SLAM Method Library Overview

This directory is the method-level SLAM library. It should help a reader answer four questions before opening any individual method page:

1. What role does SLAM play relative to odometry, map construction, and production localization?
2. Which method family should be evaluated for a given road, airside, warehouse, yard, port, mining, construction, agriculture, delivery, or campus environment?
3. Which benchmark and metric suite gives a fair result?
4. Which open-source stack is a reasonable starting point, and which is only a research reference?

For production AVs, the practical answer is usually not "run SLAM online forever." The stack should separate offline map construction, online scan-to-map localization, high-rate state estimation, and loop-closure/relocalization. Airside remains a useful reference ODD for this separation, but the same decision pattern also applies to road AVs, warehouses, logistics yards, ports, mines, construction sites, and campuses with different sensor and operational constraints.

Priority Ratings

Priority ratings are editorial reading and deployment triage signals. Learning answers what to read early for SLAM/localization understanding. Deployment answers what to evaluate early for AV deployment in the tagged context; it is not a certification, product-readiness, or all-domain average claim. If a method's deployment score is driven by a specific domain or stack role, the reason text should name that context.

Method	Rating	Stage	Maturity	Reason
Point-to-Point ICP for 3D SLAM and LiDAR Localization	Learning: ★★★★★ Deployment: ★★★★★	foundation	fielded-pattern	Core registration primitive behind LiDAR odometry and scan-to-map localization.
DO-Removal LIO	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	DO-Removal LIO is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
DOF-LIO Lightweight Dynamic Object Filter	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	DOF-LIO Lightweight Dynamic Object Filter is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
DR-REMOVER	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	DR-REMOVER is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
Dynamic-Object-Aware SLAM	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	Dynamic-Object-Aware SLAM is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
ERASOR	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	ERASOR is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
ERASOR++	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	ERASOR++ is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
FAST-LIO and FAST-LIO2	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	Core LiDAR-inertial baseline for mapping and localization fallback.
FAST-LIVO and FAST-LIVO2	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	FAST-LIVO and FAST-LIVO2 is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
FreeDOM Dynamic Object Removal	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	FreeDOM Dynamic Object Removal is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
GLIM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	GLIM is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
HDL Graph SLAM: 3D LiDAR-Based Graph SLAM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	HDL Graph SLAM: 3D LiDAR-Based Graph SLAM is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
KISS-Matcher	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	KISS-Matcher is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
KISS-SLAM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	KISS-SLAM is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
LeGO-LOAM: Lightweight and Ground-Optimized LiDAR Odometry and Mapping	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	LeGO-LOAM: Lightweight and Ground-Optimized LiDAR Odometry and Mapping is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
LiDAR Map Cleaning and Dynamic Removal	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	LiDAR Map Cleaning and Dynamic Removal is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
LIO-SAM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	Canonical factor-graph LIO reference for LiDAR, IMU, GPS, and loop factors.
LOAM: Lidar Odometry and Mapping in Real-time	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	LOAM: Lidar Odometry and Mapping in Real-time is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
LT-Mapper, Khronos, and Lifelong Mapping	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	LT-Mapper, Khronos, and Lifelong Mapping is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
LVI-SAM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	LVI-SAM is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
MapCleaner	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	MapCleaner is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
MOLA, MOLA-LO, and MOLA-LIO	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	MOLA, MOLA-LO, and MOLA-LIO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
MOVES and Label-Free Map Cleaning	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	MOVES and Label-Free Map Cleaning is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
Normal Distributions Transform (NDT) for 3D SLAM and AV Localization	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	fielded-pattern	Mature scan-to-map localization pattern used in AV and robotics stacks.
Omni-LIVO and Multi-LVI-SAM	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	Omni-LIVO and Multi-LVI-SAM is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
Point-LIO	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	Point-LIO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
R-POD Two-Stage Online Dynamic Removal LIO	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	R-POD Two-Stage Online Dynamic Removal LIO is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
Removert	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	Removert is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
RTMap, DUFOMap, and Recursive Map Maintenance	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	RTMap, DUFOMap, and Recursive Map Maintenance is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
Scan Context Family	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	fielded-pattern	Core LiDAR place-recognition pattern for loop closure and relocalization.
SD-SLAM Semantic Dynamic LiDAR	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	SD-SLAM Semantic Dynamic LiDAR is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
STATIC-LIO Dynamic Points Removal	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	STATIC-LIO Dynamic Points Removal is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
SuMa	Learning: ★★★★☆ Deployment: ★★★★★	modern-core	fielded-pattern	SuMa is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
TRLO Dynamic Tracking Removal LiDAR Odometry	Learning: ★★★★☆ Deployment: ★★★★★	deployment-pattern	pilot-proven	TRLO Dynamic Tracking Removal LiDAR Odometry is rated for dynamic-object filtering and map-cleaning workflows that protect localization maps.
Bundle Adjustment SLAM	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Bundle Adjustment SLAM is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
Continuous-Time Registration for LiDAR SLAM and AV Localization	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Continuous-Time Registration for LiDAR SLAM and AV Localization is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
CT-ICP: Continuous-Time ICP	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	CT-ICP: Continuous-Time ICP is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
Factor Graph SLAM with iSAM2 and GTSAM	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Backend pattern for smoothing, loop closure, and multi-sensor pose estimation.
FastSLAM and Particle SLAM	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	FastSLAM and Particle SLAM is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
