# SLAM Phylogenetic Taxonomy v0.1 Source: SLAM Handbook public release, compiled April 21, 2026. Unit: `Phylum > Class > Order > Genus` Rule: each paper should receive one primary path. Cross-cutting traits such as "deep", "real-time", or "multi-modal" are only primary when they are the central contribution. ## 1. Problem & System Context - **SLAM Problem Definition** - Localization vs Mapping vs SLAM - Known-map Localization - Mapping with Known Poses - Joint Localization and Mapping - Autonomy Role - Navigation Support - Manipulation Support - Exploration Support - Long-term Spatial Memory - **System Architecture** - Modern SLAM Pipeline - Front-end / Back-end Split - Odometry Stream - Loop-Closure Stream - Map Update Loop - Autonomy Integration - Control Loop Interface - Planning Loop Interface - Latency and Rate Separation - Online Operation - **Historical and Scope Framing** - SLAM Eras - Probabilistic Robotics Era - Graph Optimization Era - Robust Perception Era - Spatial AI Era ## 2. State, Geometry & Probabilistic Modeling - **Factor Graph Modeling** - Graphical Model Formulation - Variables and Factors - Landmark-based Models - Pose-Graph Models - Hybrid Factor Graphs - MAP Inference - Negative Log-Likelihood - Least-Squares Formulation - Noise Models - **State Representations** - Lie Groups and Manifolds - SO(3) Rotation States - SE(3) Pose States - Retractions and Local Coordinates - Continuous-Time Trajectories - Gaussian Process Trajectories - Spline Trajectories - Asynchronous Sensor Fusion - **Observability and Uncertainty** - Estimation Consistency - Gauge Freedom - Degeneracy - Covariance and Marginals - Inertial Observability - Bias Observability - Gravity Alignment - Scale Observability ## 3. Measurement Front-End - **Raw-to-Pseudo Measurement Processing** - Feature Extraction - Keypoints and Descriptors - Lines and Planes - Learned Features - Direct Measurement Formation - Photometric Residuals - Range Residuals - Event Residuals - Registration - ICP - Scan-to-Map Matching - Point-to-Plane Matching - **Odometry Front-End** - Frame-to-Frame Motion Estimation - Visual Odometry - LiDAR Odometry - Radar Odometry - Event Odometry - Preintegration and Propagation - IMU Preintegration - Leg Odometry Propagation - Wheel / Kinematic Propagation - **Place Recognition and Loop Closure** - Place Recognition - Visual Place Recognition - LiDAR Place Recognition - Radar Place Recognition - Cross-Modal Place Recognition - Loop Closure Validation - Geometric Verification - Temporal Consistency - Outlier-aware Loop Closure - **Data Association** - Correspondence Estimation - Landmark Association - Scan Correspondence - Object Association - Ambiguity Handling - Perceptual Aliasing - Incorrect Association Detection - Multi-Hypothesis Association ## 4. Back-End Optimization & Inference - **Least-Squares Solvers** - Linear Least Squares - Normal Equations - QR / Cholesky - Sparse Linear Algebra - Nonlinear Optimization - Gauss-Newton - Levenberg-Marquardt - Trust Region Methods - **Sparsity and Incrementality** - Variable Elimination - Bayes Trees - Fill-in Reduction - Schur Complement - Incremental SLAM - iSAM-style Updates - Fixed-Lag Smoothing - Marginalization - **Robust Back-End** - Robust Cost Functions - M-estimators - Switchable Constraints - Graduated Non-Convexity - Outlier-Robust Optimization - Max-Mixture Models - Pairwise Consistency - Robust Pose Graph Optimization - **Certifiable and Differentiable Solvers** - Certifiably Optimal SLAM - Semidefinite Relaxation - Tightness Certification - SE-Sync-style Solvers - Differentiable Optimization - Differentiating Through Least Squares - Implicit Differentiation - Differentiation on Manifolds ## 5. Map Representations - **Sparse Geometric Maps** - Landmark Maps - Point Landmarks - Line Landmarks - Plane Landmarks - Pose Graph Maps - Topological Graphs - Submap Graphs - Multi-Session Graphs - **Dense Geometric Maps** - Range-Derived Maps - Occupancy Grids - TSDF / ESDF - Point Cloud Maps - Mesh Maps - Dense Reconstruction - Real-Time Dense Reconstruction - RGB-D Reconstruction - Large-Scale Dense Mapping - **Neural and Differentiable Maps** - Neural Implicit Maps - NeRF Maps - Neural SDF Maps - Learned Occupancy Fields - Gaussian Maps - 3D Gaussian Splatting - Differentiable Rendering