Introdction

This code provides a reference implementation of method presented in the papers:

"Estimating Motion Uncertainty With Bayesian ICP"
F. Afzal Maken, F. Ramos, L. Ott
ICRA, 2020.
and

"Speeding up Iterative Closest Point Using Stochastic Gradient Descent"
F. Afzal Maken, F. Ramos, L. Ott,
ICRA 2019.

Requirements

The following libraries are needed to compile and run the code, the code is known to run on Ubuntu 16.04.

• boost
• eigen3
• pcl

Building

You will need CMake (at least version 3.1) to build the code.

cd sgd_icp
mkdir build
cd build
cmake -DCMAKE_BUILD_TYPE=Release
make

Running

Compiling the code will produce three executables:

• align_pcd
• pcl_align_pcd
• transform_cloud

align_pcd

This program uses the Bayesian ICP implementation to estimate the transform and distribution between two point clouds, the command line syntax is as follows:

align_pcd <source_cloud> <target_cloud> <cofig_file>

The configuration file, an example of which is provided in config.json, sets the various parameters of the algorithm, including the initial estimate of the transform.

pcl_align_pcd

This is a simple program which aligns the two provided points clouds using PCL with parameters similar to the default values used by align_pcd.

pcl_align_pcd <source_cloud> <target_cloud>

transform_cloud

This program applies a user specified transform onto a given cloud.

transform_cloud --input <source_cloud> --output <output_cloud> --pose <6D vector>