GenZ-ICP and GenZ-LIO	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	GenZ-ICP and GenZ-LIO is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
GICP and VGICP for 3D SLAM and LiDAR Localization	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	GICP and VGICP for 3D SLAM and LiDAR Localization is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
GraphSLAM and Pose Graph Optimization	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Core graph formulation behind mapping, loop closure, and smoothing.
Learned LiDAR Place Recognition	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Learned LiDAR Place Recognition is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
LiDAR Bundle Adjustment Factors	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	LiDAR Bundle Adjustment Factors is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
Loop Closure and Place Recognition	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Loop Closure and Place Recognition is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
Occupancy Grid, TSDF, and ESDF Mapping	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Occupancy Grid, TSDF, and ESDF Mapping is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
Point-to-Plane ICP for 3D SLAM and LiDAR Localization	Learning: ★★★★★ Deployment: ★★★★☆	foundation	fielded-pattern	Point-to-Plane ICP for 3D SLAM and LiDAR Localization is rated for foundational SLAM modeling, optimization, registration, or mapping concepts.
BEV-LIO(LC)	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	BEV-LIO(LC) is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
Certifiable Pose Graph Optimization	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Certifiable Pose Graph Optimization is rated for robust or collaborative backend design in multi-session SLAM and validation.
CLIC and Coco-LIC	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	CLIC and Coco-LIC is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
CM-LIUW-Odometry	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	CM-LIUW-Odometry is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
COVINS and COVINS-G	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	COVINS and COVINS-G is rated for robust or collaborative backend design in multi-session SLAM and validation.
D2SLAM	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	D2SLAM is rated for robust or collaborative backend design in multi-session SLAM and validation.
Distributed Multi-Robot Pose Graph Optimization	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Distributed Multi-Robot Pose Graph Optimization is rated for robust or collaborative backend design in multi-session SLAM and validation.
Dynamic Map Cleaning Benchmarks	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	Dynamic Map Cleaning Benchmarks is rated as a SLAM benchmark or reference page for comparing methods and deployments.
Dynamic-Aware LIO BTSA	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Dynamic-Aware LIO BTSA is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
FusionPortableV2 Multi-Platform SLAM Dataset	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	FusionPortableV2 Multi-Platform SLAM Dataset is rated as a SLAM benchmark or reference page for comparing methods and deployments.
GEODE Degenerate LiDAR Benchmark	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	GEODE Degenerate LiDAR Benchmark is rated as a SLAM benchmark or reference page for comparing methods and deployments.
Ground-Fusion, M2DGR, and M3DGR	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	Ground-Fusion, M2DGR, and M3DGR is rated as a SLAM benchmark or reference page for comparing methods and deployments.
HeRCULES Radar Benchmark	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	HeRCULES Radar Benchmark is rated as a SLAM benchmark or reference page for comparing methods and deployments.
Kimera-Multi	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Kimera-Multi is rated for robust or collaborative backend design in multi-session SLAM and validation.
Kimera-RPGO and Pairwise Consistency Maximization	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Kimera-RPGO and Pairwise Consistency Maximization is rated for robust or collaborative backend design in multi-session SLAM and validation.
Kimera-VIO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Kimera-VIO is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
KISS-ICP: Keep It Small and Simple ICP	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	prototype	Strong LiDAR-only odometry baseline for evaluating registration stacks.
LiDAR-IMU Temporal Initialization	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	LiDAR-IMU Temporal Initialization is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
LIR-LIVO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	LIR-LIVO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
MA-LIO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	MA-LIO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
MM-LINS	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	MM-LINS is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
Multi-Agent Neural and Gaussian SLAM	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Multi-Agent Neural and Gaussian SLAM is rated for robust or collaborative backend design in multi-session SLAM and validation.
OKVIS2-X	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	OKVIS2-X is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
Open-Source SLAM Stack Comparison	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	Open-Source SLAM Stack Comparison is rated as a SLAM benchmark or reference page for comparing methods and deployments.
OpenVINS	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Practical VIO baseline for camera-IMU state estimation and fallback odometry.
PG-LIO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	PG-LIO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
R2LIVE and R3LIVE	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	R2LIVE and R3LIVE is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
Robust Pose Graph Optimization with GNC and riSAM	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Robust Pose Graph Optimization with GNC and riSAM is rated for robust or collaborative backend design in multi-session SLAM and validation.
Semantic-LiDAR-Inertial-Wheel Odometry	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Semantic-LiDAR-Inertial-Wheel Odometry is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
SLAM Benchmarking Metrics and Datasets	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	SLAM Benchmarking Metrics and Datasets is rated as a SLAM benchmark or reference page for comparing methods and deployments.
SLAM Decision Matrix for AV, Indoor, and Outdoor Systems	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	SLAM Decision Matrix for AV, Indoor, and Outdoor Systems is rated as a SLAM benchmark or reference page for comparing methods and deployments.
SNAIL Radar Benchmark	Learning: ★★★★☆ Deployment: ★★★★☆	reference	fielded-pattern	SNAIL Radar Benchmark is rated as a SLAM benchmark or reference page for comparing methods and deployments.
SPLIN ISDOR PPLIO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	SPLIN ISDOR PPLIO is rated for LiDAR odometry, mapping, or scan-matching coverage in AV localization stacks.
SVO	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	SVO is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
VINS-Mono and VINS-Fusion	Learning: ★★★★☆ Deployment: ★★★★☆	modern-core	fielded-pattern	Widely used visual-inertial baseline for GNSS-denied motion estimation.
4D Imaging Radar RIO and SLAM	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	4D Imaging Radar RIO and SLAM is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Cartographer 3D	Learning: ★★★☆☆ Deployment: ★★★★☆	classic-baseline	fielded-pattern	Mature submap SLAM reference for indoor and robotics mapping.
Radar Teach-Repeat Localization	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	Radar Teach-Repeat Localization is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Radar-Inertial Odometry	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	Radar-Inertial Odometry is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Radar-Inertial Online Temporal Calibration	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	Radar-Inertial Online Temporal Calibration is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Radar-to-LiDAR Map Localization	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	Radar-to-LiDAR Map Localization is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
RadarSplat-RIO	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	prototype	RadarSplat-RIO is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
RTAB-Map	Learning: ★★★☆☆ Deployment: ★★★★☆	deployment-pattern	fielded-pattern	Practical multi-sensor robotics SLAM stack with broad deployment use.
BundleFusion	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	BundleFusion is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
Co-SLAM and ESLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	Co-SLAM and ESLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
DPVO and DPV-SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	DPVO and DPV-SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
DROID-SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	DROID-SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
ElasticFusion	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	ElasticFusion is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
Event-Camera VIO and SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	Event-Camera VIO and SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
KinectFusion	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	KinectFusion is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
LSD-SLAM and DSO	Learning: ★★★★☆ Deployment: ★★★☆☆	classic-baseline	historical	LSD-SLAM and DSO are rated for direct visual SLAM foundations and camera-only fallback concepts.
MASt3R-SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	MASt3R-SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
Object-Level SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	Object-Level SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
ORB-SLAM2 and ORB-SLAM3	Learning: ★★★★☆ Deployment: ★★★☆☆	classic-baseline	fielded-pattern	Strong visual SLAM baseline, but not a primary AV localization backbone.
Semantic SLAM	Learning: ★★★★☆ Deployment: ★★★☆☆	modern-core	fielded-pattern	Semantic SLAM is rated for visual or visual-inertial SLAM coverage, especially fallback and GNSS-denied use.
GPR Localization and Ground Encoding	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	GPR Localization and Ground Encoding is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
GS-LIVM	Learning: ★★★☆☆ Deployment: ★★★☆☆	modern-core	prototype	GS-LIVM is rated as a supporting SLAM method for autonomy-stack triage and follow-up reading.
LO-Net Learned LiDAR Odometry	Learning: ★★★☆☆ Deployment: ★★★☆☆	modern-core	prototype	LO-Net Learned LiDAR Odometry is rated as a supporting SLAM method for autonomy-stack triage and follow-up reading.
Radar Odometry and Radar SLAM	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	Radar Odometry and Radar SLAM is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Radar Place Recognition: 4DRaL and SHeRLoc	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	Radar Place Recognition: 4DRaL and SHeRLoc is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
Radar-LiDAR-Inertial Fusion for Robust Odometry and SLAM	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	Radar-LiDAR-Inertial Fusion for Robust Odometry and SLAM is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
RegFormer Learned Registration	Learning: ★★★☆☆ Deployment: ★★★☆☆	modern-core	prototype	RegFormer Learned Registration is rated as a supporting SLAM method for autonomy-stack triage and follow-up reading.
ROMAN Object Map Alignment	Learning: ★★★☆☆ Deployment: ★★★☆☆	modern-core	prototype	ROMAN Object Map Alignment is rated as a supporting SLAM method for autonomy-stack triage and follow-up reading.
Super4DR	Learning: ★★★☆☆ Deployment: ★★★☆☆	modern-core	prototype	Super4DR is rated as a supporting SLAM method for autonomy-stack triage and follow-up reading.
Thermal-Inertial SLAM and Odometry	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	Thermal-Inertial SLAM and Odometry is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
UWB and Radio Ranging SLAM	Learning: ★★★☆☆ Deployment: ★★★☆☆	deployment-pattern	prototype	UWB and Radio Ranging SLAM is rated for alternative-sensor localization under adverse weather, weak LiDAR, or GNSS-denied conditions.
EKF-SLAM	Learning: ★★★★★ Deployment: ★★☆☆☆	foundation	historical	Foundation for estimator thinking, but rarely the direct modern AV stack.
4dNDF	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	4dNDF is rated for neural or Gaussian SLAM research and future dense map representation workflows.
Dynamic 4D Gaussian SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	Dynamic 4D Gaussian SLAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
Gaussian-LIC and Gaussian-LIC2	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	Gaussian-LIC and Gaussian-LIC2 is rated for neural or Gaussian SLAM research and future dense map representation workflows.
GigaSLAM: Large-Scale Monocular SLAM with Hierarchical Gaussian Splats	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	GigaSLAM: Large-Scale Monocular SLAM with Hierarchical Gaussian Splats is rated for neural or Gaussian SLAM research and future dense map representation workflows.
GS-SLAM and MonoGS	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	GS-SLAM and MonoGS is rated for neural or Gaussian SLAM research and future dense map representation workflows.
iMAP	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	iMAP is rated for neural or Gaussian SLAM research and future dense map representation workflows.
NeRF-SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	NeRF-SLAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
NICE-SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	NICE-SLAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
Photo-SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	Photo-SLAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
PIN-SLAM Neural LiDAR Mapping	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	PIN-SLAM Neural LiDAR Mapping is rated for neural or Gaussian SLAM research and future dense map representation workflows.
S3PO-GS	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	S3PO-GS is rated for neural or Gaussian SLAM research and future dense map representation workflows.
SLAM3R and VGGT Foundation SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	SLAM3R and VGGT Foundation SLAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
Splat-LOAM: Gaussian Splatting LiDAR Odometry and Mapping	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	Splat-LOAM: Gaussian Splatting LiDAR Odometry and Mapping is rated for neural or Gaussian SLAM research and future dense map representation workflows.
Splat-SLAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	Useful Gaussian SLAM reference, but not a runtime pose backbone.
SplaTAM	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	SplaTAM is rated for neural or Gaussian SLAM research and future dense map representation workflows.
VIGS-SLAM and VINGS-Mono	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	VIGS-SLAM and VINGS-Mono is rated for neural or Gaussian SLAM research and future dense map representation workflows.
WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments	Learning: ★★★☆☆ Deployment: ★★☆☆☆	frontier	research	WildGS-SLAM: Monocular Gaussian Splatting SLAM in Dynamic Environments is rated for neural or Gaussian SLAM research and future dense map representation workflows.

