1. CULLIDE: Interactive Collision Detection Between Complex Models in Large Environments using Graphics Hardware Presented by Marcus Parker By Naga K. Govindaraju, Stephane Redon, Ming C. Lin and Dinesh Manocha http://gamma.cs.unc.edu/CULLIDE

2. Collision Detection on GPUs 2 categories: – Depth/stencil buffer techniques for computing interference – Fast computation of distance fields for proximity queries Limitations: – Bandwidth Issues – Closed Objects – Multiple Object-Pair Culling

3. Contributions Potentially Colliding Set (PCS) – Initially objects, then sub-objects Exact Collision Detection Advantages: – Simple, no assumption about input model – Handles “polygon soups”, deformable and breakable objects – No pre-computation or data structures

4. Previous Work Collision Detection – 2 phases: broad (culling) and narrow (primitive tests) Acceleration using GPUs – Accelerate geometric computations – Applies to rigid & deformable models – Other interference algorithms are limited

5. PCS n triangulated objects: O 1,…,O n PCS – Objects that are overlapping or in close proximity Lemma 1: – An object O does not collide with a set of objects S if O is fully-visible with respect to S. – Sufficient, but not necessary, condition

6. Visibility Based Pruning Lemma 2: – Given n objects O 1,…,O n, an object O i does not belong to the PCS if it does not intersect with O 1,…,O i-1,O i+1, …,O n, 1 ≤ i ≤ n. This test can be easily decomposed as : an object O i does not belong to the PCS if it does not intersect with O 1,…,O i-1 and with O i+1, …,O n, 1 ≤ i ≤ n. 2 pass rendering approach Avoids frame-buffer readbacks Computes less conservative PCS than other algorithms

7. Localizing Overlapping Features Decompose objects to sub-objects Lemma 3: – Given n objects and each object O i is composed of mi sub-objects T 1 i, T 2 i,…,T mi i, a sub-object T k i of O i does not belong to the object’s potentially intersecting region if it does not intersect with the sub-objects of O 1,…,O i-1,O i+1,…,O n, 1 ≤ i ≤ n. This can decomposed similarly to Lemma 2. No checks for self-intersections or self- occlusion

8. Collision Detection Object Pruning – AABBs, then triangulated representation Sub-Object Pruning Exact Collision Detection (Intersection Tests)

9. Pruning Algorithm 2 pass rendering algorithm Use Lemma 2 to compute the PCS 1 st pass: Render objects in the order O 1,…,O n 2 nd pass: Render objects in reverse order (O n,…,O 1 ) Check if object is fully-visible with respect to objects rendered before it An object fully-visible in both passes is not in the PCS

10. Visibility Queries Overall algorithm based on hardware visibility query Some modifications are necessary to better use the existing tests Bandwidth Requirements – Doesn’t require frame buffer readbacks

11. Optimizations AABBs and Orthographic Projections Visibility Query returning Z-fail Avoid stalls Rendering Acceleration

12. Environments

13. Breaking Objects

14. Performance

17. Conclusions & Limitations Makes no assumption about object motion Doesn’t readback frame-buffer or depth- buffer Only checks overlapping objects (no distance/penetration information) Currently can’t handle self-collision