Distance and Collision Detection

Transcription

1 Distance and Collision Detection Efi Fogel School of computer science, Tel Aviv University Fall 2003/4 Motion Planning seminar 1/33

2 The Papers A Fast Procedure for Computing the Distance Between Complex Objects in Three-Dimensional Space Computing Minimum and Penetration Distances between Convex Polyhedra Fall 2003/4 Motion Planning seminar 2/33

3 About the Algorithms First came Barr, Gilbert, and Wolfe. For example, Finding the nearest point in a polytope by Wolfe, In 1988 appeared A Fast procedure for computing the distance between... known as GJK. S. Cameron enhanced GJK in Computing Minimum and Penetration Distances between.... and also described modifications to compute penetration distances Lin & Canny, A fast Algorithm for incremental distance calculation, 1991 Fall 2003/4 Motion Planning seminar 3/33

4 Introduction In many fields (e.g., Robotics, CAD, Graphics, etc) it is important to know whether two objects in 3D intersect or are in close proximity. Fall 2003/4 Motion Planning seminar 4/33

5 Minimum Translation Distance (Cameron) When two simulated objects interpenetrate, we may need to know how to extricate the system from this condition. is in contact with objects overlap objects do not overlap Fall 2003/4 Motion Planning seminar 5/33

6 The Approach Compute the distance between convex sets in -dimensional space Efficient when Terminate after a finite number of iterations Linear in the total number of vertices Practical Fall 2003/4 Motion Planning seminar 6/33

7 Handled Object Shapes and Representations Objects that are the union of convex polytopes and their spherical extensions Spherical extensions are valuable May be used to cover an object with a safety shell Economical representations Fall 2003/4 Motion Planning seminar 7/33

8 Preliminaries The affine hull of a set, denoted by the intersection of all affine subspaces of containing., is The convex hull of a set, denoted by, is the intersection of all convex sets in containing. Fall 2003/4 Motion Planning seminar 8/33

9 Convex and Affine hulls in Fall 2003/4 Motion Planning seminar 9/33

10 Caratheodory s theorem Theorem 1 Let. Then each point of convex combination of at most points of. For example, in the plane, is the union of all triangles with vertices at points of. is a Fall 2003/4 Motion Planning seminar 10/33

11 The nearest point to the origin, nearest point in to origin Fall 2003/4 Motion Planning seminar 11/33

12 Translational C-space Obstacle (Cameron) Recognized as Minkowski Sum Fall 2003/4 Motion Planning seminar 12/33

13 Witness Points and are the witness points - realize the minimum distance Each is a surface point on and resp. Witness points are not necessary unique is the -witness point (Cameron) A surface point on Fall 2003/4 Motion Planning seminar 13/33

14 Tracking The distance algorithm is called many times in time steps Make sense to use the witness points found at the last step Fall 2003/4 Motion Planning seminar 14/33

15 Algorithm Sketch Finding the nearest point to the origin An example in Fall 2003/4 Motion Planning seminar 15/33

16 Algorithm Sketch Finding the nearest point to the origin An example in Fall 2003/4 Motion Planning seminar 15/33

17 Algorithm Sketch Finding the nearest point to the origin An example in Fall 2003/4 Motion Planning seminar 15/33

18 Algorithm Sketch Finding the nearest point to the origin An example in Fall 2003/4 Motion Planning seminar 15/33

19 The algorithm (Cameron) Require: is a compact convex set in _ while _ do _ end while computes the initial points returns true if the simplex contains the witness point, and false otherwise. computes a neighboring _ simplex Fall 2003/4 Motion Planning seminar 16/33

20 Inner (dot) Product The projection of whose length is between and. onto the unit vector, is the vector times the cosine of the angle, Fall 2003/4 Motion Planning seminar 17/33

21 Notations the support function of,, the support vertex, any witness of, Fall 2003/4 Motion Planning seminar 18/33

22 Minkowski Sum Fall 2003/4 Motion Planning seminar 19/33

23 Theorem be compact and convex, and by: Theorem 1 Let define. Then: Suppose Fall 2003/4 Motion Planning seminar 20/33

24 Theoretical Algorithm is compact and convex, Require: then 1: 2: 3: if 4: has d 5: stop 6: end if, where elements or less and satisfies 8: goto step 2. 7: Fall 2003/4 Motion Planning seminar 21/33

25 Distance Subalgorithm -th iteration, Consider the We need to compute: Fall 2003/4 Motion Planning seminar 22/33

26 Distance Subalgorithm, D.W. Johnson The number of all possible subsets of is: For example, In 4 vertices, 6 open edges, 4 open faces, 1 open simplex. Fall 2003/4 Motion Planning seminar 23/33

27 Distance subalgorithm. simplex an ordering of the subsets of. Define in be the complement of Let :, and Define real numbers Fall 2003/4 Motion Planning seminar 24/33

28 Distance Subalgorithm = 1 = 1 = 1 = = = = = = = = = Fall 2003/4 Motion Planning seminar 25/33

29 Theorem Theorem 1 if and only if, and 1., and 2., and Fall 2003/4 Motion Planning seminar 26/33

30 Distance Subalgorithm Require:, and an ordering 1: 2: if, and then, and 3: Stop 4: end if 5: if then 6: Increment and proceed to step 2 7: end if 8: Stop and report failure Fall 2003/4 Motion Planning seminar 27/33

31 Robustness Issues How reliable is it in the presence of roundoff errors Errors do not accumulate! Each iteration is recomputed based on Fall 2003/4 Motion Planning seminar 28/33

32 Making the Main Algorithm Robust Translate the origin to a point on the line segment joining the centroids of and Helps when is small and the is large Replace the convergence criterion to: related to the number-type accuracy Fall 2003/4 Motion Planning seminar 29/33

33 Making the Sub-Algorithm Robust The condition in the distance subalgorithm is not satisfied for any May happen when is affinely dependent or nearly so in all 4 points are nearly coplanar Resort to a backup procedure: Require: a simplex Compute the distance to all candidates {Compute for Return the best } Fall 2003/4 Motion Planning seminar 30/33

34 Hill Climbing (Cameron) Expediting the computing of the support vertex Given new support direction, and previous support vertex compare with for every vertex connected to if is not the smallest then, such that is the smallest else return end if Fall 2003/4 Motion Planning seminar 31/33

35 Solving each Simplex Fall 2003/4 Motion Planning seminar 32/33

36 Estimating Penetration Distance TCSO TC-space origin Objects overlap Fall 2003/4 Motion Planning seminar 33/33