A human-following robot for indoor navigation using deep learning and LiDAR-based path planning

dc.contributor.authorBremer, Luis Bernardo
dc.contributor.authorZuñiga , Hiram
dc.contributor.authorOrozco Rosas, Ulises
dc.contributor.authorPicos, Kenia
dc.date.accessioned2026-11-13T19:35:47Z
dc.date.created2026
dc.date.issued2026
dc.description.abstractHuman-following mobile robots can support indoor security and defense scenarios in which autonomous platforms must accompany designated operators through cluttered spaces. In Global Navigation Satellite System (GNSS)-denied environments, satellite positioning is unavailable, and vision-only following robots may fail due to occlusion, crowding, or target switching. This paper presents a human-following robot built on a computationally lightweight architecture that integrates deep-learning perception with a LiDAR-based reactive obstacle-avoidance layer. Built on a TurtleBot2 platform with a Kobuki base, the system combines YOLOv3 for person detection, Channel and Spatial Reliability Tracking (CSRT) for target continuity, and Hue, Saturation, and Value (HSV) color histograms for identity locking to ensure that the correct target is tracked across multi-person scenes and partial occlusions. The reactive layer partitions a 90° frontal LiDAR arc into twelve angular sectors, enabling the robot to steer around obstacles rather than merely halt. During experiments, computation is distributed between the onboard NVIDIA Jetson TX2, which handles sensor publishing and actuation, and an external GPU machine that performs the deep-learning inference, the reactive obstacle-avoidance layer, and control; this split serves as a development proxy for a single, more capable onboard compute module onto which the same pipeline collapses in deployment, eliminating any external processor or wireless link. The system was evaluated over twenty logged obstacle course runs, completing the course each time, reproducing one avoidance maneuver per obstacle, and steering away from the obstructed side in 91% of turns commanded while an obstacle was present. The system supports follow-me logistics, patrol support, escort assistance, and facility inspection in critical infrastructure, command centers, urban interiors, and subterranean defense scenarios.
dc.description.sponsorshipSPIE.Digital Library
dc.description.urlhttps://www.spiedigitallibrary.org/conference-proceedings-of-spie/14291/3116973/A-human-following-robot-for-indoor-navigation-using-deep-learning/10.1117/12.3116973.full
dc.format.page142910P
dc.identifier.doihttps://doi.org/10.1117/12.3116973
dc.identifier.indexacionScopus
dc.identifier.urihttps://repositorio.cetys.mx/handle/60000/2110
dc.language.isoen_US
dc.relation.ispartofseries14291
dc.subjectHuman-Robot Interaction, Computer Vision, Occlusion, Deep Neural Networks, Convolutional Neural Networks, Socially Assistive Robots
dc.subject.sedeCampus Tijuana
dc.titleA human-following robot for indoor navigation using deep learning and LiDAR-based path planning
dc.title.alternativeAutonomous Systems for Security and Defence III
dc.typeArticle

Files

License bundle

Now showing 1 - 1 of 1
Loading...
Thumbnail Image
Name:
license.txt
Size:
1.39 KB
Format:
Item-specific license agreed upon to submission
Description: