pointcloud_cuda_mapping

CUDA Optimized PointCloud based Mapping Repository

Introduction:

An Elevation mapping package using capabilities of GPU, Pointclouds and No-Copy Message Publishing from ROS.

Purpose:

• Use Stereocamera-generated RGB Pointclouds and the PointCloud Library as mediums of perception for mapping the environment
• Understand and use ROS Nodelets.
• Identify parallelizable parts of code.
• Use appropriate CUDA Libraries achieve desired parallelization.

Proposed Solution:

1. Use ROS Tf2 package for synchronization of odom/map-baselink transfroms.
2. Use ROS Nodelets for zero copy message passing between nodes.
3. Use Linear transformation to bring Pointcloud from Rover to Map frame.
4. Use cubalsSgemm() to accelerate the matrix multiplication.
5. Use grid_map_msgs for making the elevation map.

Testing:

Testing is has been done on Gazebo Simulations using the package provided by the European Rover Challenge 2022. The Module was also tested using ZED2i Stereocamera at 1920*1080 resolution. The results can be seen here