We introduce a broad-phase contrinuous collision detection algorithms designed to take advantage of modern GPU hardware. Our apporoach is based on a simple and suprising experimental observation that brute force checking all collision pairs is, surprisingly, much faster than competing approaches when executed on a modern GPU, due to the massive compute power and memory bandwidth. When paired with a simple splitting strategy to avoid the asymptotically quadratic computation cost, our algorithm is more efficient, both in term of times and memory, than 11 state of the art methods on a large set of representative scenes based on the work of [Wang et al].
翻译:我们引入了一种广泛阶段的相撞探测算法,旨在利用现代GPU硬件。我们的相邻算法基于一个简单而令人惊讶的实验观察,即由于巨大的计算功率和内存带宽度,在现代GPU上执行时,粗力检查所有碰撞对子的粗力比竞争性方法要快得多,令人惊讶。当我们采用简单的分解策略以避免非现四边计算成本时,我们的算法在时间和记忆方面比以[Wang 的工作为基础的大量具有代表性的场景的11种先进方法效率更高。