IMPORTANT: the complete and detailed report is available inside the folder
/report
This project implements a planar monocular Simultaneous Localization and Mapping (SLAM) system for a differential drive robot equipped with a single camera. The system integrates:
- 📍 Wheeled odometry
- 🎯 Point projections for trajectory estimation
- 🗺️ Landmark mapping
- 🔧 Robust Bundle Adjustment using Huber, Cauchy, and Tukey M-estimators to handle outliers effectively.
- Monocular SLAM using a single camera
- Odometry integration for improved accuracy
- Triangulation for 3D point estimation
- Bundle Adjustment (BA) for refining trajectory and landmark positions
- Robust Optimization using M-estimators:
- Huber (smooth transition between quadratic and linear loss)
- Cauchy (fast suppression of large errors)
- Tukey (aggressive outlier down-weighting)
conda create --name myenv --file packageslist.txtRun the main script with the appropriate arguments:
python main.py --kind BA --iterations 20 --damping 1.0 --threshold 1e3 --optimize False --method HUBER --param 1.0--kind: Type of run (BA,RBA,COMPARISON)--iterations: Number of iterations for BA/RBA--damping: Damping factor for BA/RBA--threshold: Inlier kernel threshold for BA/RBA--optimize: Boolean flag for performing an "only-landmark" pre-optimization--method: Robustifier method (CAUCHY,HUBER,TUKEY,NONE)--param: Specific value for the chosen robustifier
Modify parameters inside the code or via bash for tuning the system behavior.
The system has been tested under two configurations:
- Without pre-optimization: Directly applying Bundle Adjustment (BA/RBA) on the full system.
- With pre-optimization: Refining landmark estimates first, then running BA/RBA.
- Robust BA improves accuracy and resilience to noisy measurements.
- Pre-optimization further enhances mapping precision.
- Tuning M-estimator parameters significantly impacts system performance.
This project is licensed under Creative Commons Attribution 4.0 International. See LICENSE for details.
🤖 Happy Mapping! 🎉