Struct GicpRegistration 

pub struct GicpRegistration { /* private fields */ }

Generalized ICP registration.

Implementations

impl GicpRegistration

pub fn align_with_gpu_reduction( &self, runtime: &WgpuRuntime, source: &PointCloud, target: &PointCloud, ) -> SpatialResult<GpuGicpReductionDiagnostics>

Retains source/target/covariances on GPU for transform, nearest matching, rotated covariance inverse, Hessian/gradient and Euclidean fitness passes. CPU solves normalized 6x6 equations and converts the linearized step to the existing world update/stopping convention. Each evaluation explicitly downloads 256 bytes; post-update equations feed the next solve/final fitness. Covariance preparation remains CPU k-NN or explicit radius-GPU readback and re-upload with full receipts. CPU k-NN rows are packed directly with bounded scoped workers for large native inputs; small/WASM inputs stay serial. Per-point neighbor and f64 accumulation order is preserved. No automatic CPU fallback occurs. Configuration/conditioning restrictions are stricter than legacy GICP; f32 weights, transforms and sums can change pose/matching/stopping.

pub fn align_with_gpu_resident_covariances( &self, runtime: &WgpuRuntime, source: &PointCloud, target: &PointCloud, ) -> SpatialResult<GpuGicpReductionDiagnostics>

Computes k-NN neighborhoods and plane covariances on GPU, retaining the covariance buffers for the existing resident equation/update loop. No neighborhood/covariance download or re-upload occurs. CPU builds both trees, solves normalized 6x6 equations and composes/stops the pose. Requires covariance_radius=None, k_neighbors=3..=64, epsilon (as f32) in 1e-4..=1, and the same valid configuration as align_with_gpu_reduction. Centered/scaled f32 covariance/eigenvectors may differ from CPU raw products, particularly at repeated eigenvalues and search boundaries. Invalid device preparation rejects equations without CPU fallback.

pub fn align_with_gpu_resident_radius_covariances( &self, runtime: &WgpuRuntime, source: &PointCloud, target: &PointCloud, neighbor_capacity: u32, ) -> SpatialResult<GpuGicpReductionDiagnostics>

Computes complete inclusive radius neighborhoods and covariances on GPU and retains them for the resident equation/update loop. Requires configured covariance_radius=Some(finite positive radius) with a normal f32 square, epsilon (as f32) in 1e-4..=1 and positive neighbor_capacity per point. Capacity includes self/duplicates; overflow fails without truncation or fallback. Fewer than three neighbors use epsilonI as in the grid route. Centered/scaled compensated f32 covariance and max-pivot Jacobi can differ from grid accumulation/sweeps, especially at repeated eigenvalues/gates. CPU builds both trees, solves 6x6 equations and composes/stops the pose. Upload is 32(S+T+source_nodes+target_nodes) bytes; downloads are only 256 bytes per evaluation. No neighborhood/covariance payload crosses CPU.

pub fn align_with_gpu_resident_grid_covariances( &self, runtime: &WgpuRuntime, source: &PointCloud, target: &PointCloud, neighbor_capacity: u32, ) -> SpatialResult<GpuGicpReductionDiagnostics>

Prepares radius covariances with CPU counting-sort grids and GPU gather, retaining the canonical grid’s raw f32 means/central sums/Jacobi sweeps. No source KD-tree, per-point neighborhood buffer or covariance readback is needed. Positive capacity includes self/duplicates; overflow rejects all equations. Configured radius/epsilon and normal-equation constraints match the resident-radius API. Dense grids must fit existing cell/device limits; no automatic sparse-path/CPU fallback occurs. CPU prepares grids and the target tree and solves 6x6 equations. Downloads remain 256 bytes per evaluation; all grid-index uploads are included in the receipt.

impl GicpRegistration

pub const fn new(config: GicpConfig) -> Self

Creates a GICP algorithm from config.

pub const fn config(&self) -> GicpConfig

Returns the config.

pub fn align_with_diagnostics( &self, source: &PointCloud, target: &PointCloud, ) -> SpatialResult<RegistrationResult>

Aligns source to target using Generalized ICP.

pub fn align_with_gpu_correspondences( &self, runtime: &WgpuRuntime, source: &PointCloud, target: &PointCloud, ) -> SpatialResult<GpuGicpDiagnostics>

Aligns with explicit GPU correspondence uploads/readbacks and a reused target. Covariance estimation uses CPU k-NN when covariance_radius is None, or explicit radius-GPU execution on this runtime when configured. Hessian/gradient reduction, pose solving, transforms and fitness sums remain CPU. All GPU errors propagate without the legacy CPU fallback. This explicit path requires finite input/transformed XYZ.

Trait Implementations

impl Clone for GicpRegistration

fn clone(&self) -> GicpRegistration

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for GicpRegistration

impl Debug for GicpRegistration

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PartialEq for GicpRegistration

fn eq(&self, other: &GicpRegistration) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl PointCloudRegistration for GicpRegistration

fn name(&self) -> &'static str

Human-readable registration algorithm name.

fn align( &self, source: &PointCloud, target: &PointCloud, ) -> SpatialResult<RegistrationResult>

Aligns source to target and returns the estimated transform.

impl StructuralPartialEq for GicpRegistration