costmap_2d is an ROS package in its navigation stack. In order to boost performance, three of its plugins are modified to use make use of CUDA for parallel processing.
This is also the source code of work mentioned in the paper CUDA-based parallel optimization of ROS costmap submitted to Computing Frontiers 2019. (But rejected unfortunately) We are here to make source code of modified costmap_2d and test dataset public.
If this optimized component or the idea itself makes you feel useful, please consider citing the following literature:
[1]刘海涛,缪晓伟. 一种使用GPU加速的代价地图快速更新方法[P]. CN109947118A,2019-06-28.
[2]缪晓伟. 无人车自主定位与导航研究[D].南京大学,2019.
- Make sure CUDA Toolkit and other necessary softwares are installed.
- Override the original version of
costmap_2dby following the instructions from Overlaying with catkin workspaces.
If multiple workspaces are in use, consider chaining catkin workspaces. - Manually compile this workspace. If all environments are corretly configured, it should be working properly.
- A computer running Linux and ROS
Only tested in Linux Ubuntu 16.04 LTS and ROS Kinetic Kame. - GPU that supports CUDA acceleration
CUDA Toolkit 8.0 is recommended.
Please refer to the Release page.