Domain Fit Guidance

Generic SLAM method pages should use Domain Fit, not Airside Fit, as the default deployment lens. Keep the section compact and focus on where method assumptions match the operating domain.

Domain	Fit	Note
Road AV	strong / conditional / weak / insufficient evidence	Check speed, map freshness, GNSS availability, and road-scale validation evidence.
Airside	strong / conditional / weak / insufficient evidence	Check open-apron geometry, aircraft/GSE dynamics, low-speed routes, FOD, and airport map operations.
Warehouse / logistics yard / port / mining / construction / agriculture / delivery robot / outdoor campus	strong / conditional / weak / insufficient evidence	Add only the domains where sensor, map, localization, and operational assumptions materially transfer.

Airside-specific pages may stay airside-first, but generic pages should not make airside the only deployment lens.

Repo Cross-Links

Topic	Read next	Why it matters for this library
Modern LiDAR odometry and SLAM front ends	LiDAR SLAM Algorithms	Detailed treatment of KISS-ICP, KISS-SLAM, MOLA, LIO-SAM, LVI-SAM, FAST-LIO2, FAST-LIVO2, R2LIVE/R3LIVE, CT-ICP, and Point-LIO.
Production scan-to-map localization	Production LiDAR Map Localization	Explains why production AV localization should usually match live scans to a prebuilt map instead of relying only on online SLAM.
Loop closure and kidnapped-robot recovery	LiDAR Place Recognition and Re-Localization and Scan Context Family	SLAM without robust place recognition becomes odometry with drift; these docs cover descriptor, retrieval, and verification pipelines.
Offline survey processing	Map Construction Pipeline	Shows where SLAM outputs become fleet-deployable HD maps, geodetic alignment, QA artifacts, and OTA packages.
Ego-state fusion and uncertainty	Robust State Estimation Multi-Sensor	SLAM factors must land in an estimator with sane gating, covariance, fallback, and sensor-fault behavior.
Factor graph foundations	GTSAM Factor Graphs and LiDAR Bundle-Adjustment Factors	The common backend language for LIO-SAM, map optimization, loop closure, IMU preintegration, LiDAR BA, and production smoothing.
Robust, certifiable, and collaborative backends	Robust PGO / GNC / riSAM, Kimera-Multi, COVINS/COVINS-G, D2SLAM	Adds robust graph optimization, certifiable PGO, PCM loop-closure validation, and collaborative SLAM systems as explicit backend families rather than hidden implementation details.
Alternative and degraded-sensor localization	UWB / Radio Ranging SLAM, Event-Camera VIO/SLAM, Thermal-Inertial SLAM, 4D Imaging Radar RIO/SLAM, Radar-to-LiDAR Map Localization	Covers GNSS-denied indoor/outdoor transitions, HDR/low-light operation, smoke/dust/night, and all-weather map localization fallbacks.
Dynamic map cleaning and object removal	LiDAR Map Cleaning and Dynamic Removal	Connects ERASOR, Removert, MapCleaner, ERASOR++, 4dNDF, FreeDOM, STATIC-LIO, temporal visibility, semantic masks, MOS/scene-flow evidence, and multi-session consensus to production map construction.
Lifelong and alternative localization	LT-Mapper, Khronos, and Lifelong Mapping	Connects recursive map maintenance, MOVES-style label-free map cleaning, GPR localization, and radar teach-repeat fallbacks for changed scenes and adverse weather.
Dense/neural scene representations	Photoreal City-Scale 4D Reconstruction and Gaussian Splatting for Driving	Connects Splat-SLAM, S3PO-GS, Gaussian-LIC, GS-LIVM, VIGS-SLAM, Dynamic 4D Gaussian SLAM, and RadarSplat-RIO to future dense mapping, semantic map QA, simulation, and photoreal 4D reconstruction.
Coverage audit and backlog	SLAM Coverage Audit and Backlog	Tracks missing first-class method pages found by parallel web-search agents so the library does not silently omit major techniques.

