Fast and accurate 3D global localization using branch-and-bound scan matching.
Please refer to our paper.

The latest implementation demonstrates faster processing times than those published in the paper.
Specifically, when tested in a real environment with the following hardware configuration (Intel Core i7-10700K 3.8GHz, 32GB RAM, and NVIDIA GeForce RTX2060), the processing times are as follows:

• Hierarchical voxelmap construction
  • Paper: 9,272 ms on average
  • Latest: 3,494 ms on average
  • Use saved voxelmap: 130 ms on average
• Localize
  • Paper: 878 ms on average
  • Latest: 189 ms on average

• bbs3d (Lower versions are not tested)
  • Eigen3 version 3.4.0 or higher
  • CMake version 3.15 or higher
  • GPU version: CUDA version 12.0 or higher
• test
  • (All bbs3d dependencies)
  • PCL
• ros2 test
  • (All bbs3d dependencies)
  • ros2 humble

If nvidia driver is 525.60.11 or higher, try docker!
For more information, you can check docker_start.md

git clone https://github.com/KOKIAOKI/3d_bbs.git
cd 3d_bbs
mkdir build && cd build

• CPU ver. & GPU ver. (Please ignore the large number of warnings)

cmake .. -DCMAKE_BUILD_TYPE=Release
make -j8
sudo make install

• CPU ver. only

cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_CUDA=OFF
make -j8
sudo make install

See test_code.md

See ros2_test_code.md