You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Implement voxel grid downsampling in Awkernel to reduce point cloud size before NDT scan matching, matching the behavior of the selected Autoware reference implementation.
Scope:
Configurable voxel dimensions
Voxel indexing and representative-point computation consistent with the reference implementation
Explicit handling of additional point fields, invalid coordinates, and output ordering
Integration with the Awkernel point cloud preprocessing pipeline
Acceptance criteria:
Output voxel membership and representative-point coordinates match the reference within predefined tolerances.
Tests cover empty inputs, negative coordinates, voxel boundaries, and multiple points within a voxel.
Timestamp and coordinate-frame metadata are handled consistently with the defined interface.
Output ordering is either matched or explicitly treated as irrelevant in comparisons.
Execution time, memory usage, and point-count reduction are measured across representative inputs.
Test datasets, parameters, and reproduction steps are documented.
Activity
romancardenas commentedon Nov 17, 2026
After #803 is merged, I can work on this issue.
Reference implementation: https://github.com/autowarefoundation/autoware_core/blob/16045061a9c10da468b60e190b8fab02110fa501/sensing/autoware_downsample_filters/src/voxel_grid_downsample_filter/faster_voxel_grid_downsample_filter.hpp