Scope Boundaries

Term	Primary output	Updates a map?	Drift behavior	Production role	Typical methods
Odometry	Relative pose stream	Local map only	Unbounded without correction	Fallback, prediction, survey front-end	KISS-ICP, FAST-LIO2, FAST-LIVO2, Point-LIO, CT-ICP
SLAM	Trajectory plus map	Yes	Bounded by loop closures and global constraints	Survey mapping, map repair, exploratory operation	LIO-SAM, LVI-SAM, KISS-SLAM, Cartographer, RTAB-Map, GLIM, MOLA
Localization	Pose in an existing map	No, except quality overlays	Bounded by map quality and scan matching	Normal AV runtime	Autoware NDT, VGICP, MOLA localization, KISS-Matcher, GTSAM scan-to-map factors
Relocalization	Global pose hypothesis	No	Recovers after losing track	Startup and fault recovery	Scan Context, MinkLoc3D, LCDNet, ICP/NDT verification
Mapping pipeline	Validated HD map package	Offline	Optimized globally	Airport onboarding and maintenance	Multi-session SLAM, GCP alignment, AMDB/Lanelet2 overlays

Canonical SLAM Pipeline

Stage	Main job	Common choices	Outputs to preserve	Failure signal
Time sync and calibration	Make sensors geometrically and temporally coherent	PTP/hardware sync, Kalibr, targetless LiDAR-camera calibration, extrinsic graph	Timestamp residuals, extrinsic covariance, sensor health	Rolling-shutter/deskew artifacts, inconsistent IMU gravity, scan "swimming"
Motion compensation	Remove intra-scan motion distortion	Constant-velocity deskew, IMU preintegration, continuous-time trajectory	Deskewed cloud plus deskew model used	Sharp poles become curved; ICP residual depends on azimuth
Front-end association	Create frame-to-frame or frame-to-map constraints	Point-to-point ICP, point-to-plane ICP, GICP/VGICP, NDT, visual reprojection, photometric residuals	Residuals, inlier masks, degeneracy eigenvalues	Low inlier ratio, anisotropic information matrix, poor convergence
Local mapping	Maintain bounded map for matching	Voxel hash, ikd-tree, iVox, submaps, surfels, occupancy grids	Keyframes/submaps, map timestamps, dynamic-object masks	Memory growth, stale dynamic objects, repeated walls/stands
Loop closure	Detect revisits and add constraints	Scan Context, M2DP, DBoW2/3, NetVLAD, MinkLoc3D, geometric verification	Candidate score, verified transform, covariance	Perceptual aliasing, false positive loop, topology tear
Back-end optimization	Solve global state consistency	GTSAM/iSAM2, g2o, Ceres, pose graph, factor graph, bundle adjustment	Full graph, marginalized priors, robust kernels	Graph jumps, bad loop factor dominates, overconfident covariance
Map export and QA	Convert research map to operational artifact	Dense point cloud, voxel map, occupancy grid, Lanelet2, Gaussian/mesh overlays	Map version, georeference transform, QA report	Double walls, map-frame drift, unbounded map entropy

Method Taxonomy

Family	Sensors	Core residual	Backend	Strengths	Weaknesses	Best fit
2D LiDAR grid SLAM	2D LiDAR, wheel odom, optional IMU	Scan correlation or point-to-line	Pose graph, occupancy grid	Simple, explainable, works well in planar indoor spaces	Cannot model ramps, multi-level structures, overhangs, aircraft geometry	Indoor mobile robots, AGVs, warehouses
3D LiDAR odometry	3D LiDAR	ICP/GICP/NDT	Local map, optional pose graph	Lighting independent, robust outdoors, simple failure metrics	Degenerate in open flat areas and repeated geometry; drift without loops	Survey validation, fallback odometry, outdoor robots
LiDAR-inertial odometry	3D LiDAR plus IMU	Point-to-plane or direct point residual with IMU propagation	IEKF, factor graph, smoother	Best real-time geometry-based odometry for fast motion and vibration	Requires tight sync and good IMU calibration; can double-count IMU if fused again naively	AV survey mapping, UAVs, handheld mapping, rough ground
Visual SLAM	Mono/stereo/RGB-D camera	Reprojection or photometric residual	Bundle adjustment, pose graph	Cheap sensors, rich semantics, mature sparse mapping	Lighting, weather, texture, motion blur, scale for monocular	Indoor AR, inspection, visual QA
Visual-inertial odometry	Camera plus IMU	Reprojection plus IMU preintegration/MSCKF	EKF, sliding window, factor graph	High rate, scale observable with IMU, compact maps	Initialization and calibration sensitive; degraded by low texture	Drones, handheld, camera-rich indoor robots
RGB-D dense SLAM	RGB-D camera	ICP plus photometric/depth residual	Pose graph, TSDF/surfels	Dense indoor maps, object-level QA	Range limited, sunlight interference, not AV-range	Indoor mapping, manipulation, labs
Radar SLAM	FMCW radar, optional IMU	Correlation, learned descriptors, Doppler constraints	Pose graph/factor graph	Weather and dust robust; long-range structure	Lower angular resolution; clutter and multipath	All-weather outdoor localization research
Neural/Gaussian SLAM	RGB-D, RGB, LiDAR-camera	Rendering loss, depth loss, learned features	Differentiable optimization plus pose graph	Dense appearance maps, semantic rendering, simulation reuse	Compute-heavy, immature uncertainty, hard certification	Offline QA, simulation assets, inspection overlays

