UGV · LiDAR · SLAM
Autonomous Infrastructure Inspection
Graduate research building an autonomous ground-vehicle platform that fuses LiDAR, RGB-D cameras and computer vision to map and inspect civil infrastructure. The system explores how visual-LiDAR SLAM, sensor fusion and offline trajectory refinement can produce reliable maps for structural inspection and digital-twin generation.
Status
Active Research
Timeline
2026
Role
MESc Researcher — Lead
Category
Robotics
Technology Stack
Problem
Manual infrastructure inspection is slow, subjective and often unsafe. Reliable autonomous mapping requires robust perception across varied, GPS-denied environments.
Constraints
- GPS-denied and cluttered indoor/outdoor environments
- Real-time compute budget on mobile hardware
- Sensor calibration and time synchronization
Solution
A modular ROS-based perception stack combining LiDAR odometry, visual features and sensor fusion, with an offline refinement stage that improves trajectory and map quality after data capture.
Key Decisions
- Modular ROS graph so sensors and algorithms can be swapped independently
- Offline refinement to decouple capture from heavy optimization
- Containerized environment for reproducible experiments
System Architecture
Diagram placeholder — replace with a detailed architecture diagram when available.
Hardware
- Unmanned ground vehicle (UGV)
- 3D LiDAR
- RGB-D camera
- IMU
- Onboard compute
Software
- ROS / ROS 2
- Point Cloud Library
- OpenCV
- C++ & Python nodes
- Docker
- RViz
Challenges
- Cross-sensor calibration and synchronization
- Drift management in feature-poor corridors
- Balancing map density against processing cost
Results
- Research in progress — results, datasets and evaluation will be published here.
Lessons Learned
- Reproducible tooling and disciplined data recording matter as much as the algorithms.