![]() |
Tesseract 0.28.4
|
Speed and accuracy of the Bullet signed distance field paths. More...
#include <tesseract/common/macros.h>#include <algorithm>#include <chrono>#include <array>#include <cmath>#include <cstdio>#include <cstdlib>#include <fstream>#include <functional>#include <iterator>#include <memory>#include <random>#include <string>#include <vector>#include <BulletCollision/CollisionShapes/btCollisionShape.h>#include <tinyvdb_io.h>#include <tinyvdb_nanovdb.h>#include <tinyvdb_sparse_tree.h>#include <tesseract/collision/bullet/bullet_utils.h>#include <tesseract/collision/bullet/tesseract_sdf_collision_shape.h>#include <tesseract/collision/implicit_sdf_collision_solver.h>#include <tesseract/geometry/geometries.h>#include <tesseract/geometry/impl/signed_distance_field_utils.h>Classes | |
| class | SignedDistanceField |
| A volumetric signed distance field geometry. More... | |
Enumerations | |
| enum class | SignedDistanceFieldInterpolation |
| How a grid-backed or sparse field is read between its samples. More... | |
Functions | |
| int | main (int argc, char **argv) |
Speed and accuracy of the Bullet signed distance field paths.
Compares the ways TesseractSdfCollisionShape can read a field in place (trilinear, tricubic Catmull-Rom, and split: trilinear distance with tricubic direction), for dense and sparse (NanoVDB) fields. Every query goes through the same helpers Bullet's SDF algorithms use (querySdfShapePoint, makeImplicitSdfShape + collideImplicitSDF), so only the field representation differs.
Accuracy is measured against exact analytic distances on curved surfaces (a sphere, a cylindrical bore) and sharp features (a box's edges and creases, and a 90 degree inside corner). The inside corner is the hardest of these: where two surfaces meet at 90 degrees, the distance has a crease along their bisector and its gradient flips across it. It is also where a tool often works, for example along the seam between two plates in welding.
Usage: sdf_interpolation_benchmark [spacing_m] (default 0.001) sdf_interpolation_benchmark field.nvdb (a real field, mm)
|
strong |
How a grid-backed or sparse field is read between its samples.
Both pass through the samples and reproduce a linear field exactly. They differ in how many samples a read uses and whether the gradient is continuous across cell faces.