Real-time collision culling of a million bodies on graphics processing units

Abstract

Figure 1: GPU-based collision culling for massive bodies. N -body collision detection for 1M arbitrarily moving boxes (first image), realtime simulation of 0.3M particles of random size on GPUs (second and third images), real-time rigid-body dynamics for 16K torus models of varying sizes on GPUs (fourth and fifth images). In these challenging benchmarks, our algorithm can find all the colliding bodies at interactive rates. AbstractWe cull collisions between very large numbers of moving bodies using graphics processing units (GPUs). To perform massively parallel sweep-and-prune (SaP), we mitigate the great density of intervals along the axis of sweep by using principal component analysis to choose the best sweep direction, together with spatial subdivisions to further reduce the number of false positive overlaps. Our algorithm implemented entirely on GPUs using the CUDA framework can handle a million moving objects at interactive rates. As application of our algorithm, we demonstrate the real-time simulation of very large numbers of particles and rigid-body dynamics.

A parallel constraint solver for a rigid body simulation

“…3 is a screenshot from a simulation of half-million rigid bodies on four GPUs. At the moment, the broad-phase collision detection is performed on the CPU, but we are going to integrate our GPU broad-phase collision detection to complete full rigid body simulation pipeline on the GPU [Liu et al 2010]. …”

Optimizing Pairwise Box Intersection Checking on GPUs for Large-Scale Simulations

“…To handle large datasets, the GPU-accelerating algorithms have been proposed in the literature [BULLET 2011;Le Grand 2007;Liu et al 2010;Lo et al 2011]. Le Grand [2007] employed a uniform partition strategy to separate boxes into different cells and sorted the cells to filter out empty ones.…”

“…The boxes that cross a certain number of cells, which are considered large boxes, are directly matched with one another without mapping them into cells. Liu et al [2010] presented an algorithm that applies a coarse partition strategy, which divides the space into subdivisions and then employs the sweep-line technique to find box intersections for each subdivision. Lo et al [2011] proposed a 2-D rectangle intersection algorithm on GPUs for overcrowded situations, in which many intersecting box pairs are reported.…”

“…Several GPU-based algorithms [BULLET 2011;Le Grand 2007;Liu et al 2010] have been proposed to accelerate broad-phase collision detection for graphics applications. These algorithms adopt the uniform partition strategy to fit the GPU execution; however, they do not address the issues that have arisen for large-scale datasets, such as the large output problem.…”

Virtual subdivision for GPU based collision detection of deformable objects using a uniform grid

“…In general, these approaches use a uniform grid whose cell size is object-independent [20], a uniform grid with an objectdependent cell size [15], a two-level hierarchical grid [7], or a general multilevel hierarchical grid [6]. Recently, several papers have focused on utilizing the GPU's parallelism to achieve high performance for CD by taking advantage of the features of spatial subdivision [12,15], and [7].…”