Practical Selection Guidance

Environment	First-choice family	Good candidate pages	Add-ons that make it production-ready	Avoid as primary source when
Airside AV apron, known map	LiDAR-to-map localization plus state estimation	Production LiDAR Map Localization, KISS-ICP, LIO-SAM	RTK/GCP anchored maps, GTSAM scan-to-map factors, place recognition, degeneracy gating	Treating online SLAM as the only global reference; open tarmac is geometrically weak
Road AV in mapped ODD	Prebuilt HD map localization plus LiDAR/radar/camera odometry	FAST-LIO2, GLIM, Autoware NDT	GNSS/INS fusion, dynamic object removal, map versioning, online map-change detection	Using indoor RGB-D or monocular-only SLAM for safety-critical pose
Indoor warehouse, planar floors	2D LiDAR SLAM or 3D LiDAR-inertial if tall racks matter	SLAM Toolbox, Cartographer, RTAB-Map	Wheel odometry, reflector/AprilTag anchors, floor-zone maps, periodic relocalization	Forklifts or racks create persistent dynamic clutter without filtering
Underground/construction	3D LiDAR-inertial with multi-session SLAM	FAST-LIO2, Point-LIO, GLIM, CT-ICP	Loop closure, cross-session map merging, robust kernels, lidar intensity if geometry repeats	Pure visual tracking in dust/dark or pure GNSS outdoors/indoors mixed
Outdoor campus/service robot	LiDAR-inertial plus place recognition	KISS-ICP, KISS-SLAM, LIO-SAM, MOLA	Long-term dataset validation, seasonal map maintenance, semantic/dynamic filtering	Assuming one sunny-day map covers all seasons and construction changes
UAV/handheld inspection	Visual-inertial or LiDAR-inertial depending payload	OpenVINS, VINS-Fusion, FAST-LIO2, Point-LIO	Rolling-shutter handling, aggressive-motion IMU validation, loop closure	Low-grade IMU plus unsynchronized camera/LiDAR under fast motion

Airside SLAM Architecture Recommendation

Layer	Recommended role	Candidate implementation	Why
Survey odometry	Generate per-session trajectories and submaps	FAST-LIO2 or GLIM as primary; KISS-ICP as independent check	LIO gives robust deskewing and fast motion handling; LiDAR-only validation catches IMU/calibration-specific failures.
Loop closure	Correct drift across long apron loops and repeated passes	LIO-SAM loop module, KISS-SLAM, KISS-Matcher, Scan Context from place-recognition library	Required to prevent multi-kilometer survey maps from accumulating meter-scale drift.
Global optimization	Fuse odometry, loop closure, GCP, RTK, and prior map constraints	GTSAM/iSAM2, see GTSAM Factor Graphs	The same factor-graph representation can be reused by map construction and runtime localization.
Production runtime pose	Match live LiDAR to the validated HD map	GPU VGICP/NDT plus Production LiDAR Map Localization	Bounded drift and calibrated uncertainty are more important than online map growth during normal operation.
Recovery	Reinitialize after startup, tow, GPS loss, or bad scan matching	LiDAR Place Recognition and Re-Localization plus ICP/NDT verification	Avoids blindly trusting a local optimizer when the initial pose is wrong.
Dense/semantic QA	Inspect map quality and create simulation/visualization artifacts	Gaussian Splatting for Driving	Valuable for map QA and digital twins; not yet mature enough as the certified pose backbone.

Decision Rules

Rule	Practical test	Reason
Prefer localization over online SLAM when a validated map exists	Can the vehicle load a current map tile with known georeference and uncertainty?	Runtime map growth creates certification and fleet-consistency problems; localization against a validated map is easier to monitor.
Treat LiDAR-only odometry as a drift source, not a global truth source	Disable loop closures and run a 1-3 km loop; measure closure error	Even excellent odometry drifts; loop closure or external anchors are required for maps.
Add IMU only if timing and extrinsics are under control	Compare deskewed pole/edge sharpness and IMU residual statistics	Bad sync makes LIO worse than LiDAR-only ICP because errors become systematic.
Publish covariance that reflects degeneracy	Inspect Hessian eigenvalues, innovation gates, and scan-matching score distributions	Overconfident bad factors corrupt GTSAM more severely than missing factors.
Benchmark on target-like negatives	Include open aprons, repeated gates, wet tarmac, aircraft changes, night, rain, and GPS multipath	Public datasets rarely include airport-specific perceptual aliasing and dynamic aircraft/GSE clutter.
Keep method evaluation separate from product stack selection	Evaluate accuracy, then license, ROS version, maintenance, and compute	A top leaderboard method can still be unusable because of GPL obligations, missing ROS 2 support, or fragile calibration assumptions.

