Session 7A – Robot Navigation 2
Date: Friday, 20th November 2026
Time: 09:00 – 10:30
Room: Sala d’Actes
Session Chair: TBA
Presentation format: Each paper is allocated 15 minutes: approximately 10–12 minutes for the presentation and 3–5 minutes for questions.
09:00 – Simulation-Based Fault Injection for Performance Evaluation of a Quadrotor Controller
Authors: Mustafa Saraoglu, Janneck Hamann, Frederik Luca Zimmermann and Klaus Janschek
Affiliations: Institute of Automation, Technische Universität Dresden, Dresden, Germany.
Abstract
This work presents a modular fault-injection framework for simulating unmodeled real-world effects and evaluating the performance of a quadrotor control system. The study is based on the cascaded control architecture of a Crazyflie quadrotor modeled and implemented in Webots. Crazyflie flight data obtained from real-world experiments are used as a reference for investigating the unmodeled effects to be implemented in the Webots simulation. To this end, the simulation model is extended with configurable fault-injection functions for aerodynamic drag, ground effect, and position sensor noise. The gap between the simulation results and the real-flight data is analyzed through a series of experiments involving altitude step response and trajectory tracking, and the mismatch is used to manually adjust the parameters of the fault-injection functions. The proposed framework is then used to evaluate the performance and quantify the remaining model mismatch of the cascaded control architecture implemented in Webots using post-simulation metrics such as tracking error, control effort, mission success, and successful mission duration.
09:15 – Acoustic Drone Localization Using a Stereo System Based on Deep Neural Networks
Authors: Guilherme Silva, Sedat Dogru and Lino Marques
Affiliations: Institute of Systems and Robotics, Department of Electrical and Computer Engineering, University of Coimbra, Coimbra, Portugal.
Abstract
Malevolent use of drones has been around for some time, requiring implementation of both custom hardware and specific techniques for detection and localization. To address this problem, in this work we propose a new system consisting of a stereo microphone pair mounted on a pan tilt unit to continuously listen to the environment, and with the aid of a multi-head Deep Neural Network (DNN) detects and continuously localizes the invading drone. The DNN based approach relies both on Mel features and inter-aural level and time difference measurements. We validate the proposed approach using field data, showing very high detection performance, and acceptable localization error.
09:30 – Guidance and Control of a Wing-in-Ground-Effect Vehicle for Low-Altitude Waypoint Tracking
Authors: Sarthak Mishra, Ramon A. Suarez Fernandez, Zorana Milošević, Claudio Rossi and Sergio Domínguez
Affiliations: Centre for Automation and Robotics (CAR), UPM-CSIC, Universidad Politécnica de Madrid, Madrid, Spain.
Abstract
Wing-In-Ground (WIG) effect vehicles face tightly coupled lateral–longitudinal dynamics and safety-critical constraints when operating at near-surface altitudes, making autonomous control particularly challenging. This paper presents a system of cascaded PID controllers, together with a classical guidance algorithm based on great-circle bearing calculations to achieve waypoint tracking tasks. The architecture was validated using X-Plane 12 with a custom WIG vehicle model and a ROS 2 communication underlay, imposing strict roll limits of ±5° for near-surface safety. The approach used in this work combines a PI-PID loop for pitch control, a PI throttle loop for energy management, a PID loop combined with lateral guidance and a cascaded PD-PI loop with a cross-track-to-roll mapping bounded for low-altitude operation. Testing across square, figure-eight, and 3D low-altitude trajectories at 50 Hz, the altitude tracking achieves RMSE of 0.072 −0.321 m, the roll obtains an RMSE of 0.170 −0.271°, and stable low-altitude flight, while heading transitions induce larger variations that motivate circular-error handling. The results validate this modular control architecture and establish a baseline for future comparison with advanced non-linear controllers.
09:45 – RL-Based Charging Control of a Shuttle Robot for Always-On Multi-Service-Robot Operation
Authors: Mojgan Ghanbari¹, Shahed Rasekh², Amaro Fernandes de Sousa¹ and Pedro Fonseca¹
Affiliations: ¹ Instituto de Telecomunicações, Universidade de Aveiro, Aveiro, Portugal; ² INMA (CSIC-Universidad de Zaragoza), Zaragoza, Spain.
