gDist: Efficient Distance Computation between 3D Meshes on GPU

TL;DR

This work introduces a highly parallel algorithm (gDist) optimized for Graphics Processing Units (GPUs), which is capable of computing the distance between two meshes with over 15 million triangles in less than 0.4 milliseconds.

Abstract

Computing maximum/minimum distances between 3D meshes is crucial for various applications, i.e., robotics, CAD, VR/AR, etc. In this work, we introduce a highly parallel algorithm (gDist) optimized for Graphics Processing Units (GPUs), which is capable of computing the distance between two meshes with over 15 million triangles in less than 0.4 milliseconds (Fig. 1). By testing on benchmarks with varying characteristics, the algorithm achieves remarkable speedups over prior CPU-based and GPU-based algorithms on a commodity GPU (NVIDIA GeForce RTX 4090). Notably, the algorithm consistently maintains high-speed performance, even in challenging scenarios that pose difficulties for prior algorithms.

Figures (18)

• Figure 1: Distance Bounds for AABBs: Compared to the conventional bound computation (a), our new method returns a tighter bound (b). The upper bounds for the minimum distance are highlighted in red.
• Figure 2: BVTT expansion: We traverse all BVTT nodes in the input front $\rm buffer_1$ and, based on the input adaptive depth step, obtain the expanded BVTT front $\rm buffer_2$.
• Figure 3: F12-BVH: We propose f12-BVH, a full binary tree with 1 or 2 triangles on each leaf node, designed for enhanced GPU performance.
• Figure 4: Performance Comparison: we conduct a comprehensive comparison with several optimized CPU algorithms (PQP Larsen14, FCL Pan12, and SSE Shellshear14) and a 'naïve' GPU implementation (based on BVHCD18) (quantified in milliseconds). Across all benchmarks, our GPU algorithms consistently exhibit remarkable speedups, notably outperforming the tested algorithms on an NVIDIA GeForce RTX 4090.
• Figure 5: Performance on GPUs: We rigorously evaluate the performance (quantified in milliseconds) of our GPU-based algorithm on three commodity GPUs, establishing its efficacy across diverse GPU architectures.