Flexible Use of Temporal and Spatial Reasoning for Fast and Scalable CPU Broad‐Phase Collision Detection Using KD‐Trees

Abstract

Realistic computer simulations of physical elements such as rigid and deformable bodies, particles and fractures are commonplace in the modern world. In these simulations, the broad‐phase collision detection plays an important role in ensuring that simulations can scale with the number of objects. In these applications, several degrees of motion coherency, distinct spatial distributions and different types of objects exist; however, few attempts have been made at a generally applicable solution for their broad phase. In this regard, this work presents a novel broad‐phase collision detection algorithm based upon a hybrid SIMD optimized KD‐Tree and sweep‐and‐prune, aimed at general applicability. Our solution is optimized for several objects distributions, degrees of motion coherence and varying object sizes. These features are made possible by an efficient and idempotent two‐step tree optimization algorithm and by selectively enabling coherency optimizations. We have tested our solution under varying scenario setups and compared it to other solutions available in the literature and industry, up to a million simulated objects. The results show that our solution is competitive, with average performance values two to three times better than those achieved by other state‐of‐the‐art AABB‐based CPU solutions.Realistic computer simulations of physical elements such as rigid and deformable bodies, particles and fractures are commonplace in the modern world. In these simulations, the broad‐phase collision detection plays an important role in ensuring that simulations can scale with the number of objects. In these applications, several degrees of motion coherency, distinct spatial distributions and different types of objects exist; however, few attempts have been made at a generally applicable solution for their broad phase. In this regard, this work presents a novel broad‐phase collision detection algorithm based upon a hybrid SIMD optimized KD‐Tree and sweep‐and‐prune, aimed at general applicability. Our solution is optimized for several objects distributions, degrees of motion coherence and varying object sizes. These features are made possible by an efficient and idempotent two‐step tree optimization algorithm and by selectively enabling coherency optimizations.