Session 1B – Localization and Mapping
Date: Wednesday, 18th November 2026
Time: 11:00 – 12:15
Room: Aula Capella
Session Chair: TBA
Presentation format: Each paper is allocated 15 minutes: approximately 10–12 minutes for the presentation and 3–5 minutes for questions.
11:00 – Approximate reconstruction of 3D objects from 3D SLAM using frameworks integrating optics, geometry, and semantics
Authors: Kavy Nihar Mittal¹,² and Alberto Sanfeliu¹,²
Affiliations: ¹ Institut de Robòtica i Informàtica Industrial (IRI, CSIC-UPC), Spain; ² Universitat Politècnica de Catalunya – BarcelonaTech (UPC), Barcelona, Spain.
Abstract
Precise 3D object reconstruction has been achieved through exact sensor localization; however, accurately reconstructing 3D objects using the robot’s sensor pose estimation (via 3D SLAM) remains a challenge. This paper presents a workflow for object-level 3D reconstruction using RGB-D data captured by a ZED 2i camera, the pose of which is determined by RTAB-Map using both the RGB-D data and the camera’s IMU. This information, along with the RGB-D frames is passed through a segmentation pre-processing and is subsequently fed into the OpenGS-Fusion workflow to generate a semantically enriched, voxel-based Gaussian reconstruction. This reconstruction is refined through semantic label fusion and queried to obtain the object’s point cloud; this cloud is extracted and can be placed in any other reconstructed environment at a user-defined location. This approach highlights the potential of semantic Gaussian representations for object manipulation. The results of 3D object reconstruction in real-world environments demonstrate that the methods employed are consistent with the pose accuracy achieved with 3D SLAM.
11:15 – Unsigned Distance Maps on 2D Point Cloud Registration
Authors: Ricardo B. Sousa¹,², Giorgio Grisetti³, Héber Miguel Sobreira², Carlos André Silva⁴ and António Paulo Moreira¹,²
Affiliations: ¹ Faculty of Engineering, University of Porto (FEUP), Porto, Portugal; ² INESC TEC – Institute for Systems and Computer Engineering, Technology and Science, Porto, Portugal; ³ Sapienza University of Rome, Rome, Italy; ⁴ Flowbotic Mobile Systems, S.A., Viseu, Portugal.
Abstract
2D point cloud registration is the front-end of Simultaneous Localization and Mapping (SLAM) for mobile robots with 2D lasers, where the back-end consumes both the pose estimate and its uncertainty. Iterative Closest Point (ICP), one of the most used approaches, recomputes correspondences via nearest-neighbor search at every iteration, whereas correspondence-free alternatives focus on scan-to-map alignment. This paper proposes 2D point cloud registration based on unsigned distance maps, precomputing the distance to the nearest reference point over a grid, along with its derivatives, replacing the per-iteration search with O(1) lookups. Point-to-point and point-to-plane formulations are derived on SE(2), solved via Gauss-Newton, and evaluated for accuracy and uncertainty consistency against ICP baselines. On a synthetic benchmark and IILABS 3D dataset, the precomputed point-to-point variant outperforms its analytical counterparts, achieving competitive odometry drift compared to point-to-plane formulations, as the precomputed gradient regularizes correspondences in the presence of noise.
11:30 – Autonomous Dungeon Navigation and Mapping with Frontier-Based Exploration and Incremental Path Planning
Authors: João Rocha¹,², Paulo Pinto¹,², Rui Alvarilhão¹,², Gonçalo Leão²,³, Armando Sousa²,³ and Luís Paulo Reis²,⁴
Affiliations: ¹ Faculty of Sciences, University of Porto (FCUP), Portugal; ² Faculty of Engineering, University of Porto (FEUP), Portugal; ³ INESC TEC – INESC Technology and Science, Portugal; ⁴ LIACC – Artificial Intelligence and Computer Science Laboratory, Portugal.
Abstract
Autonomous exploration in unknown environments requires the simultaneous integration of mapping, planning, and mission-level reasoning. This work presents a complete pipeline for a mobile robot tasked with exploring a procedurally-generated dungeon, collecting keys to unlock doors, retrieving a final goal item, and returning to its starting position. The pipeline combines a probabilistic occupancy grid based on the inverse sensor model, frontier-based exploration with Configuration Space inflation, D* Lite incremental path planning, and a mission-level Finite State Machine that arbitrates between exploration and goal-directed navigation. Path execution is performed by a Pure Pursuit controller augmented with proximity-based velocity damping, frontal symmetry breaking, and an emergency reverse mechanism. The system is evaluated in the Webots simulator using a parametric procedural generator that produces six dungeons grouped into three difficulty levels, each tested under twelve combinations of LiDAR range, resolution, and noise, for a total of 360 runs. The results validate the chosen integration of classical algorithms as a viable architecture for mission-driven autonomous exploration in environments that are non-intuitive to humans.
11:45 – Coupled consensus for distributed perception
Authors: Sergio Pardina, Rosario Aragüés and Gonzalo López-Nicolás
Affiliation: Instituto de Investigación en Ingeniería de Aragón (I3A), Universidad de Zaragoza, Spain.
Abstract
This paper presents a multiagent framework where agents track a set of beacons that form, in aggregate, a dynamic deformable body whose shape changes over time. The agents aim to reach dynamic consensus on the average state of these beacons, which represents the centroid of the deformable body. Each agent computes and leverages a reference signal from a linear combination of the subset of beacons it locally perceives. To prevent over-representation of the beacons perceived by multiple agents, a secondary perception consensus process is introduced to regulate the construction of reference signals. The system is modeled as a coupled consensus process, where the primary state consensus directly depends on the secondary perception consensus. We demonstrate that the proposed scheme converges within a finite neighborhood and illustrate its performance in simulation.
12:00 – An Empirical Benchmark of State Estimation, Localization, and Mapping for Scaled Autonomous Racing
Authors: Fernando Villa-Castell, Hugo A. Fitor-Ricote, Santiago Montiel-Marín, Franck Fierro, Fabio Sánchez-García, Manuel Ocaña and Luis M. Bergasa
Affiliation: Department of Electronics, University of Alcalá, Madrid, Spain.
Abstract
Reliable odometry provides an accurate estimate of the vehicle’s state, which is critical for safe autonomous navigation. In high-speed racing scenarios, confidently assessing the surroundings and precisely localizing the vehicle enables obstacle avoidance and optimal racing lines. In this paper, we benchmark wheel and LiDAR odometry, localization, and mapping algorithms to establish the foundations of a robotic platform for scaled autonomous racing. We calibrate the platform and fuse wheel and LiDAR odometry through an Extended Kalman Filter to obtain a reliable state estimate. We then evaluate three map-based localization algorithms (AMCL, SLAM Toolbox and Cartographer) against a mapless RF2O baseline, and two mapping algorithms (SLAM Toolbox and Cartographer) across simulated F1-inspired and synthetic circuits, prior to deploying the selected stack on the real vehicle. Results show that sensor fusion reduces full-circuit odometry error by over 90% with respect to wheel encoders alone. We further validate the mapping pipeline through a qualitative on-car experiment, mapping a real-world circuit onboard with both SLAM Toolbox and Cartographer during a teleoperated lap.