/SwarmMap

SwarmMap: Scaling Up Real-time Collaborative Visual SLAM at the Edge

Primary LanguageC++OtherNOASSERTION

SwarmMap

Watch the video

🚧 Under Construction

1. Related Publications

Xu, J., Cao, H., Yang, Z., Shangguan, L., Zhang, J., He, X., & Liu, Y. (2022). {SwarmMap}: Scaling Up Real-time Collaborative Visual {SLAM} at the Edge. In 19th USENIX Symposium on Networked Systems Design and Implementation (NSDI 22) (pp. 977-993).

2. License

SwarmMap is released under a GPLv3 license. For a list of all code/library dependencies (and associated licenses), please see Dependencies.md.

For a closed-source version of SwarmMap for commercial purposes, please contact the author(s):
admin (at) sense-lab (dot) org

If you use SwarmMap in academic work, please cite:

@inproceedings{xu2022swarmmap,
  title={$\{$SwarmMap$\}$: Scaling Up Real-time Collaborative Visual $\{$SLAM$\}$ at the Edge},
  author={Xu, Jingao and Cao, Hao and Yang, Zheng and Shangguan, Longfei and Zhang, Jialin and He, Xiaowu and Liu, Yunhao},
  booktitle={19th USENIX Symposium on Networked Systems Design and Implementation (NSDI 22)},
  pages={977--993},
  year={2022}
}

3. Installation

We have tested SwarmMap with Ubuntu 18.04/20.04, CentOS 7. We recommend using a powerful computer to ensure the performance of multi-agent SLAM.

3.1 Install dependencies

  • spdlog: log utility
  • Pangolin: visualization (tag: v0.5)
  • Eigen3: optimization-related
  • CUDA: ORB feature extraction speedup
  • OpenCV with cuda (3.4.6/4.2.0 tested)
  • Boost: serialization (version >= 1.70.0)
  • Boost beast: network communication

3.2 Build third-party libraries

  1. g2o:
cd SwarmMap/code/Thirdparty/g2o/ && mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release && make -j
  1. DBoW2
cd SwarmMap/code/Thirdparty/DBoW2/ && mkdir build && cd build
cmake .. -DCMAKE_BUILD_TYPE=Release && make -j

3.3 Build SwarmMap

  1. Clone the repository
git clone https://github.com/MobiSense/SwarmMap
  1. Build the project
cd SwarmMap/
mkdir build
cd build
# set your cuda path (tested version: 10.2)
cmake .. -DCMAKE_BUILD_TYPE=Release -DCUDA_TOOLKIT_ROOT_DIR=/usr/local/cuda 
make -j

4. Usage

4.1 Command line arguments

Server-client combined mode

Executable: swarm_map

SwarmMap - Scaling Up Real-time Collaborative Visual SLAM at the Edge:
  -h, --help                print this message
  -v, --voc arg             path to vocabulary
  -d, --dataset arg         path to dataset config file
  -l, --log arg (=debug)    log level: error/warn/info/debug
  -u, --viewer arg (=1)     use frame viewer
  -m, --mapviewer arg (=1)  use map viewer
  -c, --client arg (=2)     client number

Server-client standalone mode

  • Client - swarm_client 
    ./swarm_client --voc ../code/Vocabulary/ORBvoc.bin --log info --viewer 1 --mapviewer 1 --dataset PATH_TO_DATASE_FILE
  • Server - swarm_server 
    ./swarm_server --voc ../code/Vocabulary/ORBvoc.bin --log info --viewer 1 --mapviewer 1 --dataset PATH_TO_DATASE_FILE

4.2 Config file format

SwarmMap designs a config format to launch multi-agent SLAM, which specifies the following parameters:

mh01-03.yaml

%YAML:1.0 
# EuRoC MH01 - 03 (3 agents) dataset config file

# dataset type [euroc/tum/kitti]
TYPE: 'euroc'
# calibration and config file path
SETTING: '/Examples/Monocular/EuRoC.yaml'
# image sequnce path (each agent's sequence path)
IMAGES: ['/MH_01_easy/mav0/cam0/data', '/MH_02_easy/mav0/cam0/data', '/MH_03_medium/mav0/cam0/data']
# timestamps file path (each agent's timestamp file)
TIMES: ['/Examples/Monocular/EuRoC_TimeStamps/MH01.txt', '/Examples/Monocular/EuRoC_TimeStamps/MH02.txt', '/Examples/Monocular/EuRoC_TimeStamps/MH03.txt']

# SERVER HOST
HOST: 'xxx.xxx.xxx.xxx'
# port the server listens (when the config file is used by the server)
# port the client connects (when the config file is used by the client)
PORT: 1234

kitti00-02.yaml:

%YAML:1.0
# KITTI 00-02 (3 agents) dataset config file
TYPE: 'kitti'
SETTING: '/Examples/Monocular/KITTI00-02.yaml'
IMAGES: ['/kitti/sequences/00', '/kitti/sequences/01', '/kitti/sequences/02']
# SERVER HOST
HOST: 'xxx.xxx.xxx.xxx'
# port the server listens (when the config file is used by the server)
# port the client connects (when the config file is used by the client)
PORT: 1234

5. Examples

  • Remember to set the correct path to the dataset config file and vocabulary file
  • The vocabulary file SwarmMap used is in binary format with suffix .bin instead of the origin .txt
  • The program binary swarm_map is output to the folder bin

5.1 Example on the EuRoC dataset

cd SwarmMap/bin/
./swarm_map -d ../config/mh01-03.yaml -v ../code/Vocabulary/ORBvoc.bin -c 3 -l info

5.2 Example on the KITTI dataset

cd SwarmMap/bin/
./swarm_map -d ../config/kitti00-02.yaml -v ../code/Vocabulary/ORBvoc.bin -c 3 -l info

5.3 Example on the TUM dataset

cd SwarmMap/bin/
./swarm_map -d ../config/fr2-large12.yaml -v ../code/Vocabulary/ORBvoc.bin -c 2 -l info

5.4 Standalone mode

  • server

    # start the server first
./swarm_server --voc ../code/Vocabulary/ORBvoc.bin --log info --viewer 1 --mapviewer 1 --dataset PATH_TO_DATASE_FILE

  • client #1

    # start the first client
./swarm_client --voc ../code/Vocabulary/ORBvoc.bin --log info --viewer 1 --mapviewer 1 --dataset ../config/mh1.yaml

  • client #2

      # start the second client
./swarm_client --voc ../code/Vocabulary/ORBvoc.bin --log info --viewer 1 --mapviewer 1 --dataset ../config/mh2.yaml

5.4 Output Files

  • The maps generated by the clients are saved as map-client-<client_id>.bin
  • The maps generated by the server are saved as map-server-<client_id>.bin
  • The global map on the server is saved as map-global.bin
  • The trajectory of the agents is saved as KeyFrameTrajectory=CURRENT_TIME-<client_id>.txt and the format is TUM (<timestamp> <tx> <ty> <tz> <qx> <qy> <qz> <qw>)
  • Trajectories in TUM format can be directly evaluated using the evo evaluation tool.

6. Acknowledgements

SwarmMap is based on the following repositories: