SLAM Coverage Audit and Backlog
This audit exists to prevent method-library blind spots like the missing dedicated GLIM page. It consolidates parallel web-search agent findings across LiDAR SLAM, LiDAR-inertial odometry, visual SLAM, dense/RGB-D SLAM, LiDAR-visual-inertial SLAM, radar SLAM, registration, loop closure, and optimization backends.
The current SLAM library is useful, but it is not complete. Treat the P0 backlog below as required coverage before calling the method-level SLAM section comprehensive.
Current Status
| Item | Status |
|---|---|
| Dedicated GLIM page | Added as GLIM. |
| GTSAM coverage | Present in Factor Graph SLAM with iSAM2 and GTSAM and GTSAM Factor Graph Optimization. |
| Latest promotion status | The 2026-05-09 deep-dive waves added MOLA, KISS-SLAM, KISS-Matcher, LVI-SAM, FAST-LIVO/FAST-LIVO2, R2LIVE/R3LIVE, Splat-SLAM, S3PO-GS, Gaussian-LIC, GS-LIVM, VIGS-SLAM, Dynamic 4D Gaussian SLAM, RadarSplat-RIO, MapCleaner, ERASOR++, 4dNDF, FreeDOM, STATIC-LIO dynamic-point removal, dynamic-map cleaning benchmarks, robust/certifiable PGO, Kimera-RPGO/PCM, distributed multi-robot PGO, LT-mapper/Khronos, RTMap/DUFOMap, GPR localization, radar teach-repeat, MOVES, Scan Context family, LiDAR bundle-adjustment factors, Kimera-Multi, COVINS/COVINS-G, D2SLAM, UWB/range SLAM, OKVIS2-X, MM-LINS, event-camera VIO/SLAM, thermal-inertial SLAM, 4D imaging-radar RIO/SLAM, and radar-to-LiDAR map localization. The 2026-05-14 update promoted RMGS-SLAM. The recurring loop promoted COSMO-Bench, SLAM Toolbox, NDT Variants and NDT Maps, LaMAria, Hilti x Trimble SLAM Challenge 2026, ScaleMaster, HI-SLAM2, SEGS-SLAM, Neural/Gaussian SLAM Surveys, Semantic SLAM, Dynamic-Object-Aware SLAM, Multi-Agent Neural Gaussian SLAM, Object-Level SLAM, GV-iRIOM, GVINS/GLIO raw GNSS factor fusion, Wheel Odometry and Vehicle Motion Factors, CAO-RONet, Doppler Radar-LiDAR SLAM, Radar RIO Correspondence and Uncertainty, Lifelong 3D Map Version Control, Potentially Dynamic Object Removal by Ground Projection, Uni-Mapper Dynamic-Aware LiDAR Map Merging, LAMM Multi-Session Point-Cloud Map Merging, MapEval Point-Cloud Map-Quality Evaluation, BeautyMap, and Raymoval, broadened Radar-Inertial Online Temporal and Spatio-Temporal Calibration, refreshed Radar Place Recognition: 4D Radar Descriptor Lineage, and refreshed the benchmark matrix for Oxford Spires, IILABS 3D, SMapper-light, FusionPortableV2, and S3E. |
| Most severe structural gap | GaussianFlow-SLAM if code/source maturity improves, robust-loss uncertainty follow-ons, and sensor/foundation calibration pages still need more first-class files. Robust-loss covariance consistency is now covered in the state-estimation knowledge base. Raw GNSS factor fusion is now covered through GVINS/GLIO, wheel/vehicle-motion factors are now covered through a factor-family page, CAO-RONet covers source-mature learned 4D radar odometry, Doppler/radar-LiDAR bridge methods are now covered through Doppler Radar-LiDAR SLAM, radar point-correspondence and point-uncertainty routing is now covered through Radar RIO Correspondence and Uncertainty, radar place-recognition descriptors are refreshed through the 4D radar descriptor lineage page, and benchmark aliases are matrix-routed before dedicated pages are created. |
| How to use this audit | Add P0 files first, then P1, then P2 or mention-only aliases. Update this audit whenever a missing method is promoted into the library. |
Deep-Dive Promotion Wave (2026-05-09)
The May 9 loop took the highest-value "mentioned but not atomic" SLAM gaps and created dedicated method pages.
| Cluster | Promoted method files |
|---|---|
| Production-relevant LiDAR and LIVO | MOLA, KISS-SLAM, KISS-Matcher, LVI-SAM, FAST-LIVO/FAST-LIVO2, R2LIVE/R3LIVE |
| Gaussian and neural SLAM | Splat-SLAM, S3PO-GS, Gaussian-LIC, RMGS-SLAM, GS-LIVM, VIGS-SLAM, Dynamic 4D Gaussian SLAM |
| Radar and adverse-weather Gaussian mapping | RadarSplat-RIO |
Dynamic Map Cleaning Promotion Wave (2026-05-09)
The removal-focused loop promoted dynamic/static map cleaning from a broad synthesis topic into atomic method and benchmark files. The key distinction is between deleting transient moving points, suppressing dynamic-object trails, and rejecting static objects that do not belong in the persistent operational map.
| Cluster | Promoted files |
|---|---|
| Classical and learned map cleaners | MapCleaner, ERASOR++, 4dNDF, FreeDOM, BeautyMap, Raymoval, STATIC-LIO Dynamic-Point Removal |
| Benchmark and map-change bridge | Dynamic Map Cleaning Benchmarks, Moved-Object and Map-Change Datasets, Airside Dynamic Map-Cleaning Benchmark |
Static-Map Lifecycle Promotion (2026-05-23)
The recurring semantic-map loop expanded static-but-wrong map handling beyond single-session dynamic residual cleaning. The batch deliberately avoided duplicating ERASOR, Removert, MapCleaner, FreeDOM, STATIC-LIO, and MOVES, and instead promoted the missing lifecycle pieces: a reconstructable map-version-control architecture and a detector-based single-survey branch for parked or movable-class objects.
| Cluster | Promoted files |
|---|---|
| Static map lifecycle, heterogeneous-map merging, multi-session map merging, and movable-object quarantine | Lifelong 3D Map Version Control, Uni-Mapper Dynamic-Aware LiDAR Map Merging, LAMM Multi-Session Point-Cloud Map Merging, Potentially Dynamic Object Removal by Ground Projection |
Recurring LAMM Multi-Session Map-Merging Promotion (2026-05-24)
Five read-only scouts checked the post-Uni-Mapper queue for aggregated-map segmentation, ML-related SLAM, and dynamic/static residual cleanup. This bounded pass promoted LAMM because it has an IEEE RA-L 2025 paper, DOI metadata, an official HKU-MARS repository, and a non-duplicative large-scale multi-session point-cloud map-merging architecture that combines temporal bidirectional moving-object filtering, BTC loop discovery, false-positive loop filtering, connectivity grouping, and graph optimization.
| Cluster | Promoted files |
|---|---|
| Large-scale multi-session point-cloud map merging | LAMM Multi-Session Point-Cloud Map Merging |
Recurring MapEval Map-Quality Promotion (2026-05-24)
Five read-only scouts checked the post-LAMM aggregated-map queue and compared MapEval against LEMON-Mapping, MS-Mapping, and adjacent map-merging candidates. This bounded pass promoted MapEval because it has an IEEE RA-L 2025 paper, DOI metadata, an official repository, and a non-duplicative role as the direct point-cloud map-quality gate between map merging/dynamic cleanup and semantic segmentation. LEMON-Mapping remains a watchlist/comparison item until official code, license, or publication maturity improves.
| Cluster | Promoted files |
|---|---|
| Point-cloud map-quality evaluation | MapEval Point-Cloud Map-Quality Evaluation |
Recurring BeautyMap Promotion (2026-05-23)
Five read-only scouts checked the dynamic-removal and aggregated-map segmentation queues. This bounded pass promoted BeautyMap because it has a RA-L 2024 paper, official Python implementation, KTH DynamicMap Benchmark routing, and a distinct binary-encoded ground-matrix design that was previously mentioned only inside broader map-cleaning pages.
| Cluster | Promoted files |
|---|---|
| Binary-encoded dynamic-point removal | BeautyMap |
Recurring Raymoval Promotion (2026-05-24)
Five read-only scouts checked the remaining dynamic-removal queue. This bounded pass promoted Raymoval because it has a RiTA 2025 / arXiv 2026 paper, a distinct azimuth-elevation raycasting and spatial-validation design, and explicit benchmark numbers against Removert and ERASOR, while remaining paper-only and outside the KTH DynamicMap Benchmark list.
| Cluster | Promoted files |
|---|---|
| Raycasting dynamic-residual removal | Raymoval |
Recurring DynPurge / PRN Triage Refresh (2026-05-24)
Five read-only scouts checked whether the post-Raymoval queue justified another atomic map-cleaning page. The batch deliberately stayed at audit/backlog level. DynPurge has a verified IEEE RA-L 2025 / DBLP record and a public pseudocode/result-image repository, but no runnable implementation, license, release, machine-readable metric tables, or accessible full-paper details. PRN / Point Restoration Network is relevant to temporarily static object removal and occluded-region restoration, but current public evidence is paper-level and CARLA-oriented, so it remains a learned-restoration caveat rather than a map-publication method page.
| Cluster | Routing decision |
|---|---|
| DynPurge timestamp-distribution cleaning | Watchlist-only in the dynamic-cleaning hub and benchmark caveat page; promote only if full text, code, metric tables, or benchmark artifacts make the method non-duplicative. |
| PRN learned point restoration | Mention-only in static-but-transient cleanup; do not treat as a safety-critical permanent-map publication gate. |
Robust Backend and Lifelong Localization Promotion Wave (2026-05-09)
The latest web-gap loop promoted backend and long-term localization topics that were previously only backlog rows or hidden inside mapping documents.
| Cluster | Promoted files |
|---|---|
| Robust and certifiable backends | Robust PGO / GNC / riSAM, Certifiable Pose-Graph Optimization, Kimera-RPGO / PCM, Distributed Multi-Robot PGO |
| Lifelong and alternative localization | LT-Mapper / Khronos Lifelong Mapping, RTMap / DUFOMap Recursive Maintenance, GPR Localization, Radar Teach-Repeat Localization, MOVES Label-Free Map Cleaning |
Collaborative and Alternative-Sensor Promotion Wave (2026-05-09)
The latest gap loop promoted the strongest "mentioned but not atomic" SLAM families across loop closure, LiDAR-specific backends, collaborative SLAM, and alternative localization sensors.
| Cluster | Promoted files |
|---|---|
| Loop closure and LiDAR backends | Scan Context Family, LiDAR Bundle-Adjustment Factors |
| Collaborative SLAM systems | Kimera-Multi, COVINS/COVINS-G, D2SLAM |
| Alternative and degraded-sensor localization | UWB / Radio Ranging SLAM, OKVIS2-X, MM-LINS, Event-Camera VIO/SLAM, Thermal-Inertial SLAM, 4D Imaging Radar RIO/SLAM, Radar RIO Correspondence and Uncertainty, GV-iRIOM, GVINS/GLIO Raw GNSS Factor Fusion, Wheel Odometry and Vehicle Motion Factors, Radar-to-LiDAR Map Localization |
Recurring SLAM Benchmark Promotion (2026-05-22)
Five parallel research scouts checked perception, SLAM benchmarks, practical SLAM methods, sensor/foundation gaps, and dataset/validation gaps. The bounded SLAM promotion added the missing collaborative SLAM optimization benchmark page and marked already-promoted queue items as stale duplicates.
| Cluster | Promoted files |
|---|---|
| Collaborative SLAM backend benchmark | COSMO-Bench |
The remaining high-value SLAM queue is now visual/VIO benchmark freshness and fast-moving Gaussian/foundation SLAM coverage.
Recurring Practical SLAM Promotion (2026-05-23)
Five parallel research scouts checked practical SLAM methods, visual/VIO benchmarks, perception methods, sensor/foundation gaps, and stale queue entries. The bounded practical SLAM promotion added the missing ROS 2 indoor-SLAM reference and split NDT map/variant coverage from the existing core NDT page.
| Cluster | Promoted files |
|---|---|
| Practical 2D and NDT localization | SLAM Toolbox, NDT Variants and NDT Maps |
Next SLAM promotion queue: newer Gaussian/foundation SLAM variants where source maturity justifies atomic pages, specialized visual/VIO follow-ons beyond the now-promoted benchmark and wheel-factor pages, and radar place-recognition/odometry variants not already covered by the radar family pages.
Recurring Wheel/Vehicle-Motion Factor Promotion (2026-05-23)
Five parallel research scouts checked wheel/vehicle-motion factors, sparse/e2e planning follow-ons, collaborative Gaussian perception, embodied 3D benchmarks, and stale queue metadata. The bounded SLAM promotion added a factor-family page for wheel odometry, nonholonomic constraints, visual-inertial-wheel odometry, skid-steer calibration, and learned/adaptive wheel kinematic factors without duplicating the existing encoder primer, semantic-LIW page, CM-LIUW page, or raw GNSS factor page.
| Cluster | Promoted files |
|---|---|
| Vehicle-motion localization factors | Wheel Odometry and Vehicle Motion Factors |
Next SLAM promotion queue: source-mature radar place-recognition/odometry variants and GaussianFlow-SLAM if code/source maturity improves. Further wheel pages should be created only when they are narrower than the new factor-family page.
Recurring Learned 4D Radar Odometry Promotion (2026-05-23)
Five parallel research scouts checked CAO-RONet, adjacent radar odometry routes, GaussianFlow-SLAM maturity, sparse/e2e planning follow-ons, and source-registry freshness. The bounded SLAM promotion added CAO-RONet as a focused learned 4D radar odometry page because the arXiv paper and official repository are public and reviewable, while GaussianFlow-SLAM remained watchlist-only due to missing runnable code.
| Cluster | Promoted files |
|---|---|
| Learned 4D radar odometry | CAO-RONet |
Next SLAM promotion queue after this learned-radar pass: radar place-recognition variants, Doppler/radar-LiDAR bridge methods such as Radarize/Doppler-SLAM/DRO where they fit existing radar routing, and GaussianFlow-SLAM only after its repository contains runnable code, configs, setup instructions, and evaluation scripts. The following recurring pass resolved the Doppler bridge item into one page.
Recurring Doppler Radar Bridge Promotion (2026-05-23)
Five parallel research scouts checked Doppler radar/LiDAR SLAM routing, radar correspondence and uncertainty follow-ons, radar place-recognition/benchmark freshness, GaussianFlow-SLAM maturity, and sensor/calibration alternatives. The bounded SLAM promotion added one bridge page for Radarize, DRO, and Doppler-SLAM instead of splitting three partially overlapping pages. Radarize and DRO are source/code mature enough for comparison, while Doppler-SLAM remains paper-backed but code-maturity-limited because the public repository still marks usage/source release as incomplete.
| Cluster | Promoted files |
|---|---|
| Doppler radar and radar-LiDAR SLAM bridge | Doppler Radar-LiDAR SLAM |
Next SLAM promotion queue: radar point-correspondence and uncertainty methods are resolved by the following combined promotion, and radar place-recognition variants are resolved by the later descriptor-lineage refresh. Keep GaussianFlow-SLAM watchlist-only until its repository contains runnable code, configs, setup instructions, and evaluation scripts.
Recurring Radar RIO Correspondence/Uncertainty Promotion (2026-05-23)
Five parallel research scouts checked radar point-correspondence learning, point-uncertainty radar SLAM, radar place-recognition freshness, and adjacent sensor/foundation gaps. The bounded SLAM promotion added one combined radar RIO hardening page instead of creating two duplicate pages for correspondence and uncertainty. The page routes radar_transformer, HKUST point-uncertainty RIO, continuous point-pose uncertainty, and UNRIO as a front-end/backend confidence-weighting cluster.
| Cluster | Promoted files |
|---|---|
| Radar RIO association and uncertainty | Radar RIO Correspondence and Uncertainty |
Next SLAM promotion queue: radar place-recognition variants beyond the existing 4DRaL/SHeRLoc page are resolved by the following descriptor-lineage refresh. GaussianFlow-SLAM remains watchlist-only until its repository contains runnable code, configs, setup instructions, and evaluation scripts. Robust-loss uncertainty/foundation pages remain queued from the cross-architecture backlog.
Recurring Radar Place-Recognition Refresh (2026-05-23)
Five parallel research scouts checked TransLoc4D, 4D RadarPR, TDFANet, DIDLM, navigation deduplication, and robust-loss uncertainty fallback scope. The bounded batch refreshed the existing radar place-recognition page instead of creating duplicate one-method pages. TransLoc4D has author-maintained research code, 4D RadarPR is peer-reviewed but code-maturity-limited, TDFANet is ICRA 2025 source-backed but public-code URL unclear, and DIDLM is routed as a dataset/benchmark watch item rather than a descriptor method.
| Cluster | Refreshed files |
|---|---|
| 4D radar place-recognition descriptor lineage | Radar Place Recognition: 4D Radar Descriptor Lineage |
Next SLAM promotion queue: GaussianFlow-SLAM only after runnable code/configs/evaluation scripts are public, robust-loss uncertainty follow-ons beyond the now-promoted covariance-consistency page, and non-duplicative benchmark refreshes where a dataset needs deeper evaluation than the matrix.
Recurring Robust-Loss Covariance Foundation Promotion (2026-05-23)
Five parallel research scouts checked Ceres/GTSAM covariance semantics, robust M-estimator covariance theory, repo deduplication, adjacent page structure, and competing queue items. The bounded batch added one state-estimation foundation page rather than a SLAM method page, because the gap applies across SLAM, VIO, calibration, localization fusion, and tracking.
| Cluster | Promoted file |
|---|---|
| Robust covariance reporting and release evidence | Robust-Loss Covariance Consistency |
Next SLAM promotion queue: GaussianFlow-SLAM remains watchlist-only until its repository contains runnable code, configs, setup instructions, and evaluation scripts. Robust-loss uncertainty follow-ons should only be added if they are narrower than covariance reporting after robust weights, gates, and GNC.
Recurring Visual/VIO Benchmark Promotion (2026-05-23)
Five parallel research scouts checked visual/VIO benchmarks, Gaussian/foundation SLAM candidates, perception occupancy follow-ons, sensor/foundation gaps, and stale queue metadata. The bounded benchmark promotion added current visual-inertial and monocular scale-evaluation pages instead of expanding another method family.
| Cluster | Promoted files |
|---|---|
| Visual/VIO and monocular scale benchmarks | LaMAria, Hilti x Trimble SLAM Challenge 2026, ScaleMaster |
Recurring Gaussian/Foundation SLAM Promotion (2026-05-23)
Five parallel research scouts checked Gaussian/foundation SLAM, perception occupancy follow-ons, sensor/foundation gaps, radar/alternative-sensor SLAM, and stale queue metadata. The bounded promotion added two current visual Gaussian SLAM method pages plus a taxonomy page so the fast-moving neural/Gaussian family can be routed without duplicating adjacent method pages.
| Cluster | Promoted files |
|---|---|
| Neural/Gaussian SLAM methods and taxonomy | HI-SLAM2, SEGS-SLAM, Neural/Gaussian SLAM Surveys |
Recurring Multi-Agent Neural/Gaussian SLAM Refresh (2026-05-24)
Five read-only scouts checked ML-related SLAM, collaborative Gaussian SLAM, source hygiene, and watchlist maturity. The bounded refresh kept the consolidated family page rather than creating duplicate one-method pages, added CoMA-SLAM and MAGS-SLAM routing, and downgraded the family maturity to frontier research because the current methods remain camera/RGB-D/RGB Gaussian pipelines rather than LiDAR-primary production localization stacks.
| Cluster | Refreshed files |
|---|---|
| Multi-agent neural/Gaussian SLAM family | Multi-Agent Neural and Gaussian SLAM now routes CoMA-SLAM's AAAI 2026 distributed 2D Gaussian surfel branch, MAGS-SLAM's RGB-only branch, Coko-SLAM's bandwidth branch, and the existing CP-SLAM/MAGiC-SLAM/MAC-Ego3D/GRAND-SLAM/MNE-SLAM lineage without creating duplicate atomic pages. |
Recurring Radar/Calibration SLAM Promotion (2026-05-23)
Five parallel research scouts checked radar/SLAM hardening, perception occupancy follow-ons, sensor/foundation gaps, SLAM benchmarks, and stale metadata. The bounded radar promotion added the missing GV-iRIOM page, broadened radar-inertial calibration from temporal-only to temporal plus extrinsic calibration, and kept CFEAR-TR in the existing radar teach-repeat page rather than creating a duplicate atomic file.
| Cluster | Promoted or refreshed files |
|---|---|
| Radar hardening and globally referenced mapping | GV-iRIOM, Radar-Inertial Online Temporal and Spatio-Temporal Calibration, Radar Teach-Repeat Localization |
Recurring Raw GNSS Factor Fusion Promotion (2026-05-23)
Five parallel research scouts checked GVINS/GLIO raw GNSS factor fusion, collaborative Gaussian perception, embodied 3D benchmarks, sparse/e2e perception follow-ons, and stale queue metadata. The bounded SLAM promotion added the missing raw-GNSS factor page and left GSCOOP, EmbodiedScan/MMScan, SparseDriveV2, and DiffusionDriveV2 queued for later perception/planning passes.
| Cluster | Promoted files |
|---|---|
| Raw GNSS factor fusion | GVINS and GLIO Raw GNSS Factor Fusion |
Recurring Benchmark Matrix Refresh (2026-05-23)
Five parallel research scouts checked SLAM benchmarks, perception occupancy, sensor/foundation candidates, world-model hygiene, and repository checks. This pass did not add atomic pages; it routed source-mature benchmark aliases into SLAM Benchmarking Metrics and Datasets so dataset selection can be reviewed without creating one-off pages for every public dataset.
| Cluster | Matrix-covered datasets |
|---|---|
| Current SLAM benchmark matrix | Oxford Spires, IILABS 3D, SMapper-light, FusionPortableV2, and S3E |
Second Latest-Method Sweep (2026-05-08)
This second sweep used parallel web-search agents plus direct spot checks for 2025-2026 SLAM methods. It focused on methods likely to be missed by older SLAM surveys: degeneracy-robust LIO, LiDAR-visual-inertial odometry, wheel/GNSS factors, 4D radar localization, Gaussian/foundation-model visual SLAM, and current benchmarks.
Interpretation:
- P0 latest means create a dedicated file before describing the SLAM library as current.
- P1 latest means create a dedicated file after P0s, or cross-link clearly from the closest existing method page.
- Watch means the technique is promising but venue status, code release, official source availability, or independent benchmark maturity is not strong enough yet.
- Venue claims are treated cautiously when only an arXiv comment, project page, or third-party index was found.