Abstract
Mobile service robots (SRs) are constrained by limited battery capacity and require periodic recharging. We consider three SRs executing predefined cyclic missions and one mobile charging shuttle robot (ShR) that transfers energy in situ while the SRs remain in motion and periodically returns to a base station (BS). Target selection is addressed with an event-driven linear state-action-reward–state-action (SARSA) controller. Dynamic replanning permits early BS departure and mid-charge switching, while reach-and-return energy screening and a slot-level return guard protect the ShR energy reserve. To separate learning from replanning, an additional comparison baseline (H1) retains the lowest-energy-first target rule and energy-safety logic of the original heuristic (H0), but is given the same replanning opportunities as RL. Ten independently trained policies are evaluated on the same 100 randomized operating cases. H1 improves mean Network Yield over H0, while both deterministic heuristics achieve only a 1% always-on success rate in these randomized cases. The RL policies achieve an across-seed mean Network Yield of about 98.5% and an always-on success rate of about 38%, where success requires zero stopped-robot slots over the complete test episode. Nine of ten RL training seeds provide a further Yield gain over H1. The results support an additional contribution from learned target selection while revealing residual training-seed sensitivity.
10:00 – Comparative Evaluation of Autonomous Racing Controllers in a RoboRacer Simulation Environment
Authors: Rubén López-Ruiz, Rodrigo Gutiérrez-Moreno, Miguel Antunes-García, Santiago Montiel-Marín, Pedro Revenga and Luis M. Bergasa
Affiliations: Electronics Department, University of Alcalá (UAH), Spain.
Abstract
Reliable motion control at the limit of traction is a foundational challenge in autonomous racing. Although numerous path-tracking algorithms have been proposed, comparing them fairly remains difficult due to heterogeneous environments, incompatible planning assumptions, and manual tuning bias. This paper investigates how four representative control paradigms—Proportional-Integral-Derivative (PID), Pure Pursuit, Model- and Acceleration-based Pursuit (MAP), and Model Predictive Control (MPC)—compare in lap time, tracking accuracy, and reliability when co-optimized with their reference trajectory under an identical simulation environment and tuning budget. Using ROS 2 and MVSIM on a 1:10 scale RoboRacer vehicle, we pair an offline trajectory optimizer (generating minimum-curvature racing lines and kinematically feasible speed profiles from track centerlines) with automated hyperparameter calibration via Gaussian Process-based Bayesian Optimization. By jointly tuning trajectory and controller parameters over a fixed 100-iteration budget, we evaluate baseline and optimized planning-and-control pipelines without human bias. Our results quantify the trade-offs between speed, tracking precision, and reliability, offering actionable guidelines for controller selection.
10:15 – TRex: A Compact Telerobotic System for Archaeological Underground Cistern Exploration
Authors: Michael A. Saliba¹, Natasha Micallef¹ and John C. Betts²
Affiliations: ¹ Department of Mechanical Engineering, University of Malta, Msida, Malta; ² Department of Classics and Archaeology, University of Malta, Msida, Malta.
Abstract
Archaeologically significant underground spaces such as cisterns, wells, shafts, and subsurface cavities frequently remain inaccessible due to restricted apertures, structural instability, hazardous conditions, and preservation-sensitive environments. This paper presents the design, development, and experimental validation of TRex (Telescopic Reach explorer), a compact telerobotic system specifically developed for archaeological underground cistern exploration and surveying. The robot employs a lightweight telescopic architecture capable of deployment through narrow openings while maintaining sufficient operational reach for underground visual inspection tasks. A hybrid actuation strategy combining manual and electrical actuation was adopted to minimise power consumption, structural mass, and overall system complexity while preserving operational controllability and portability. The design methodology incorporated customer-driven requirements analysis, concept generation, embodiment design, structural analysis, and prototype testing. Analytical calculations and finite element analysis validated the structural feasibility of the proposed robotic architecture. A functional prototype was manufactured using low-cost fabrication techniques and locally available components. Experimental validation demonstrated successful deployment and image acquisition within representative underground environments and during field trials conducted at an archaeological excavation site in Marsaxlokk, Malta. Although specifically developed for archaeological underground surveying, the resulting robotic platform additionally demonstrates potential applicability to broader confined-space inspection applications.