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
{{ message }}
Repository navigation
Integrate Motion Distortion Correction Before Voxel Grid Downsampling #761
Integrate LiDAR motion distortion correction into the Awkernel preprocessing pipeline before downsampling stages that discard or aggregate per-point timing information. Ensure that NDT receives downsampled points expressed in a consistent scan reference frame.
Scope:
Connect motion distortion correction upstream of voxel grid downsampling
Preserve per-point timestamps and other required fields through all stages preceding correction
Define the scan reference time, coordinate frame, and metadata passed between stages
Integrate motion-data availability checks and the correction component’s failure policy
Prevent duplicate correction and accidental bypass of correction during normal operation
Acceptance criteria:
Pipeline dependencies enforce correction before voxel grid downsampling.
Tests verify that preceding filters preserve the original timing information of retained points.
Downsampling operates on corrected coordinates, and its output metadata identifies the common reference time and frame.
Invalid inputs or failed correction cannot silently proceed as successfully corrected data.
Integrated outputs match an explicit correction-then-downsampling reference sequence within predefined tolerances.
Tests include moving-sensor scans for which reversing the processing order produces a detectable difference.
Regression tests confirm correct behavior when crop box filtering is enabled or disabled.
Pipeline configuration, data contracts, and reproduction steps are documented.
Out of scope:
Changes to the distortion-correction or downsampling algorithms, reconstruction of timestamps already lost by earlier processing, and full NDT accuracy or performance benchmarking.
Activity