P0 Latest Additions and Upgrades
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
| Promoted: OKVIS2-X | OKVIS2-X | Multi-modal VI/LiDAR/GNSS SLAM | Modern open keyframe VI SLAM configurable with dense depth, LiDAR, GNSS, volumetric submaps, and map-alignment factors; good companion to OKVIS/ROVIO/BASALT and VILENS. | https://arxiv.org/abs/2510.04612, https://github.com/ethz-mrl/OKVIS2-X |
| Promoted: MM-LINS | MM-LINS | Degeneracy-robust LIO | Multi-map LiDAR-inertial system for over-degenerate environments such as warehouses, crowds, smoke, temporary LiDAR blindness, and long corridors. | https://arxiv.org/abs/2503.19506 |
| Promoted: PG-LIO | PG-LIO | Intensity/photometric-geometric LIO | Uses LiDAR intensity to stabilize self-similar tunnels and corridors; directly relevant to GPS-denied indoor/outdoor transitions and long feature-poor routes. | https://arxiv.org/abs/2506.18583 |
| Promoted: LIR-LIVO | LIR-LIVO | LiDAR-visual-inertial odometry | Lightweight LIVO using illumination-resilient learned features, LiDAR depth association, SuperPoint, and LightGlue; upgrades FAST-LIVO-style coverage for poor lighting. | https://arxiv.org/abs/2502.08676 |
| Promoted: Semantic LiDAR-Inertial-Wheel Odometry | Semantic LiDAR-Inertial-Wheel odometry | Vehicle/industrial multi-sensor odometry | Strong deployment signal from automated-port operation; important for low-speed industrial vehicles where wheel odometry, semantics, and LiDAR need to coexist. | https://arxiv.org/abs/2509.14999 |
| Promoted: GV-iRIOM | GV-iRIOM | GNSS/visual/4D-radar inertial odometry and mapping | Large-scale 4D radar SLAM extension combining radar inertial odometry, visual-inertial odometry, GNSS, loop closure, and map fusion for adverse weather. | https://www.sciencedirect.com/science/article/pii/S0924271625000449 |
| Covered in Radar Teach-Repeat Localization | CFEAR-Teach-and-Repeat | Radar-only teach-and-repeat localization | March 2026 radar-only localization pipeline with Boreas evaluation, reported 29 Hz operation, and strong adverse-weather relevance. | https://arxiv.org/abs/2603.06501 |
radarsplat-rio.md | RadarSplat-RIO | Radar bundle adjustment / Gaussian radar map | April 2026 radar-inertial method that frames radar SLAM drift reduction as local bundle adjustment over range-azimuth-Doppler data with Gaussian splatting. | https://arxiv.org/abs/2604.13492 |
| Promoted: RMGS-SLAM | RMGS-SLAM | LiDAR-inertial-visual Gaussian SLAM | April 2026 real-time multi-sensor 3DGS SLAM with feed-forward plus voxel-PCA Gaussian initialization and Gaussian-GICP loop closure on a global Gaussian map. | https://arxiv.org/abs/2604.12942 |
vggt-slam-plus-plus.md | VGGT-SLAM++ | Foundation-model visual SLAM | April 2026 VGGT-based SLAM with DEM graph construction, DINOv2 retrieval, and local optimization; belongs beside SLAM3R/VGGT/AIM-SLAM coverage. | https://arxiv.org/abs/2604.06830 |
vista-slam.md | ViSTA-SLAM | Foundation/pointmap monocular SLAM | Calibration-free monocular SLAM using lightweight pointmap/pose prediction and Sim(3) loop closure; relevant to the foundation-geometry SLAM lineage. | https://ganlinzhang.xyz/vista-slam/, https://github.com/zhangganlin/vista-slam, https://arxiv.org/abs/2509.01584 |
| Promoted: S3PO-GS | S3PO-GS | Outdoor monocular Gaussian SLAM | ICCV 2025 outdoor RGB-only Gaussian SLAM with global scale-consistent 3D Gaussian pointmaps; fills the outdoor 3DGS gap beyond indoor MonoGS/SplaTAM. | https://arxiv.org/abs/2507.03737, https://openaccess.thecvf.com/content/ICCV2025/html/Cheng_Outdoor_Monocular_SLAM_with_Global_Scale-Consistent_3D_Gaussian_Pointmaps_ICCV_2025_paper.html |
| Promoted: VIGS-SLAM | VIGS-SLAM | Visual-inertial Gaussian SLAM | Tightly couples visual, inertial, depth/pose optimization, and Gaussian mapping for blur, low texture, and exposure variation; promote out of the watchlist. | https://arxiv.org/abs/2512.02293, https://vigs-slam.github.io/ |
gaussianflow-slam.md | GaussianFlow SLAM | Monocular Gaussian SLAM | April 2026 monocular 3DGS SLAM using optical-flow-guided GaussianFlow constraints to regularize pose and structure. | https://arxiv.org/abs/2604.15612 |
| Promoted: HI-SLAM2 | HI-SLAM2 | Geometry-aware monocular Gaussian SLAM | T-RO 2025 RGB-only Gaussian SLAM with monocular priors, scale alignment, loop closure, and instant map updates. | https://hi-slam2.github.io/, https://doi.org/10.1109/TRO.2025.3626627 |
| Promoted: SEGS-SLAM | SEGS-SLAM | Structure-enhanced Gaussian SLAM | ICCV 2025 3DGS SLAM across monocular, stereo, and RGB-D inputs; useful for organizing the fast-moving Gaussian SLAM family. | https://segs-slam.github.io/ |
| Promoted: ScaleMaster | ScaleMaster | Learned monocular SLAM benchmark | February 2026 benchmark for scale consistency and map-to-map quality in deep monocular SLAM, especially large indoor and multi-floor sequences. | https://arxiv.org/abs/2602.18174 |
| Promoted: SNAIL Radar Benchmark | SNAIL Radar | 4D radar SLAM benchmark | IJRR 2025 radar benchmark with 4D radar, stereo, LiDAR, IMU, GNSS/INS, rain/night/tunnel coverage, and updated 2025 poses/extrinsics. | https://snail-radar.github.io/, https://arxiv.org/abs/2407.11705 |
| Promoted: HeRCULES Radar Benchmark | HeRCULES | Heterogeneous radar benchmark | ICRA 2025 benchmark combining 4D radar, spinning radar, FMCW LiDAR, IMU, GPS, and cameras for radar SLAM, place recognition, and fusion. | https://sites.google.com/view/herculesdataset, https://arxiv.org/abs/2502.01946 |
| Promoted: GEODE Degenerate LiDAR Benchmark | GEODE | Degenerate LiDAR benchmark | Large multi-LiDAR benchmark with many geometrically degenerate settings; directly supports the degeneracy-robust LIO backlog. | https://thisparticle.github.io/geode/, https://arxiv.org/abs/2409.04961 |
P1 Latest Additions
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
genz-icp.md | GenZ-ICP | Degeneracy-robust LiDAR odometry / ICP | Already mentioned in broader LiDAR docs, but now deserves a focused file because degeneracy-robust registration is a 2025-2026 hotspot. | https://github.com/cocel-postech/genz-icp |
bev-lio-lc.md | BEV-LIO-LC | LIO plus learned BEV loop closure | Bridges FAST-LIO-style odometry with BEV image place recognition for long warehouse, campus, and outdoor loops. | https://github.com/HxCa1/BEV-LIO-LC |
omni-livo.md | Omni-LIVO | Multi-camera LiDAR-visual-inertial odometry | Extends LIVO toward surround/multi-camera FoV coverage, a better fit for AV-style sensor suites than monocular LIVO. | https://arxiv.org/abs/2509.15673 |
| Covered in Wheel Odometry and Vehicle Motion Factors | Online neural LiDAR-IMU-wheel factor graph | Wheel/vehicle state estimation | Learns skid-steer or terrain-dependent wheel kinematics inside the estimator rather than treating wheel odometry as fixed noise; keep as part of the factor-family page unless a narrower implementation deep dive is needed. | https://arxiv.org/abs/2407.08907 |
| Promoted: CM-LIUW-Odometry | CM-LIUW-Odometry | LiDAR-IMU-UWB-wheel odometry | Underground/tunnel-focused fusion with nonholonomic constraints, lever-arm compensation, and adaptive mode switching. | https://arxiv.org/abs/2511.01379 |
