Feature SLAM & Map Management
Feature-based SLAM with online map switching and AdaBoost loop closure for ultra-large dynamic environments.
Problem
Ultra-large warehouses and manufacturing sites exceed the memory and compute budget of resource-constrained edge computers when loading monolithic maps. Dynamic environments further degrade loop closure reliability.
Approach
I led the mapping team toward a production feature-based SLAM product with two key contributions:
Online map switching
- Divides large environments into small map chunks
- Loads chunks on demand based on navigation requirements
- Avoids loading entire facility maps on resource-constrained host computers
Robust loop closure detection
- AdaBoost classifier-based loop closure model robust to environmental changes
- Integrated with range-based and feature-based SLAM implementations
- First major project at Novus Hi-Tech (NHRSL)
Results
- Enabled SLAM-based navigation in ultra-large industrial facilities
- Map switching critical for scalable deployment across multi-site customers
- Loop closure robustness improved under environmental change and dynamism
Technologies
ROS 2 · Feature-based SLAM · Loop closure · Map management · AdaBoost · Edge computing