Author: Ruetz
Year: 2019
- sf On Visible Point Cloud Mesh, applied to rough terrain navigation, fullfil RT constraints
- Uses the General hidden point removal operator that reduces a PC to visible points from a certain viewpoint
- Octomap : uses an octree structure (slow map queries for insertion or lookup)
- Voxblox : voxel hashing strategy which allocates fixed size blocks in regions of the map that contain data
- On the contrary, this method does not rely on grid discretization
- On fast surface reconstruction methods for large and noisy point clouds reaches near RT
- Super slides on convex hull algorithms: https://tildesites.bowdoin.edu/~ltoma/teaching/cs3250-CompGeom/spring17/Lectures/cg-hull3d.pdf
- performs a radial transformation of every LiDAR point and forms a convex hull on it
- Use the gravity aligned world frame to compute the surface angle
$\alpha$ - A polygon is traversable if
$\alpha < \alpha_{max}$ and$\in [-\pi/2, \pi/2]$ , anfd if the maximum height difference between the vertices is smaller than a thresh - Generates a point cloud from the mesh with all the vertices as a map representation for path planning
- Use a robot centric point cloud buffer of 20-30 scans
- Use RRTConnect path planner
- Simulates noisy pc => normal estimation is better using OCPV mesh
- Three runs on rough terrain of several hectometers
- Has a smaller and less spreaded run time than octomap and elevation mapping
- able to handle hoverhang scenarios (roof)
- Very sensitive to outliers in LiDAR data => voxel filter needed
- Convex hull algorithm ensures a watertight mesh
@INPROCEEDINGS{ovpcmeshRuetz, author={Ruetz, Fabio and Hernández, Emili and Pfeiffer, Mark and Oleynikova, Helen and Cox, Mark and Lowe, Thomas and Borges, Paulo}, booktitle={2019 International Conference on Robotics and Automation (ICRA)}, title={OVPC Mesh: 3D Free-space Representation for Local Ground Vehicle Navigation}, year={2019}, volume={}, number={}, pages={8648-8654}, doi={10.1109/ICRA.2019.8793503} }