Thematic Section: Vision-based Navigation in Mobile Robotics

Abstract

In this tehmatic section of Computación y Sistemas CyS we present a selection of 10 research papers showcasing recent advances in intelligent mobile robotics through innovations in autonomous navigation visual perception motion planning and AIdriven decisionmaking A prominent theme is the growing reliance on vision as a primary sensing modality particularly for UAVs operating in complex and dynamic environments The issue highlights both learningbased and classical methodologies demonstrating how deep learning generative AI geometric reasoning and traditional planning techniques can complement one another to achieve robust autonomy Contributions span emergencyresponse UAVs autonomous drone racing humanoid and biped robot navigation intelligent transportation systems and machinelearningenabled perception reflecting the multidisciplinary nature of modern mobile robotics Together these works emphasize practical deployment realtime performance and robust operation in realworld scenarios which are key characteristics of nextgeneration intelligent mobile robotic systems The review process for this Special Issue was conducted by the Guest Editors All submissions were subject to a rigorous doubleblind peerreview process involving at least three independent reviewers Manuscripts were evaluated on the basis of their technical merit novelty methodological rigor quality of the stateoftheart analysis and overall contribution to the field of intelligent mobile robotics Special consideration was given to studies presenting validated robotic systems and demonstrating their effectiveness in realworld deployment scenarios Next we provide a summary of the contributions included in this thematic section Osorio and MartinezCarranza present a visionlanguage navigation framework for Unmanned Aerial Vehicle UAV deployment in an environment that is GPSdenied searchandrescue This proposed approach enables operators to issue freeform natural language instructions which are translated into imagegrounded shorthorizon waypoints instead of directly generated text actions that are combined with distance estimations to produce threedimensional displacement commands while adaptive steplength control and continuous replanning improve navigation under changing visibility moving targets and emerging obstacles This proposal experiments in highfidelity simulated rescue scenarios achieving an online XY RMSE of 0098 meters and demonstrating shorter paths reduced completion times and reliable success rates compared with baseline methods These results support the potential of visionlanguage models for obstacleaware UAV navigation nevertheless realworld flight validations are necessary