gtsam_points

This is a collection of GTSAM factors and optimizers for range-based SLAM.

Tested on Ubuntu 22.04 / 24.04 and CUDA 12.2, and NVIDIA Jetson Orin with GTSAM 4.2a9.

Factors

Scan Matching Factors

• IntegratedICPFactor & IntegratedPointToPlaneICPFactor
  The conventional point-to-point and point-to-plane ICP [1].
• IntegratedGICPFactor
  Generalized ICP based on the distribution-to-distribution distance [2].
• IntegratedVGICPFactor
  GICP with voxel-based data association and multi-distribution-correspondence [3][4].
• IntegratedVGICPFactorGPU
  GPU implementation of VGICP [3][4].
  To enable this factor, set -DBUILD_WITH_CUDA=ON.
• IntegratedLOAMFactor
  Matching cost factor based on the combination of point-to-plane and point-to-edge distances [5][6].

Colored Scan Matching Factors

• IntegratedColorConsistencyFactor
  Photometric ICP error [7].
• IntegratedColoredGICPFactor
  Photometric ICP error + GICP geometric error [2][7].

Continuous-time ICP Factors

• IntegratedCT_ICPFactor
  Continuous Time ICP Factor [8].
• IntegratedCT_GICPFactor
  Continuous Time ICP with GICP's D2D matching cost [2][8].

Bundle Adjustment Factors

• PlaneEVMFactor and EdgeEVMFactor
  Bundle adjustment factor based on Eigenvalue minimization [9].
• LsqBundleAdjustmentFactor
  Bundle adjustment factor based on EVM and EF optimal condition satisfaction [10].

Optimizers for GPU Factors

All the following optimizers were derived from the implementations in GTSAM.

• LevenbergMarquardtOptimizerExt
• ISAM2Ext
• IncrementalFixedLagSmootherExt

Nearest Neighbor Search

• KdTree KdTree with parallel tree construction. Derived from nanoflann.
• IncrementalVoxelMap Incremental voxel-based nearest neighbor search (iVox) [11].
• IncrementalCovarianceVoxelMap Incremental voxelmap with online normal and covariance estimation.

Continuous-Time Trajectory (Under development)

• B-Spline Cubic B-Spline-based interpolation and linear acceleration and angular velocity expressions [12].
• ContinuousTrajectory Cubic B-Spline-based continuous trajectory representation for offline batch optimization.

Installation

Install from source

# Install gtsam
git clone https://github.com/borglab/gtsam
cd gtsam
git checkout 4.2a9
 
mkdir build && cd build
cmake .. \
  -DGTSAM_BUILD_EXAMPLES_ALWAYS=OFF \
  -DGTSAM_BUILD_TESTS=OFF \
  -DGTSAM_WITH_TBB=OFF \
  -DGTSAM_BUILD_WITH_MARCH_NATIVE=OFF
 
make -j$(nproc)
sudo make install
 
# [optional] Install iridescence visualization library
# This is required for only demo programs
sudo apt install -y libglm-dev libglfw3-dev libpng-dev
git clone https://github.com/koide3/iridescence --recursive
mkdir iridescence/build && cd iridescence/build
cmake .. -DCMAKE_BUILD_TYPE=Release
make -j$(nproc)
sudo make install
 
## Build gtsam_points
git clone https://github.com/koide3/gtsam_points
mkdir gtsam_points/build && cd gtsam_points/build
cmake .. -DCMAKE_BUILD_TYPE=Release
 
# Optional cmake arguments
# cmake .. \
#   -DBUILD_DEMO=OFF \
#   -DBUILD_TESTS=OFF \
#   -DBUILD_WITH_CUDA=OFF \
#   -DBUILD_WITH_MARCH_NATIVE=OFF
 
make -j$(nproc)
sudo make install

Install from <a href="https://github.com/koide3/ppa">PPA</a> [AMD64, ARM64]

Setup PPA

Ubuntu 24.04

curl -s --compressed "https://koide3.github.io/ppa/ubuntu2404/KEY.gpg" | gpg --dearmor | sudo tee /etc/apt/trusted.gpg.d/koide3_ppa.gpg >/dev/null
echo "deb [signed-by=/etc/apt/trusted.gpg.d/koide3_ppa.gpg] https://koide3.github.io/ppa/ubuntu2404 ./" | sudo tee /etc/apt/sources.list.d/koide3_ppa.list