Failure Modes and Mitigations

Failure mode	Common in	Symptom	Detection	Mitigation
Planar/open-space degeneracy	Aprons, parking lots, long corridors	Pose update is confident along observable axes but unconstrained laterally/yaw	Small Hessian eigenvalues, high condition number, low vertical/edge diversity	Inflate covariance, fuse GNSS/wheel/IMU, require landmarks or map priors
Dynamic-object contamination	Airports, roads, warehouses	Map contains ghost aircraft, trucks, pallets, pedestrians	Repeated observations disagree by time/session; semantic dynamic masks	Dynamic filtering, temporal occupancy, multi-session consensus
Perceptual aliasing	Similar gates, warehouse aisles, tunnels	False loop closure or wrong relocalization	Descriptor top-K ambiguity, failed geometric verification	Use geometry verification, semantic priors, map-zone constraints, robust loop kernels
Time-sync error	LIO, VIO, multi-LiDAR rigs	Curved poles, residual depends on scan angle, IMU bias grows	Deskew residual by azimuth, calibration replay, timestamp diagnostics	Hardware sync/PTP, online temporal calibration, reject suspect sensors
Extrinsic drift or mounting flex	Multi-LiDAR AVs, handheld rigs	Per-sensor clouds disagree after motion	Cross-sensor residuals, loop-consistency by sensor	Rigid mounting, periodic targetless calibration, per-sensor health factors
Poor visual texture or lighting	Indoor/off-road/airside night glare	Visual tracker loses features or scale	Track count, reprojection error, exposure/blur metrics	Prefer LiDAR/radar, active illumination, inertial propagation, visual only as auxiliary
Map staleness	Long-term AV deployments	Good odometry but poor map matching in changed zones	Local residual clusters, change detection, fleet disagreement	Map-change workflow, dynamic layers, AIRAC/survey updates

Method Pages This Library Should Contain

Page	Method class	Primary question it should answer
KISS-ICP	LiDAR-only odometry	How far can a simple point-to-point ICP pipeline go, and when is it the right baseline?
KISS-SLAM	LiDAR-only SLAM	When is a lightweight LiDAR-only SLAM system enough for survey mapping?
KISS-Matcher	Global point-cloud registration	When can a robust matcher support relocalization, loop verification, or map merging?
LIO-SAM	Factor-graph LiDAR-inertial SLAM	How should LiDAR, IMU, GPS, and loop factors be structured in GTSAM?
LVI-SAM	LiDAR-visual-inertial SLAM	When does adding visual information to LiDAR-inertial smoothing improve robustness?
FAST-LIO2	Direct LiDAR-inertial odometry	When is a tightly coupled IEKF front-end the best real-time mapper?
FAST-LIVO2	Direct LiDAR-inertial-visual odometry	When should a stack use camera constraints with FAST-LIO-style direct mapping?
R2LIVE/R3LIVE	LiDAR-inertial-visual reconstruction	When are colorized maps and dense LIV reconstruction useful for survey QA?
Faster-LIO family	iVox LiDAR-inertial odometry	What are the speed/accuracy trade-offs of incremental voxels versus trees?
Point-LIO	High-bandwidth point-wise LIO	When do aggressive motion and high-rate control justify point-level updates?
CT-ICP	Continuous-time LiDAR odometry	How should a method model intra-scan motion without relying on IMU?
Cartographer	Submap and branch-and-bound SLAM	Why is it still relevant for 2D/3D submap SLAM and loop closure?
RTAB-Map	RGB-D/visual/LiDAR graph SLAM	When is a mature multi-sensor robotics stack more useful than a leaderboard method?
ORB-SLAM3	Visual and visual-inertial SLAM	What is the strongest sparse feature baseline for cameras?
OpenVINS	Filter-based VIO	When is MSCKF-style VIO preferable to full bundle adjustment?
GLIM	Range-inertial factor-graph mapping	How do GPU scan-matching factors, GTSAM, and manual map correction fit together?
MOLA	Modular LiDAR odometry, mapping, and localization	When is a ROS 2-ready modular mapping/localization framework useful?
Autoware NDT	Production scan-to-map localization	What can the AV open-source ecosystem teach about diagnostics and integration?
Scan Context Family, LiDAR Bundle-Adjustment Factors	Loop-closure and LiDAR backend factors	How should LiDAR descriptors and LiDAR-specific BA factors support relocalization, map refinement, and offline QA?
Robust PGO / GNC / riSAM, Certifiable Pose-Graph Optimization, Kimera-RPGO / PCM, Distributed Multi-Robot PGO, Kimera-Multi, COVINS/COVINS-G, D2SLAM	Robust and collaborative SLAM backends	How should loop-closure outliers, high-outlier graph optimization, certifiable initialization, and multi-session/multi-robot SLAM systems be handled?
ERASOR, Removert, MapCleaner, ERASOR++, 4dNDF, FreeDOM, STATIC-LIO, Dynamic Map Cleaning Benchmarks, LiDAR Map Cleaning and Dynamic Removal	Dynamic map cleaning	How should a static operational map be built from dynamic scenes without deleting valid structure or preserving ghost objects?
LT-Mapper / Khronos, RTMap / DUFOMap, GPR Localization, Radar Teach-Repeat Localization, MOVES	Lifelong map maintenance and alternative localization	How should a robot maintain maps across long-term changes, localize when visual/LiDAR assumptions degrade, and clean static-but-wrong objects without labels?
OKVIS2-X, MM-LINS, Event-Camera VIO/SLAM, Thermal-Inertial SLAM, UWB / Radio Ranging SLAM	Alternative sensor and degraded-scene SLAM	Which non-standard sensors or robustness mechanisms help in GNSS-denied, low-light, smoke/dust, corridor, or weak-geometry settings?
4D Imaging Radar RIO/SLAM, Radar-to-LiDAR Map Localization	Radar localization and cross-modal map matching	How should radar support all-weather odometry or localization against existing LiDAR maps?
Splat-SLAM, S3PO-GS	Gaussian visual SLAM	What can RGB-only Gaussian maps do, and why are scale and uncertainty still limiting?
Gaussian-LIC, GS-LIVM, VIGS-SLAM	Multi-sensor Gaussian SLAM	How do LiDAR, camera, and IMU constraints stabilize neural/Gaussian maps?
Dynamic 4D Gaussian SLAM, RadarSplat-RIO	Dynamic/radar Gaussian SLAM	How should dynamic scenes and radar measurements be handled before these maps are trusted?

