Leveraging 3-D Data for Whole Object Shape and Reflection Aware 2-D Map Building

Abstract

Two-dimensional laser scan sensors stand out as the preferred choice for robot mapping applications. However, these sensors have a significant drawback. Encountering objects with varying shapes at different heights, such as tables, poses challenges for these sensors due to their limited detection capability resulting from their dimensionality. This limitation increases the risk of potential collisions. Additionally, there are multiple polished materials that generate noise due to reflection. In order to have a robust occupancy grid map representation, these problems must be addressed. This paper proposes the usage of a 3D laser scan sensor to generate a 2D occupancy grid map that incorporates the complete geometry of objects and effectively filters out noise from reflective materials like glass. The main novelty of the method is that it takes advantage of all the available 3D data to avoid any information loss about objects' shapes. Additionally, a new approach for filtering reflection noise based on the analysis of indoor structural elements is proposed. Both approaches are merged for the creation of a robust indoor representation that allows to safely navigate the environment. Finally, a recursive Bayesian filter is applied for merging data, so noise due to dynamic elements that appeared during data collection is also filtered. Experimental evaluations in indoor environments with diverse objects and reflective surfaces, including dynamic elements like people, demonstrate the effectiveness of the proposed approach.

ADAM: a robotic companion for enhanced quality of life in aging populations

“…A sensor profile is extracted for every 3D scan that is captured, so for every data capturing step, there is a corresponding 2D sensor profile. Then, 2D data is merged using a recursive Bayesian filter modeled as a Markov Random Field of order 0 to create the final map, meaning that each cell in the map is estimated as an independent variable (Mora et al, 2023a). However, geometric maps are difficult for people to interpret.…”