Adapting SegMap: A LiDAR place recognition framework for standalone use in C++ applications

¹ Centre For Robotics and Future Production, CSIR, Pretoria, South Africa
² Electrical & Electronic Engineering, University of Cape Town, Cape Town, South Africa

^* Corresponding author: tmaweni@csir.co.za

Abstract

Autonomous mobile robots rely on accurate environmental mapping and continuous self-localisation for effective navigation, often achieved through complex algorithms that combine data from multiple sensors. Aru-SegMap is an adaptation of SegMap, a widely used 3D point cloud segment-based map representation, for modern ROS2-based and standalone C++ applications focused on localisation. SegMatch, a 3D point cloud segmentation and matching library integral to SegMap, reliably estimates a robot’s position and detects loop closures. This adaptation involved modularising the original library, decoupling it from a deprecated TensorFlow C++ API and ROS1, and integrating visualisation capabilities, enabling greater flexibility and usability for continued robotics research and development. Aru_SegMap was validated using datasets of varied agricultural environments. It is functional, produces consistent segments, and provides reliable localisation.