Feature detection and matching for collaborative multi-robot SLAM

This thesis presents a novel framework for multi-robot Collaborative Simultaneous Localization and Mapping (C-SLAM) based on distributed optimization and keypoint-based localization. The system leverages an incremental version of the Alternating Direction Method of Multipliers (ADMM), known as Incremental Manifold Edge-based Separable ADMM (iMESA) iMESA, to efficiently solve pose graph optimization problems in a decentralized and communication-limited setting. The main contribution of this work is the integration of FALKO (Fast Adaptive Laser Keypoint Orientation-invariant) FALKO, a 2D LIDAR-based algorithm for detecting and describing stable geometric features in laser scans. FALKO enables accurate and repeatable keypoint extraction, allowing robots to estimate relative poses without relying on visual markers or external infrastructure. Keypoints are matched across agents using viewpoint-invariant descriptors, providing a robust basis for data association and inter-robot localization. The proposed method is implemented and validated in a simulated multi-robot environment, demonstrating its effectiveness in marker-less scenarios. Experimental results show that FALKO improves localization accuracy, feature stability, and overall map consistency, offering a scalable and vision-free solution for collaborative SLAM.