Accurate Registration of Transparent Objects in 2D LiDAR SLAM

Gurtajbir Singh Herr, Lasitha Weerakoon, Nikhil Chopra

Abstract: Transparent objects and surfaces present a unique challenge to the application of LiDAR for indoor mobile robotics. This is due to the fact that the majority of the incident light (except for normal incidence) passes through such objects and results in them becoming “invisible” to the resulting map. In this work, we exploit the sharp intensity spikes that occur when an incident beam strikes a transparent surface perpendicularly, and keep an accurate account of such points. This enables us to make the representation unaffected by the drift accumulated as the robot moves through the environment. Our method utilizes the Pose Graph Optimization (PGO) framework and exhibits superior results when compared to an existing approach that utilizes intensity spikes for detecting transparent surfaces but uses a particle filter based approach for generating the map of the environment. We showcase how, by leveraging the inherent structure of a PGO framework, the transparent objects can be mapped accurately.