Ubuntu 22.04

curl -s --compressed "https://koide3.github.io/ppa/ubuntu2204/KEY.gpg" | gpg --dearmor | sudo tee /etc/apt/trusted.gpg.d/koide3_ppa.gpg >/dev/null
echo "deb [signed-by=/etc/apt/trusted.gpg.d/koide3_ppa.gpg] https://koide3.github.io/ppa/ubuntu2204 ./" | sudo tee /etc/apt/sources.list.d/koide3_ppa.list

Ubuntu 20.04

curl -s --compressed "https://koide3.github.io/ppa/ubuntu2004/KEY.gpg" | gpg --dearmor | sudo tee /etc/apt/trusted.gpg.d/koide3_ppa.gpg >/dev/null
echo "deb [signed-by=/etc/apt/trusted.gpg.d/koide3_ppa.gpg] https://koide3.github.io/ppa/ubuntu2004 ./" | sudo tee /etc/apt/sources.list.d/koide3_ppa.list

Install GTSAM and gtsam_points

Without CUDA

sudo apt update && sudo apt install -y libgtsam-points-dev

With CUDA 12.2

sudo apt update && sudo apt install -y libgtsam-points-cuda12.2-dev

With CUDA 12.5

sudo apt update && sudo apt install -y libgtsam-points-cuda12.5-dev

Demo

cd gtsam_points
./build/demo_matching_cost_factors
./build/demo_bundle_adjustment
./build/demo_continuous_time
./build/demo_continuous_trajectory
./build/demo_colored_registration

Videos

• Multi-scan registration of 5 frames (= A graph with 10 registration factors)
• Bundle adjustment factor
• Continuous-time ICP factor
• Colored ICP factor
• Incremental voxel mapping and normal estimation
• SE3 BSpline interpolation

License

This library is released under the MIT license.

Citation

@software{gtsam_points,
  author       = {Kenji Koide},
  title        = {gtsam_points : A collection of GTSAM factors and optimizers for point cloud SLAM},
  month        = Aug,
  year         = 2024,
  publisher    = {Zenodo},
  version      = {1.0.4},
  doi          = {10.5281/zenodo.13378352},
  url          = {https://github.com/koide3/gtsam_points)}}
}

Dependencies

• Eigen
• nanoflann
• GTSAM
• [optional] OpenMP
• [optional] CUDA
• [optional] iridescence

Disclaimer

The test data in data directory are generated from The KITTI Vision Benchmark Suite and The Newer College Dataset. Because they employ Creative Commons BY-NC-SA License 3.0 and 4.0, the test data must not be used for commercial purposes.

References

[1] Zhang, "Iterative Point Matching for Registration of Free-Form Curve", IJCV1994
[2] Segal et al., "Generalized-ICP", RSS2005
[3] Koide et al., "Voxelized GICP for Fast and Accurate 3D Point Cloud Registration", ICRA2021
[4] Koide et al., "Globally Consistent 3D LiDAR Mapping with GPU-accelerated GICP Matching Cost Factors", RA-L2021
[5] Zhang and Singh, "Low-drift and real-time lidar odometry and mapping", Autonomous Robots, 2017
[6] Tixiao and Brendan, "LeGO-LOAM: Lightweight and Ground-Optimized Lidar Odometry and Mapping on Variable Terrain", IROS2018
[7] Park et al., "Colored Point Cloud Registration Revisited", ICCV2017
[8] Bellenbach et al., "CT-ICP: Real-time Elastic LiDAR Odometry with Loop Closure", 2021
[9] Liu and Zhang, "BALM: Bundle Adjustment for Lidar Mapping", IEEE RA-L, 2021
[10] Huang et al, "On Bundle Adjustment for Multiview Point Cloud Registration", IEEE RA-L, 2021
[11] Bai et al., "Faster-LIO: Lightweight Tightly Coupled Lidar-Inertial Odometry Using Parallel Sparse Incremental Voxels", IEEE RA-L, 2022
[12] Sommer et al., "Efficient Derivative Computation for Cumulative B-Splines on Lie Groups", CVPR2020