| Covered in Radar Place Recognition: 4D Radar Descriptor Lineage | 4DRaL | 4D radar / LiDAR place recognition | LiDAR-teacher distillation for radar-to-radar and radar-to-LiDAR retrieval; important for all-weather map lookup and loop closure. | https://arxiv.org/abs/2603.26206 |
| Covered in Radar Place Recognition: 4D Radar Descriptor Lineage | SHeRLoc | Heterogeneous radar place recognition | Cross-modal place recognition across spinning radar, 4D radar, and FMCW LiDAR for mixed-sensor fleets. | https://arxiv.org/abs/2506.15175 |
| Covered in Radar Place Recognition: 4D Radar Descriptor Lineage | TransLoc4D / 4D RadarPR / TDFANet | 4D radar place-recognition descriptors | Sparse-convolution/Transformer, context-aware RCS attention, and sequential BEV aggregation variants are now routed together; create deeper pages only when a non-duplicative implementation or benchmark question appears. | https://openaccess.thecvf.com/content/CVPR2024/html/Peng_TransLoc4D_Transformer-based_4D_Radar_Place_Recognition_CVPR_2024_paper.html, https://github.com/phatli/TransLoc4D, https://doi.org/10.1016/j.isprsjprs.2025.01.033, https://arxiv.org/abs/2504.05103 |
| Promoted: CAO-RONet | CAO-RONet | Learning-based 4D radar odometry | 2025 ICRA line for low-quality 4D radar point completion and context-aware association, useful for AV radar odometry comparison. | https://arxiv.org/abs/2503.01438, https://github.com/NEU-REAL/CAO-RONet |
| Promoted: Radar RIO Correspondence and Uncertainty | Radar point correspondence learning and point uncertainty for RIO | Radar-inertial frontend/backend hardening | Transformer-based radar point correspondences, explicit point uncertainty, point-pose uncertainty, and learned velocity uncertainty are routed together to avoid duplicate radar-RIO assumption pages. | https://arxiv.org/abs/2506.18580, https://github.com/aau-cns/radar_transformer, https://arxiv.org/abs/2402.16082, https://github.com/HKUST-Aerial-Robotics/RIO |
| Covered in Radar-Inertial Online Temporal and Spatio-Temporal Calibration | Online radar-IMU spatial and temporal calibration | Radar-inertial deployment hardening | March 2026 continuous-time calibration for radar-inertial odometry, important for productizing radar fusion. | https://arxiv.org/abs/2603.19958 |
gs-livm.md | GS-LIVM | LiDAR-inertial-visual Gaussian mapping | ICCV 2025 real-time outdoor Gaussian mapping using LIVO poses, voxel GPR, and CUDA; compare with Gaussian-LIC and FAST-LIVO2. | https://openaccess.thecvf.com/content/ICCV2025/html/Xie_GS-LIVM_Real-Time_Photo-Realistic_LiDAR-Inertial-Visual_Mapping_with_Gaussian_Splatting_ICCV_2025_paper.html, https://github.com/xieyuser/GS-LIVM |
4dtam.md | 4DTAM | Dynamic/non-rigid 4D Gaussian SLAM | CVPR 2025 dynamic-surface-Gaussian tracking and mapping; lower AV-localization priority but important for dynamic Gaussian maps. | https://github.com/muskie82/4dtam |
ace-slam.md | ACE-SLAM | Neural implicit RGB-D SLAM | Scene-coordinate regression as live implicit map, compact relocalization angle; promote after code maturity is checked. | https://arxiv.org/abs/2512.14032, https://ialzugaray.github.io/ace-slam/ |
dropd-slam.md | DropD-SLAM | Monocular dense/RGB-D replacement | Uses pretrained metric depth, learned keypoints, and segmentation to drive an RGB-D-style backend from monocular video. | https://arxiv.org/abs/2510.06216 |
levio.md | LEVIO | Embedded visual-inertial odometry | Ultra-low-power VIO for RISC-V/embedded devices; useful if the corpus covers small robots or auxiliary wearable/camera nodes. | https://arxiv.org/abs/2602.03294 |
| Matrix-covered in SLAM Benchmarking Metrics and Datasets | IILABS 3D | Indoor LiDAR SLAM benchmark | Indoor wheeled-robot benchmark with heterogeneous 3D LiDARs, IMU, wheel odometry, and MoCap ground truth; create a dedicated file only if deeper sequence-by-sequence evaluation is needed. | https://jorgedfr.github.io/3d_lidar_slam_benchmark_at_iilab/ |
| Matrix-covered in SLAM Benchmarking Metrics and Datasets | SMapper / SMapper-light | Open-hardware SLAM benchmark platform | Reproducible acquisition platform with public indoor/outdoor sequences and offline LiDAR-SLAM-derived sub-cm ground truth; create a dedicated file only if platform replication details become the target. | https://snt-arg.github.io/smapper_docs/, https://arxiv.org/abs/2509.09509 |
agrilira4d-benchmark.md | AgriLiRa4D | UAV LiDAR/radar benchmark | Agricultural UAV benchmark with 3D LiDAR, 4D radar, IMU, and RTK; useful for vegetation, repetitive texture, and outdoor robustness. | https://zhan994.github.io/AgriLiRa4D/, https://arxiv.org/abs/2512.01753 |
diter-plus-plus-benchmark.md | DiTer++ | Multi-robot/multi-session benchmark | Legged-robot, thermal, RGB/RGB-D, LiDAR, IMU, and proprioception coverage for outdoor multi-session SLAM. | https://arxiv.org/abs/2412.05839 |
Latest Watchlist
| Method or technique | Why watch | Current concern | Sources |
|---|---|---|---|
| LTR^2 / LiDAR Teach, Radar Repeat | Very relevant cross-modal LiDAR-teach and 4D-radar-repeat navigation, with long-term deployment claims across smoke/night/changed environments. | Primary arXiv link surfaced in the third sweep, but keep watch until code and independent validation mature. | https://arxiv.org/abs/2605.02809 |
| CUBE-LIO | Intensity-assisted cubemap projection for degenerate LIO. | Venue listing found, but paper/repo not found in this sweep. | https://ras.papercept.net/conferences/conferences/ICRA26/program/ICRA26_ContentListWeb_4.html |
| R-VoxelMap | 2026 VoxelMap successor candidate. | Code/release maturity unclear. | https://arxiv.org/abs/2601.12377 |
| LIO-MARS | Non-uniform continuous-time B-spline LIO. | Submitted status; wait for venue/code. | https://arxiv.org/abs/2511.13985 |
| AKF-LIO | Adaptive covariance and Gaussian map for degraded/dynamic scenes. | Code appears not released yet. | https://arxiv.org/abs/2503.06891 |
| Super4DR, DNOI-4DRO, Equi-RO | Fast-moving 4D radar odometry and Gaussian/radar learning line. | Good technical signal, but code and repeated external evaluation are not yet mature. | https://arxiv.org/abs/2512.09608, https://arxiv.org/abs/2505.12310, https://arxiv.org/abs/2509.20674 |
| SCE-SLAM and GSO-SLAM | Scene-coordinate and Gaussian/direct-VO visual SLAM variants worth tracking. | Strong preprint signal, but not yet core AV 3D SLAM coverage without code/benchmark maturity checks. | https://arxiv.org/abs/2601.09665, https://arxiv.org/abs/2602.11714 |
| D-GVIO and CT-VIR | Decentralized GNSS-VIO and visual-inertial-ranging fusion. | Interesting for multi-agent or UWB work, but not yet core AV 3D SLAM coverage. | https://arxiv.org/abs/2603.01404, https://arxiv.org/abs/2604.14545 |
Third Gap-Discovery Sweep (2026-05-08)
This sweep deliberately looked outside the latest neural/Gaussian and LIO-heavy lanes. Five parallel agents searched for backend/tooling, specialized visual/VIO, LiDAR map representation, alternative-sensor, and collaborative/lifelong SLAM gaps. The result is a more structural backlog: the library is missing several mature method families, not just new papers.
