CART-SLAM: CUDA-Accelerated Real-Time SLAM

This is an implementation of SLAM using CUDA to accelerate processing. The current implementation focuses on identifying planes in the environment.

Running

The program is run from the command line, and has the following syntax:

./cart-slam <path-to-config>

or

./cart-slam <path-to-source-config> <path-to-modules-config>

Some sample configurations are provided in the config-folder.

Building

Use CMake to build the project. The following libraries are required:

• OpenCV
• CUDA

Command:

cmake -S . -B build

Optionally you can use -DCMAKE_CXX_COMPILER=<your compiler> and -DCMAKE_CUDA_COMPILER=<your CUDA compiler> to specify which compilers to use. A useful flag for debugging is -DCMAKE_BUILD_TYPE=Debug.

CMAKE Options

Available options are

• ENBALE_TIMING (default: OFF): Enable generation of timing csv files to timings-folder.
• ENABLE_SAVE_SAMPLES (default: OFF): Enable saving of samples to disk in samples-folder.
• ENABLE_RECORD_SAMPLES (default: OFF): Enable recording of samples to disk in samples-folder.

These can be set using -D<option>=<value> when running CMake.

ZedSDK on Manjaro

The ZedSDK is intended for Ubuntu. To get it working on Manjaro, you need to modify the installed CMake file (found at /usr/local/zed/zed-config.cmake).

Change the line:

set(LIB_PATH_64 "/usr/lib/${CMAKE_SYSTEM_PROCESSOR}-linux-gnu/")

to

set(LIB_PATH_64 "/usr/lib/")

Optionally also add the following line to suppress the warning spam about policy CMP0153:

cmake_policy(SET CMP0153 OLD)