Lidar Attenuation Through a Physical Model of Grass-like Vegetation

Abstract

Off-road autonomous vehicles face a unique set of challenges compared to those designed for road use. Lane markings and road signs are unavailable, with soft soils, mud, steep slopes, and vegetation taking their place. Autonomy struggles with shrubbery, saplings, and tall grasses. It can be difficult to determine if this vegetation or what it obscures is drivable. Modeling and simulation of autonomy sensors and the environments they interact with enhances and accelerates autonomy development, but analytical models found in the literature and our in-house simulation software did not agree on how well lidar penetrates grass-like vegetation. To test our simulator against the analytical model, we constructed vegetation mock-ups that conform to the assumptions of the analytical model and measured the pass-through rate on calibrated lidar targets. Vegetation density, lidar-to-vegetation distance, and target reflectivity were varied. A random effects model was used to address the dependence introduced by repeated measures, which increased accuracy while reducing time and cost. Stem density impacted total beam return count and grass patch pass-through rate. Target reflectivity results varied by lidar unit, and three-way factor interaction was significant. Results suggest benchmarking experiments could be useful in autonomy development.