P0 Discovered Families
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
| Promoted: Robust PGO / GNC / riSAM | Graduated Non-Convexity, Black-Rangarajan duality, riSAM | Robust graph optimization backend | Complements switchable constraints, DCS, max-mixtures, and robust kernels with modern high-outlier and incremental robust-estimation tools. | https://arxiv.org/abs/1909.08605, https://arxiv.org/abs/2210.07097 |
| Promoted: Certifiable Pose-Graph Optimization | SE-Sync and Shonan Averaging | Certifiable / globally initialized PGO | Explains when pose-graph solutions can be certified or initialized beyond local Gauss-Newton; important for offline map QA. | https://arxiv.org/abs/1612.07386, https://arxiv.org/abs/2003.01032, https://gtsam.org/doxygen/a05136.html |
| Promoted: Kimera-RPGO / PCM | Kimera-RPGO and Pairwise Consistency Maximization | Loop-closure verification / robust PGO | False loop closure is safety-critical in repeated stands, gates, corridors, and warehouses; PCM deserves first-class coverage. | https://github.com/MIT-SPARK/Kimera-RPGO, https://arxiv.org/abs/2003.12932, https://arxiv.org/abs/1711.08632 |
| Promoted: Distributed Multi-Robot PGO | DPGO and distributed certifiably correct PGO | Distributed / multi-session SLAM backend | Multi-vehicle and multi-session map merging are backend problems, not only front-end SLAM-system problems. | https://github.com/MIT-SPARK/DPGO, https://arxiv.org/abs/2011.11929 |
| Promoted: Kimera-Multi | Kimera-Multi | Distributed multi-robot metric-semantic SLAM | Peer-to-peer dense metric-semantic C-SLAM with distributed GNC; not covered by the existing Kimera-VIO page. | https://arxiv.org/abs/2106.14386, https://github.com/MIT-SPARK/Kimera-Multi |
| Promoted: COVINS/COVINS-G | COVINS and COVINS-G | Centralized/cloud collaborative VI-SLAM | Strong pattern for fleet/cloud VI map sharing; COVINS-G generalizes beyond one front end. | https://arxiv.org/abs/2108.05756, https://arxiv.org/abs/2303.16641, https://github.com/VIS4ROB-lab/covins |
| Promoted: D2SLAM | D2SLAM | Decentralized aerial-swarm VI-SLAM | Covers near-field relative localization plus far-field global consistency for aerial swarm and multi-agent systems. | https://github.com/HKUST-Aerial-Robotics/D2SLAM, https://ieeexplore.ieee.org/document/10582478 |
swarm-slam.md | Swarm-SLAM | Sparse decentralized C-SLAM | ROS 2, multi-sensor support, decentralized sparse map sharing; practical baseline for indoor, underground, and swarm robotics. | https://arxiv.org/abs/2301.06230, https://github.com/MISTLab/Swarm-SLAM |
slideslam.md | SlideSLAM | Decentralized metric-semantic multi-robot SLAM | Object-based metric-semantic maps for heterogeneous robot teams and bandwidth-constrained operation. | https://arxiv.org/abs/2406.17249 |
| Promoted: LT-Mapper / Khronos | LT-mapper / LT-SLAM / LT-removert / LT-map, Khronos | LiDAR lifelong and spatio-temporal map maintenance | Explicit multi-session alignment, dynamic filtering, positive/negative change handling, delta maps, and temporal semantic maps. | https://arxiv.org/abs/2107.07712, https://github.com/gisbi-kim/lt-mapper, https://roboticsproceedings.org/rss20/p081.pdf |
| Promoted: ROMAN Object Map Alignment | ROMAN | Object-map alignment / global localization | Open-set object map alignment for view-invariant localization and multi-robot loop closure. | https://acl.mit.edu/roman/, https://github.com/mit-acl/roman, https://www.roboticsproceedings.org/rss21/p029.pdf |
| Promoted: 4dNDF | 4D implicit LiDAR mapping / 4dNDF | Dynamic LiDAR map updating / static-map extraction | AV-relevant static map construction from dynamic LiDAR logs with spatio-temporal implicit filtering. | https://arxiv.org/abs/2405.03388, https://github.com/PRBonn/4dNDF |
| Promoted: Lifelong 3D Map Version Control | Lifelong 3D Mapping Framework | Cloud-native map version control | Adds explicit map versioning, change queries, positive/negative changes, and multi-session alignment. | https://arxiv.org/abs/2501.18110 |
| Promoted: Potentially Dynamic Object Removal by Ground Projection | No More Potentially Dynamic Objects | Static-map construction / detector-ground projection | Adds single-survey removal of parked or movable-class objects through LiDAR 3D detection, ground segmentation, and projection. | https://arxiv.org/abs/2407.01073, https://github.com/woo-soojin/no_more_potentially_dynamic_objects |
| Promoted: Uni-Mapper Dynamic-Aware LiDAR Map Merging | Uni-Mapper / DynaSTD | Dynamic-aware heterogeneous-LiDAR map merging | Adds dynamic-removal-aware descriptors and anchor-node graph optimization for multi-map merging across LiDAR modalities. | https://arxiv.org/abs/2507.20538, https://sparolab.github.io/research/uni_mapper/, https://github.com/sparolab/uni-mapper |
multi-lio.md | Multi-LIO | Multi-LiDAR inertial odometry | Multiple-LiDAR LIO with parallel updates, voxel maps, and point-wise uncertainty; directly relevant to AV rigs with 4-8 LiDARs. | https://dblp.org/pid/59/2133-1, https://lightingooo.github.io/ |
lta-om.md | LTA-OM | Long-term LIO / multisession localization | Adds long-term association mapping, loop correction/rejection, and multisession map reuse on top of FAST-LIO2-style mapping. | https://repository.hku.hk/handle/10722/353311, https://github.com/hku-mars/LTAOM |
coin-lio.md | COIN-LIO | Intensity-augmented degeneracy-robust LIO | Uses LiDAR intensity as complementary constraints in tunnels, flat fields, and weak-geometry environments. | https://arxiv.org/abs/2310.01235, https://github.com/ethz-asl/COIN-LIO |
x-icp-localizability-aware-registration.md | X-ICP | Localizability-aware LiDAR registration | Detects unobservable directions and constrains ICP accordingly; directly actionable for corridor/apron scan-to-map localization. | https://ieeexplore.ieee.org/document/10328716/, https://www.research-collection.ethz.ch/items/e286a52c-2afc-492f-8b00-a9ab295665be |
| Promoted: PIN-SLAM Neural LiDAR Mapping | PIN-SLAM | Neural implicit LiDAR/RGB-D SLAM | Strong neural LiDAR SLAM line with compact implicit maps, point-to-implicit registration, and neural point loop detection. | https://arxiv.org/abs/2401.09101, https://github.com/PRBonn/PIN_SLAM |
| Promoted: Event-Camera VIO/SLAM | ESVIO, ESVO2, EVI-SAM, PL-EVIO | Event-camera visual/event-visual-inertial SLAM | Missing whole family for HDR, low light, high-speed motion, blur, and agile robotics. | https://arxiv.org/abs/2212.13184, https://github.com/arclab-hku/ESVIO, https://arxiv.org/abs/2410.09374, https://github.com/NAIL-HNU/ESVO2, https://arxiv.org/abs/2312.11911, https://arxiv.org/abs/2209.12160 |
| Promoted: Thermal-Inertial SLAM | Graph-based Thermal-Inertial SLAM, TP-TIO, embedded TIO | Thermal / infrared SLAM and odometry | Directly fills night, smoke, dust, and low-light camera gaps for indoor, tunnel, emergency, and airside night operation. | https://arxiv.org/abs/2104.07196, https://github.com/risqiutama/ti-slam, https://arxiv.org/abs/2012.03455, https://arxiv.org/abs/2603.02114 |
