Autonomous Indoor Inspection for Industry 5.0: Bridging the Reality Gap with V-SLAM and Behavior Tree Control
Faculty Mentor
Dr. Valentin Soloiu
Location
Russell Union Ballroom
Type of Research
On-going
Session Format
Oral Presentation
College
Allen E. Paulson College of Engineering & Computing
Department
Mechanical Engineering
Abstract
Unplanned downtime costs the industrial manufacturing sector an estimated $50 billion annually. This economic burden is compounded by the reliance on manual inspection for hazardous tasks, evidenced by a 41% rise in fatal injuries within confined spaces in 2022, with the warehousing sector reporting injury rates more than double the private industry average. While Industry 5.0 advocates for automation to mitigate these risks, deploying Unmanned Aerial Vehicles (UAVs) indoors is restricted by the failure of GPS signals in enclosed structures. This research tests the hypothesis that a Visual Simultaneous Localization and Mapping (V-SLAM) system, governed by Behavior Tree logic, can achieve robust, infrastructure-independent navigation where traditional automation fails. To validate this, an NVIDIA Jetson Orin Nano and Intel RealSense depth camera are integrated onto a modified Parallax ELEV-8 airframe. A key safety feature involves pre-capturing a high-fidelity map to generate a 2D occupancy grid; this acts as a static geofence to constrain the UAV while V-SLAM handles real-time positioning. Experimental validation focuses on executing autonomous inspection primitives to measure computational efficiency and responsiveness. While the Behavior Tree implementation demonstrates a threefold improvement in decision-making speed compared to legacy Finite State Machines. By bridging the gap between simulated control logic and physical deployment, this project contributes a scalable, safety-critical architecture for industrial monitoring, effectively removing human workers from the hazard loop. Future work will validate the performance for key metrics such as localization accuracy and operational robustness under varying speed and ambient lighting conditions.
Program Description
.
Start Date
4-23-2026 2:15 PM
End Date
4-23-2026 2:30 PM
Recommended Citation
Lange, Robin, "Autonomous Indoor Inspection for Industry 5.0: Bridging the Reality Gap with V-SLAM and Behavior Tree Control" (2026). GS4 Student Scholars Symposium. 231.
https://digitalcommons.georgiasouthern.edu/research_symposium/2026/2026/231
Autonomous Indoor Inspection for Industry 5.0: Bridging the Reality Gap with V-SLAM and Behavior Tree Control
Russell Union Ballroom
Unplanned downtime costs the industrial manufacturing sector an estimated $50 billion annually. This economic burden is compounded by the reliance on manual inspection for hazardous tasks, evidenced by a 41% rise in fatal injuries within confined spaces in 2022, with the warehousing sector reporting injury rates more than double the private industry average. While Industry 5.0 advocates for automation to mitigate these risks, deploying Unmanned Aerial Vehicles (UAVs) indoors is restricted by the failure of GPS signals in enclosed structures. This research tests the hypothesis that a Visual Simultaneous Localization and Mapping (V-SLAM) system, governed by Behavior Tree logic, can achieve robust, infrastructure-independent navigation where traditional automation fails. To validate this, an NVIDIA Jetson Orin Nano and Intel RealSense depth camera are integrated onto a modified Parallax ELEV-8 airframe. A key safety feature involves pre-capturing a high-fidelity map to generate a 2D occupancy grid; this acts as a static geofence to constrain the UAV while V-SLAM handles real-time positioning. Experimental validation focuses on executing autonomous inspection primitives to measure computational efficiency and responsiveness. While the Behavior Tree implementation demonstrates a threefold improvement in decision-making speed compared to legacy Finite State Machines. By bridging the gap between simulated control logic and physical deployment, this project contributes a scalable, safety-critical architecture for industrial monitoring, effectively removing human workers from the hazard loop. Future work will validate the performance for key metrics such as localization accuracy and operational robustness under varying speed and ambient lighting conditions.