Key Takeaways

Takeaway	Practical meaning
SLAM is a mapping and correction system, not a substitute for a production localization architecture.	Use it to build and maintain maps; use scan-to-map localization for normal mapped operation.
The front-end is environment-dependent; the backend pattern is reusable.	ICP/NDT/visual residuals differ, but factor graphs, robust kernels, covariance, and loop closures recur.
Public benchmark wins do not imply airside readiness.	Airports need repeated-stand negatives, dynamic aircraft/GSE changes, wet tarmac, night lighting, and geodetic map QA.
Removal belongs in both perception and mapping.	Online denoising protects detection and tracking; offline dynamic-object removal protects map quality, localization residuals, and long-term change detection.
The safest baseline is hybrid.	LiDAR-inertial for survey, LiDAR-only validation, GTSAM global optimization, place recognition for loop/recovery, and production VGICP/NDT for runtime localization.
Neural/Gaussian SLAM is a map-representation opportunity, not yet the pose backbone.	Use it for dense QA, semantic overlays, and simulation until uncertainty and failure monitoring mature.

Sources

• Vizzo et al., "KISS-ICP: In Defense of Point-to-Point ICP - Simple, Accurate, and Robust Registration If Done the Right Way", IEEE RA-L 2023, and official repo: https://github.com/PRBonn/kiss-icp
• Guadagnino et al., "KISS-SLAM: A Simple, Robust, and Accurate 3D LiDAR SLAM System With Enhanced Generalization Capabilities", arXiv 2025, and official repo: https://github.com/PRBonn/kiss-slam
• Shan et al., "LIO-SAM: Tightly-coupled Lidar Inertial Odometry via Smoothing and Mapping", IROS 2020, and official repo: https://github.com/TixiaoShan/LIO-SAM
• Xu et al., "FAST-LIO2: Fast Direct LiDAR-inertial Odometry", IEEE T-RO 2022, and official repo: https://github.com/hku-mars/FAST_LIO
• Bai et al., "Faster-LIO: Lightweight Tightly Coupled Lidar-Inertial Odometry Using Parallel Sparse Incremental Voxels", IEEE RA-L 2022, and official repo: https://github.com/gaoxiang12/faster-lio
• He et al., "Point-LIO: Robust High-Bandwidth Lidar-Inertial Odometry", and official repo: https://github.com/hku-mars/Point-LIO
• Deschaud, "CT-ICP: Real-time Elastic LiDAR Odometry with Loop Closure", ICRA 2022, and official repo: https://github.com/jedeschaud/ct_icp
• Google Cartographer official documentation: https://google-cartographer.readthedocs.io/
• RTAB-Map official project: https://introlab.github.io/rtabmap/
• Campos et al., "ORB-SLAM3: An Accurate Open-Source Library for Visual, Visual-Inertial and Multi-Map SLAM", IEEE T-RO 2021, and official repo: https://github.com/UZ-SLAMLab/ORB_SLAM3
• Geneva et al., "OpenVINS: A Research Platform for Visual-Inertial Estimation", ICRA 2020, and official docs: https://docs.openvins.com/
• Koide et al., "GLIM: 3D Range-Inertial Localization and Mapping with GPU-Accelerated Scan Matching Factors", Robotics and Autonomous Systems 2024, and official repo: https://github.com/koide3/glim
• GTSAM official docs and repo: https://gtsam.org/docs/ and https://github.com/borglab/gtsam
• Autoware NDT scan matcher official documentation: https://autowarefoundation.github.io/autoware_core/pr-602/localization/autoware_ndt_scan_matcher/
• SplaTAM and Gaussian SLAM references: https://arxiv.org/abs/2312.02126 and https://github.com/google-research/Splat-SLAM