Mobile robots must continuously estimate their position to navigate autonomously. However, satellite-based navigation systems are not always reliable: signals may degrade near buildings or become unavailable indoors. To operate safely and efficiently, robots must interpret their surroundings using onboard sensors and robust localization algorithms.
Researchers at Miguel Hernández University of Elche (UMH) in Spain have developed a hierarchical localization system that significantly improves robot positioning in large, changing environments. The method addresses one of the most challenging problems in mobile robotics: the so-called "kidnapped robot" problem, in which a robot loses knowledge of its initial pose after being moved, powered off, or displaced.
The study, published in the International Journal of Intelligent Systems , introduces MCL-DLF (Monte Carlo Localization – Deep Local Feature), a coarse-to-fine 3D LiDAR localization framework designed for long-term navigation in large environments. The system has been validated over several months on the UMH Elche campus under varying environmental conditions, including both indoor and outdoor scenarios.
A hierarchical strategy inspired by human orientation
The proposed approach mimics how humans orient themselves in unfamiliar or changing environments. First, the robot performs a coarse localization step, identifying its approximate region based on global structural features extracted from 3D LiDAR point clouds, such as buildings or vegetation.
Once this region is narrowed down, the system performs fine localization, analyzing detailed local features to estimate the robot's exact position and orientation.
"This is similar to how people first recognize a general area and then rely on small distinguishing details to determine their precise location," explains UMH researcher Míriam Máximo, lead author of the study. The work was directed by Mónica Ballesta and David Valiente, also researchers at UMH's Engineering Research Institute of Elche (I3E) . To avoid ambiguity in visually similar environments, the method integrates deep learning techniques that automatically extract discriminative local features from 3D point clouds.
Rather than relying on predefined rules, the robot learns which environmental characteristics are most informative for localization. These learned features are combined with probabilistic Monte Carlo Localization, which maintains multiple pose hypotheses and updates them as new sensor data are received.
Robust to environmental variability
A major challenge in long-term robot navigation is environmental variability. Outdoor spaces change over time due to seasonal shifts, vegetation growth, or lighting differences, which can significantly alter appearance.
The researchers report that MCL-DLF achieves higher position accuracy than conventional approaches while maintaining comparable or superior orientation estimates in certain trajectories. Importantly, the system shows lower variability across time, confirming its robustness to seasonal and structural changes.
Applications in autonomous systems
Reliable localization is fundamental for service robotics, logistics automation, infrastructure inspection, environmental monitoring, and autonomous vehicles. In all these domains, safe operation depends on stable and precise position estimation in real-world, dynamic conditions.
Although fully autonomous navigation remains a central challenge in robotics, this work brings robots closer to operating reliably in large, changing environments without external positioning infrastructure.
Authors and funding
The study was conducted by Míriam Máximo, Antonio Santo, Arturo Gil, Mónica Ballesta, and David Valiente at the Engineering Research Institute of Elche (I3E) , Miguel Hernández University of Elche (UMH).
The research was funded by the Spanish Ministry of Science, Innovation and Universities and the State Research Agency through project PID2023-149575OB-I00, co-funded by the European Regional Development Fund (ERDF), and by the Generalitat Valenciana under the PROMETEO program (CIPROM/2024/8).
