Core Concepts of SLAM
- Map is needed for localization
- Pose estimation is needed to build a map
- Sensor Noise & drift
- Ambiguous observations
- Dynamic environment
- Real-time constraints
Sensor Fusion
- Lidar (2D, 3D)
- Camera (RGB, Depth) (Mono, Stereo)
- IMU
- Wheel encoder
- Sonar (Ultrasonic)
- Radar
Structure from Motion
- Feature Extraction
- Feature Matching
- Camera Poses
- Triangulation
- Incremental Reconstruction
- Bundle Adjustment
- Theia - algorithms for Structure from Motion (SfM)
- Bundle Adjustment in the Large
- COLMAP
- openMVG
- VisualSFM : A Visual Structure from Motion System
- FLOT: Scene Flow on Point Clouds guided by Optimal Transport
- ByteTrack
Popular SLAM frameworks
- DROID-SLAM
- SLAM Toolbox
- RTAB-Map
- MOLA (Modular Optimization frameworkfor Localization and Mapping)
- LeGO-LOAM
- LIO-SAM
- Fast-LIO2
- OpenVINS
- ORB-SLAM3
AI/ML and SLAM
- Monocular Depth Prediction - MiDaS , Depth Anything
- Learned Place Recognition
- Motion Segmentation via ML
- Learned Feature Extraction & Matching
- Semantic Segmentation & Object Detection
CNN and Visual Models
Classification
Object Detection
Segmentation
Efficientcy Era
Attention and Vision Transformers
Foundation and Generative Vision Models
Datasets
References
- Vladimir Georgiev - AI за компютърно зрение и роботика
- https://github.com/VladiGGeorgiev
- Can Modern C++ SPEED UP Your Bundle Adjustment Pipeline? - Vishnu Sudheer Menon
- SLAM for Mobile Robotics: Localization and Mapping in the Real World
- https://github.com/ali-pahlevani
- Large Scale Interactive Motion Forecasting for Autonomous Driving : The WAYMO OPEN MOTION DATASET
- Multi-Agent Trajectory Prediction with Scalable Diffusion Transformer
- Vincent Sitzmann - Machine Learning for Inverse Graphics
- Rafael C. Gonzalez, Richard E. Woods - Digital Image Processing
- Richard Szeliski - Computer Vision: Algorithms and Applications