multi-camera-fisheye-visual-inertial-slam.md | BAMF-SLAM, MAVIS/OpenMAVIS, BundledSLAM | Omnidirectional/fisheye/multi-camera VI-SLAM | AVs and indoor robots use surround/fisheye rigs, not only mono/stereo pinhole; synchronization and overlap are core estimator issues. | https://arxiv.org/abs/2306.01173, https://bamf-slam.github.io/, https://arxiv.org/abs/2309.08142, https://github.com/MAVIS-SLAM/OpenMAVIS, https://arxiv.org/abs/2403.19886 |
structure-plp-slam.md | Structure PLP-SLAM, PL-VINS, point-line-plane SLAM | Structural visual SLAM | Lines and planes matter in low-texture corridors, warehouses, parking garages, terminals, and indoor/outdoor built environments. | https://arxiv.org/abs/2207.06058, https://github.com/PeterFWS/Structure-PLP-SLAM, https://arxiv.org/abs/2009.07462, https://github.com/cnqiangfu/PL-VINS |
| Covered in Wheel Odometry and Vehicle Motion Factors | VINS-on-Wheels, VIWO, PIEKF-VIWO, PL-VIWO/2, VIPS-Odom | Visual-inertial-wheel odometry | Ground vehicles often have wheel encoders and nonholonomic constraints; the family is now routed through the vehicle-motion factor page unless one implementation needs a dedicated file. | https://mars.cs.umn.edu/papers/KejianWu_VINSonWheels.pdf, https://woosiklee.com/downloads/papers/Lee2020IROS.pdf, https://arxiv.org/abs/2303.07668, https://arxiv.org/abs/2503.00551, https://arxiv.org/abs/2509.21563, https://arxiv.org/abs/2407.05017 |
dynamic-visual-inertial-slam.md | DynaVINS, DynaVINS++, ADUGS-VINS, RD-VIO | Robust dynamic-scene VIO/VISLAM | Non-Gaussian dynamic-scene VIO line for moving people, vehicles, temporary objects, and false visual constraints. | https://arxiv.org/abs/2208.11500, https://github.com/url-kaist/dynaVINS, https://arxiv.org/abs/2410.15373, https://adugs-vins.github.io/, https://arxiv.org/abs/2310.15072 |
| Promoted: GPR Localization | Localizing Ground Penetrating Radar and Ground Encoding | GPR / underground radar localization | Targets rain, snow, fog, dust, darkness, visual-feature loss, and repeatable underground features through subsurface map matching and factor-graph localization. | https://tisl.cs.toronto.edu/publication/202005-ral-lgpr/ral20-lgpr.pdf, https://arxiv.org/abs/2107.07606, https://arxiv.org/abs/2103.15317 |
| Promoted: Radar Teach-Repeat Localization | Radar Teach and Repeat / VT&R3 radar route following | Radar teach-and-repeat | Broader adverse-condition route-repeat architecture than CFEAR-TR alone; released VTR3 code. | https://arxiv.org/abs/2409.10491, https://github.com/utiasASRL/vtr3 |
| Promoted: UWB / Radio Ranging SLAM | UWB Radar SLAM / anchorless UWB radar SLAM | UWB / RF SLAM | Distinct from fixed-anchor RTLS; relevant to warehouses, tunnels, mines, and industrial corridors. | https://arxiv.org/abs/2311.14970 |
| Promoted: COSMO-Bench | COSMO-Bench | C-SLAM backend benchmark | Dedicated benchmark for distributed C-SLAM optimization with LiDAR-derived datasets, communication models, and loop outlier labels. | https://www.cosmobench.com/, https://arxiv.org/abs/2508.16731 |
| Matrix-covered in SLAM Benchmarking Metrics and Datasets | S3E | Multi-robot multimodal C-SLAM dataset | Three UGVs, indoor/outdoor, LiDAR, stereo, IMU, UWB, and RTK; fills a true collaborative sensing benchmark gap and complements COSMO-Bench's pose-graph/backend focus. | https://pengyu-team.github.io/S3E/, https://arxiv.org/abs/2210.13723 |
P1 Discovered Families
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
maplab-rovioli.md | maplab and ROVIOLI | Multi-session VI mapping toolkit | Foundational practical toolkit for VI map building, localization, map maintenance, and multi-session workflows. | https://arxiv.org/abs/1711.10250, https://github.com/ethz-asl/maplab |
| Promoted: SLAM Toolbox | ROS 2 SLAM Toolbox | Practical 2D pose-graph SLAM | Added a page for warehouse, AGV, and Nav2 deployment context, including the LGPL license correction. | https://github.com/SteveMacenski/slam_toolbox, https://docs.ros.org/en/ros2_packages/jazzy/api/slam_toolbox/ |
voxgraph-submap-sdf-slam.md | Voxgraph | Submap and map-centric volumetric SLAM | TSDF submaps plus pose-graph optimization; complements Cartographer's submap pattern. | https://arxiv.org/abs/2004.07194, https://github.com/ethz-asl/voxgraph |
quatro-quatro-plus-plus.md | Quatro / Quatro++ | Robust global registration | Complements TEASER++, FGR, Super4PCS, and Go-ICP for loop closing and global localization. | https://arxiv.org/abs/2201.13072, https://arxiv.org/abs/2311.18622, https://github.com/url-kaist/Quatro |
max-clique-point-cloud-registration.md | SC2-PCR, MAC, maximum-clique registration | Geometric loop verification / global registration | Maximum-clique consistency is a major robust PCR family for high-outlier loop verification. | https://openaccess.thecvf.com/content/CVPR2022/html/Chen_SC2-PCR_A_Second_Order_Spatial_Compatibility_for_Efficient_and_Robust_Point_CVPR_2022_paper.html, https://openaccess.thecvf.com/content/CVPR2023/html/Zhang_Robust_Point_Cloud_Registration_With_Maximal_Cliques_CVPR_2023_paper.html |
gore-qgore-registration.md | GORE / QGORE | Guaranteed outlier removal / registration | Foundational certifiable outlier-pruning line behind robust point-cloud registration. | https://arxiv.org/abs/1702.08905, https://arxiv.org/abs/2301.00697 |
s-graphs-hierarchical-semantic-slam.md | S-Graphs / S-Graphs+ / S-Graphs 2.0 | Hierarchical semantic graph SLAM | Optimizes walls, rooms, floors, objects, and robot poses as a hierarchy rather than only frame poses. | https://arxiv.org/abs/2203.15279, https://arxiv.org/abs/2507.13995, https://github.com/snt-arg/s_graphs |
ma-lio.md | MA-LIO | Asynchronous multi-LiDAR LIO | Handles asynchronous multi-LiDAR input and inter-LiDAR uncertainty; important when sensors are not identical or perfectly synchronized. | https://arxiv.org/abs/2305.16792, https://github.com/minwoo0611/MA-LIO |
multi-modal-multi-lidar-loam.md | MMLOAM / multi-modal multi-LiDAR-inertial odometry | Multi-LiDAR LIO | Tightly coupled heterogeneous multi-LiDAR-inertial system for indoor, warehouse, and hangar mapping. | https://arxiv.org/abs/2303.02684, https://github.com/TIERS/multi-modal-loam |
clic-coco-lic.md | CLIC and Coco-LIC | Continuous-time LIO/LIC | Continuous-time fixed-lag smoothing and non-uniform B-splines for asynchronous LiDAR, IMU, and camera fusion. | https://arxiv.org/abs/2302.07456, https://github.com/APRIL-ZJU/clic, https://arxiv.org/abs/2309.09808 |
wildcat-continuous-time-lio.md | Wildcat | Continuous-time LIO / field SLAM | CSIRO/SubT online continuous-time LiDAR-inertial SLAM with pose-graph support for degraded environments. | https://arxiv.org/abs/2205.12595 |
loner.md | LONER | Real-time neural implicit LiDAR-only SLAM | Bridge between classical LiDAR SLAM and neural implicit scene representations. | https://arxiv.org/abs/2309.04937, https://umautobots.github.io/loner |
nerf-loam.md | NeRF-LOAM | Neural implicit LiDAR odometry/mapping | Early large-scale incremental LiDAR NeRF/SDF LOAM method; mentioned in Splat-LOAM but not audit-promoted. | https://arxiv.org/abs/2303.10709, https://openaccess.thecvf.com/content/ICCV2023/papers/Deng_NeRF-LOAM_Neural_Implicit_Representation_for_Large-Scale_Incremental_LiDAR_Odometry_and_ICCV_2023_paper.pdf, https://github.com/JunyuanDeng/NeRF-LOAM |
