on 3 steps, because of some problems with time syncronization. You should execute it on main computer:
./src/mrslam_sim/scripts/mm_gmapping_mapmerging_step1.sh
./src/mrslam_sim/scripts/mm_gmapping_mapmerging_step2.sh
./src/mrslam_sim/scripts/mm_gmapping_mapmerging_step3.sh
2.1. Map merging algorithm m-explore/map_merge based on OpenCV library matchers. Does not require knowledge of initial positions of robots. G-mapping used as base SLAM (Rao-Blackwellised particle filter).
roslaunch mrslam_sim m_gmapping_mexplore_mapmerge.launch
2.2. The same as 2.1. but can be run on different hosts with separate copies of gazebo and ROS.
Robots connections provided by multimaster_fkie package.
Connections between hosts can established with ad-hoc WIFI network.
Run src/mrslam_sim/scripts/adhoc_net.sh on every host to setup WIFI adapters in Ad-hoc mode.
Setup SSH keys from main computer to secondary (src/mrslam_sim/scripts/install_ssh_keys.sh) to allow connection without password.
Don't forget to enter correct IP's in scripts.
Launch of simulation separated on 3 steps, because of some problems with time syncronization. You should execute it on main computer:
./src/mrslam_sim/scripts/mm_gmapping_mexplore_mapmerge_step1.sh
./src/mrslam_sim/scripts/mm_gmapping_mexplore_mapmerge_step2.sh
./src/mrslam_sim/scripts/mm_gmapping_mexplore_mapmerge_step3.sh
- Map merging algorithm m-explore/map_merge based on OpenCV library matchers. Does not require knowledge of initial positions of robots. Hector-SLAM used as base SLAM (scan-matching).
roslaunch mrslam_sim m_hectormapping_mexplore_mapmerge.launch
- Graph-SLAM algorithm based on G2O framework cg_mrslam. Merges overlaped parts of map online based on condensed laser scans. After executing it is possible to merge full maps. To perform it first merge output graphs (look for *.g2o files in working dir) with g2o_merger (part of cg_mrslam package). Then open resulting file in g2o_viewer (part of G2O framework) and optimize it.
roslaunch mrslam_sim m_cgmrslam.launch
- Graph-SLAM algorithm based on Karto mapping library nav2d_karto. Performs full map merging online based on raw laser scans. Initial positions of robots must be known.
roslaunch mrslam_sim m_nav2dkarto.launch
- ICP-based no-odometry SLAM ohm_tsd_slam. Performs full map merging online based on raw laser scans. Initial positions of robots must be known.
roslaunch mrslam_sim m_ohmtsdslam.launch
- Graph-SLAM algorithm based on MRPT framework mrpt_graphslam_2d.
Merges parts of maps by requesting chains of graph nodes of other robots.
Integrated with multimaster_fkie, so it is not possible to run simulation inside one master process.
To run it on one PC you need setup virtual network adapters in you system in
/etc/network/interfaces(you can find example of config insrc/mrslam_sim/scripts/interfaces) and restart networking service. Don't forget to enter correct IP's in scripts. Then run launch script for every robot simultaneously. Robots should start from one position because map-matching dont work for now (some bugs).
./src/mrslam_sim/scripts/ms_mrpt_graphslam2d_robot0.sh &
./src/mrslam_sim/scripts/ms_mrpt_graphslam2d_robot1.sh
- Rao-Blackwellised particle filter SLAM algorithm based on MRPT framework mrpt_rbpf_slam. Performs full map merging online based on raw localized laser scans. Initial positions of robots must be known.
roslaunch mrslam_sim m_mrpt_rbpf.launch
- ICP-based no-odometry SLAM algorithm based on MRPT framework mrpt_icp_slam_2d. Performs full map merging online based on raw localized laser scans. Initial positions of robots must be known.
roslaunch mrslam_sim m_mrpt_icp2d.launch