registration¶
How each new scan is aligned to the map. This covers the voxel grid the map is stored in and the iterated extended Kalman filter update that matches a scan against it.
| Option | Default | Starter |
|---|---|---|
max_iter |
4 |
|
neighbors_max_range |
5m |
|
map |
||
resolution |
50cm |
yes |
resolution.type |
Multi |
|
capacity |
262144 |
yes |
distribution_update_sensitivity |
10.0 |
registration.max_iter¶
Maximum filter correction iterations per scan.
Type: integer.
Minimum: 0 (inclusive).
registration.neighbors_max_range¶
Largest distance at which a map voxel is still considered a neighbor of a point. The default is fine unless you reduce map.resolution by a lot.
Type: length.
registration.map¶
registration.map.resolution¶
Edge length of one voxel, which is also the size of the smallest feature registration can resolve. This is the option with the largest effect on both accuracy and runtime, so change it before anything else.
Type: length.
Part of the starter configuration.
registration.map.resolution.type¶
Single: one fixed-resolution map. Multi: scales of 0.5x, 1x, 2x and 4x resolution.
Accepted values: single, multi
Accepted values are case-insensitive.
registration.map.capacity¶
How many voxels of the resolution above the map holds at any time. Once it is full, older voxels are dropped, so memory stays bounded on a long run. With CUDA enabled and Single resolution this is rounded up to the next power of two. The multiresolution map keeps the exact total instead.
Type: integer.
Minimum: 0 (inclusive).
Part of the starter configuration.
registration.map.distribution_update_sensitivity¶
How strongly a new scan moves the distribution stored in a voxel. Higher values track changes in the environment faster and keep less history.
Type: number.