Efficient Global Occupancy Mapping for Mobile Robots using OpenVDB

TL;DR

This paper introduces a fast, optimized occupancy mapping method for mobile robots using the OpenVDB data structure, enabling real-time map building at high sensor framerates with customizable quality.

Contribution

It presents a novel, multithreaded occupancy mapping approach leveraging OpenVDB, significantly improving speed and flexibility over traditional log-odds methods.

Findings

Achieves high-speed occupancy mapping suitable for modern sensors

Demonstrates effectiveness through ablation studies and benchmarks

Validates system on legged robot and UAV platforms

Abstract

In this work we present a fast occupancy map building approach based on the VDB datastructure. Existing log-odds based occupancy mapping systems are often not able to keep up with the high point densities and framerates of modern sensors. Therefore, we suggest a highly optimized approach based on a modern datastructure coming from a computer graphic background. A multithreaded insertion scheme allows occupancy map building at unprecedented speed. Multiple optimizations allow for a customizable tradeoff between runtime and map quality. We first demonstrate the effectiveness of the approach quantitatively on a set of ablation studies and typical benchmark sets, before we practically demonstrate the system using a legged robot and a UAV.