3d mapper

[ugsfume]

This build completes the cone mapping function in the software pipeline (without SLAM) General Software Pipeline: Perception -> Cone Recognition + Cluster Detection -> Cone Mapping -> SLAM (optional for simulation) -> Path Planning -> Path Follower

Major Features:

• Transform cone poses from tracker node to be directly associated with the global frame.
• Append newly detected cones to an array, input to the path planning algorithm
• Allow GPU acceleration for object recognition

Minor Features:

• Created friendly workspace for gazebo world building
• Detected cone visualisation
• Launch files for grouping cone mapping nodes
• Gazebo spawns headless to reduce CPU load
• Removed unused Gazebo assets in repo

To Be Improved:

• Currently, cones are mapped/recorded with crude arbitrary threshold conditions. Accuracy can be improved immensely if a robust approach is implemented

Mapping Visualised:

Compared to Ground Truth:

Node Graph:

Instructions:

1. Go to docker-compose.yml to enable/disable GPU option
2. Rebuild Docker
3. roscore
4. roslaunch ackermann_vehicle_description hokuyo_odom.launch
5. roslaunch ultralytics_ros tracker_with_cloud.launch
6. roslaunch ultralytics_ros cone_mapper.launch
7. roslaunch ackermann_vehicle_navigation path_follower.launch

TODO:

• Incorporate path planning algorithm into the software pipeline (WIP based on this build)
• Implement SLAM
• Define mission specific behaviours
• Complete hardware build
• Adjust to hardware specific conditions (odometry, HardwareInterface, Linux and Nvidia environment setup)