Loading
My graduate research explores how unmanned ground vehicles can use LiDAR, RGB-D cameras, computer vision and SLAM to map and inspect infrastructure — from raw sensor data to inspection-ready digital twins.
An autonomous ground vehicle carries the sensing payload into infrastructure environments where inspection is slow, subjective or unsafe for people.
A rotating LiDAR sweeps the environment, measuring precise distances to surfaces around the vehicle as the foundation for 3D perception.
Each laser return becomes a measured point. Geometric features are extracted from the sweep to anchor motion estimation.
Thousands of returns accumulate into a dense point cloud — a raw 3D reconstruction of the surrounding structure.
By tracking how features move between scans, the system estimates the vehicle's trajectory through the environment.
RGB-D cameras add color and dense depth. Fusing LiDAR, vision and inertial data yields perception more robust than any single sensor.
SLAM jointly refines where the robot is and what the world looks like — localizing the platform while building a consistent map.
Offline refinement corrects drift and consolidates the map into a structured, inspection-ready digital representation.
The finished map feeds infrastructure inspection, structural health monitoring and digital twins — connecting the physical world to digital analysis.
A structured overview of the research. Sections marked as in progress will be expanded with methodology, datasets, experiments and results over time.
Manual infrastructure inspection is slow, subjective and often unsafe. This research investigates how autonomous ground vehicles can reliably map and inspect infrastructure in GPS-denied environments using multi-sensor perception.
A modular perception pipeline combining LiDAR odometry, visual features and inertial data, with an offline refinement stage that improves trajectory and map quality after capture. Details will expand as the work develops.
Unmanned ground vehicle equipped with 3D LiDAR, an RGB-D camera, an IMU and onboard compute. Exact configuration is being finalized.
Built on ROS / ROS 2 with the Point Cloud Library, OpenCV and containerized environments for reproducible experiments. Nodes are decoupled so sensors and algorithms can be swapped independently.
Recorded sensor sequences captured across representative infrastructure environments. Dataset details and access will be documented here.
Planned experiments compare mapping quality and localization accuracy across sensor configurations and refinement strategies. Protocols will be published as they are finalized.
Results are in progress. Quantitative evaluation, maps and analysis will be added here — no results are claimed prematurely.
Publications will be listed here as they become available.
Interested in collaborating on this research? Get in touch.