Maps - Hybrid Explicit-Implicit Maps - **Semantic and Structured Maps** - Metric-Semantic Maps - Sparse Semantic Landmarks - Dense Semantic Maps - Object-Level Maps - Hierarchical Spatial Maps - 3D Scene Graphs - Room / Place Graphs - Open-Vocabulary Maps - **Dynamic and Deformable Maps** - Dynamic SLAM Maps - Short-Term Dynamic Objects - Long-Term Lifelong Changes - Multi-Body Scene Models - Deformable SLAM Maps - Non-Rigid Surfaces - Deformable Object Maps - Elastic Scene Models ## 6. Sensor & Odometry Modalities - **Visual SLAM** - Camera Models and Pipelines - Monocular SLAM - Stereo SLAM - RGB-D SLAM - Visual System Families - Feature-Based Visual SLAM - Direct Visual SLAM - Hybrid Visual SLAM - **LiDAR SLAM** - LiDAR Sensing - 2D LiDAR - 3D Spinning LiDAR - Solid-State LiDAR - LiDAR System Families - LOAM-style Odometry - Scan-Matching SLAM - LiDAR-Inertial SLAM - **Radar SLAM** - Radar Sensing - FMCW Radar - Imaging Radar - 4D Radar - Radar System Families - Radar Odometry - Radar Place Recognition - Radar Mapping - **Event-Based SLAM** - Event Sensor Processing - Event Representations - Event Feature Tracking - Event Time Surfaces - Event System Families - Event Odometry - Event-Inertial Odometry - Event-Based Mapping - **Proprioceptive and Aided Odometry** - Inertial Odometry - Strapdown Integration - IMU Preintegration - Visual-Inertial Odometry - Leg Odometry - Contact Estimation - Kinematic Factors - Legged State Estimation - Other Aiding Signals - Wheel Odometry - GNSS Factors - UWB / Radio Factors - **Multi-Modal Fusion** - Sensor Fusion Architectures - Tightly Coupled Fusion - Loosely Coupled Fusion - Cross-Modal Factors - Calibration and Synchronization - Extrinsic Calibration - Temporal Calibration - Online Calibration ## 7. Robustness, Evaluation & Operations - **Outlier and Failure Robustness** - Front-End Robustness - RANSAC-style Rejection - Descriptor Ambiguity Handling - Dynamic Object Rejection - Back-End Robustness - Robust Kernels - Certifiable Outlier Rejection - Failure Recovery - **Evaluation** - Metrics - Absolute Trajectory Error - Relative Pose Error - Map Quality Metrics - Place Recognition Metrics - Datasets and Benchmarks - Visual SLAM Datasets - LiDAR SLAM Datasets - Radar SLAM Datasets - Event SLAM Datasets - Legged SLAM Datasets - **Operational Constraints** - Real-Time Performance - Embedded Runtime - Latency Budgets - Memory Growth Control - Deployment Conditions - GPS-Denied Operation - Large-Scale Operation - Long-Term Operation - Resource-Constrained Operation ## 8. Learning, Semantics & Spatial AI - **Learning for Classical SLAM Components** - Learned Geometry - Depth Prediction - Camera Pose Prediction - Optical Flow - Learned Correspondence - Feature Matching - Dense Matching - Learned Place Recognition - Learned Optimization - Differentiable Bundle Adjustment - DROID-SLAM-style Recurrent Updates - Learned Priors - **Foundation and Open-World Spatial AI** - Foundation Models for Spatial AI - Vision-Language Grounding - Open-Vocabulary Recognition - Large-Scale Semantic Priors - Open-World Mapping - Open-Set Object Maps - Language-Grounded Maps - Continual Map Expansion - **Computational Structure** - Spatial AI Graphs - Factor-Graph Computation - Scene-Graph Computation - Hybrid Neural-Symbolic Graphs - Distributed and Hardware-Aware Computation - Processor / Sensor Co-Design - Gaussian Belief Propagation - Continual Learning in Graphs ## Primary Label Rules - If the contribution is a new sensor-specific full system, assign to `Sensor & Odometry Modalities`. - If the contribution is a new residual, association, registration, or loop detection module, assign to `Measurement Front-End`. - If the contribution is a solver, inference method, robust objective, or differentiable optimizer, assign to `Back-End Optimization & Inference`. - If the contribution is a representation of the world, assign to `Map Representations`. - If the contribution is mainly about benchmarks, metrics, runtime, or deployment reliability, assign to `Robustness, Evaluation & Operations`. - If the contribution uses learning as the core algorithmic novelty, assign to `Learning, Semantics & Spatial AI`; if learning is only a component inside a sensor pipeline, keep the primary label under that sensor or front-end branch.