shine-mapping.md | SHINE-Mapping | Neural implicit LiDAR mapping / compression | Sparse hierarchical implicit map representation; important for map compression taxonomy even if not full SLAM. | https://arxiv.org/abs/2210.02299, https://github.com/PRBonn/SHINE_mapping |
slamesh.md | SLAMesh | Mesh-based LiDAR SLAM/localization | CPU-only real-time LiDAR SLAM against an online mesh map; dense map/storage alternative to points and voxels. | https://arxiv.org/abs/2303.05252, https://github.com/lab-sun/SLAMesh |
kitware-lidar-slam.md | Kitware LiDAR SLAM / LidarView ecosystem | Open-source LiDAR SLAM framework | Practical ROS/ROS 2/LidarView stack missing from the open-source comparison beyond brief mentions. | https://www.kitware.com/kitware-lidar-slam-is-available-with-ros-and-ros2/, https://www.kitware.com/lidar-slam-spotlight-on-kitwares-open-source-library/ |
range-only-slam.md | Range-only SLAM with radio/acoustic/UWB beacons | Infrastructure-aided SLAM | Foundational family behind UWB, BLE, acoustic, and cooperative ranging SLAM. | https://www.ri.cmu.edu/publications/range-only-slam-for-robots-operating-cooperatively-with-sensor-networks/ |
radio-wifi-rfid-slam.md | WiFi-SLAM, WiFi GraphSLAM, radio fingerprint SLAM, RFID-SLAM | RF SLAM / weak localization factors | Weather-independent infrastructure signals already exist in warehouses, terminals, factories, ports, and tunnels. | https://www.ijcai.org/Proceedings/07/Papers/399.pdf, https://dblp.org/rec/conf/icra/HuangMQSTA11, https://repository.sutd.edu.sg/esploro/outputs/journalArticle/Exploiting-Radio-Fingerprints-for-Simultaneous-Localization/9911273209846, https://dblp.org/rec/journals/tii/WuGTTGY23 |
magnetic-slam-localization.md | Magnetic field maps / MagSLAM / magnetic anomaly localization | Magnetic SLAM / map localization | Passive, lighting/weather-independent fallback for indoor, underground, parking, warehouse, and infrastructure-rich environments. | https://elib.dlr.de/83461/, https://arxiv.org/abs/1804.01926 |
osm-map-prior-localization.md | OpenStreetSLAM and OSM semantic-prior localization | OSM / semantic map prior | Broader map-prior family beyond OPAL LiDAR-to-OSM: topology, building footprints, semantics, weak GNSS, and particle/factor graph localization. | https://vision.rwth-aachen.de/media/papers/florosicra13.pdf, https://www.ipb.uni-bonn.de/wp-content/papercite-data/pdf/yan2019ecmr.pdf, https://www.frontiersin.org/journals/robotics-and-ai/articles/10.3389/frobt.2023.1064934/full |
satellite-crossview-localization.md | Satellite-image-based cross-view localization | Satellite / map-prior localization | Route bootstrap and global prior when HD maps are absent but satellite imagery is available. | https://arxiv.org/abs/2207.13506, https://github.com/Toytiny/SIBCL |
panoptic-dynamic-visual-slam.md | Panoptic-SLAM, VAR-SLAM | Dynamic-scene visual SLAM | Concrete ORB-SLAM3-era dynamic systems with panoptic/unknown-object handling. | https://arxiv.org/abs/2405.02177, https://github.com/iit-DLSLab/Panoptic-SLAM, https://arxiv.org/abs/2510.16205, https://virgolinosoares.github.io/VAR-SLAM/ |
object-level-dynamic-visual-slam.md | VI Multi-Instance Dynamic SLAM, OA-SLAM, MCOO-SLAM | Object-level / semantic visual SLAM | Adds newer dynamic object-level relocalization and surround-camera object SLAM beyond DSP-SLAM. | https://arxiv.org/abs/2208.04274, https://arxiv.org/abs/2209.08338, https://arxiv.org/abs/2506.15402 |
semantic-visual-slam-avp.md | AVM-SLAM, Semantic VI-SLAM for AVP, VIPS-Odom | AV / parking semantic SLAM | Parking slots, BEV surround cameras, wheel speed, and semantic landmarks are relevant to mapped garages and low-speed AV operation. | https://arxiv.org/abs/2309.08180, https://github.com/yale-cv/avm-slam, https://arxiv.org/abs/2407.05017, https://openaccess.thecvf.com/content/ACCV2024/papers/Oh_Semantic_Visual-inertial_SLAM_for_Automated_Valet_Parking_ACCV_2024_paper.pdf |
visual-teach-repeat.md | Visual/VI teach-and-repeat, FLAF, topo-metric VT&R | Route-repeat navigation | Repeated-route navigation is operationally different from globally consistent SLAM and matters for airports, warehouses, mines, and ports. | https://www.sciencedirect.com/science/article/pii/S0921889020304176, https://arxiv.org/abs/2409.03457, https://arxiv.org/abs/2510.09089 |
event-thermal-vio-benchmarks.md | UZH event dataset, TUM-VIE, CART, MA-VIED | Event/thermal benchmark support | Needed so event and thermal pages are evaluated on appropriate data rather than EuRoC/KITTI only. | https://www.ifi.uzh.ch/en/rpg/software_datasets/davis_datasets.html, https://cvg.cit.tum.de/data/datasets/visual-inertial-event-dataset, https://arxiv.org/abs/2403.08997, https://data.caltech.edu/records/cks6g-ps927, https://github.com/isarlab-department-engineering/MA-VIED |
| Covered in Multi-Agent Neural and Gaussian SLAM | CP-SLAM, MAGiC-SLAM, MAC-Ego3D, GRAND-SLAM, MNE-SLAM, Coko-SLAM, CoMA-SLAM, MAGS-SLAM | Multi-agent neural/Gaussian SLAM | Collaborative dense neural/Gaussian SLAM is useful research contrast for multi-session dense-map construction, but remains less production-ready than classical LiDAR-capable C-SLAM for safety-critical pose. | https://arxiv.org/abs/2311.08013, https://arxiv.org/abs/2411.16785, https://arxiv.org/abs/2412.09723, https://arxiv.org/abs/2506.18885, https://openaccess.thecvf.com/content/CVPR2025/html/Deng_MNE-SLAM_Multi-Agent_Neural_SLAM_for_Mobile_Robots_CVPR_2025_paper.html, https://arxiv.org/abs/2604.00804, https://ojs.aaai.org/index.php/AAAI/article/view/37283, https://arxiv.org/abs/2605.10760 |
hydra-multi.md / multi-s-graphs.md | Hydra-Multi and Multi S-Graphs | Multi-robot scene graphs / semantic-relational C-SLAM | Collaborative scene graph and semantic relational maps for low-bandwidth loop closure and high-level map reasoning. | https://arxiv.org/abs/2304.13487, https://arxiv.org/abs/2401.05152, https://github.com/snt-arg/multi_s_graphs_docker |
dynemo-slam.md / obvi-slam.md | DYNEMO-SLAM and ObVi-SLAM | Dynamic and long-term object SLAM | Models dynamic entities or persistent object maps rather than only masking moving objects. | https://arxiv.org/abs/2503.02050, https://arxiv.org/abs/2309.15268, https://github.com/ut-amrl/ObVi-SLAM |
cu-multi-benchmark.md / mars-av-multiagent-benchmark.md / cse-cslam-benchmark.md / subt-mrs-benchmark.md | CU-Multi, Open MARS, CSE C-SLAM, SubT-MRS | Collaborative and degraded-environment benchmarks | Adds multi-robot, multi-traversal, service-environment, and subterranean benchmark coverage beyond single-robot SLAM datasets. | https://arxiv.org/abs/2509.19463, https://ai4ce.github.io/MARS/, https://arxiv.org/abs/2406.09383, https://arxiv.org/abs/2411.14775, https://arxiv.org/abs/2307.07607 |
Third-Sweep Watchlist
| Method or technique | Why watch | Current concern | Sources |
|---|---|---|---|
| ALIVE-LIO, D2-LIO, LODESTAR | New degeneracy-aware LIO variants covering learned inertial velocity fallback, directional degeneracy, and adaptive Schmidt-Kalman filtering. | Fresh 2025-2026 preprints; code and independent validation need checking. | https://arxiv.org/abs/2604.02706, https://arxiv.org/abs/2508.14355, https://arxiv.org/abs/2511.09142 |
| BUFFER-X / BUFFER-X-Lite and TurboReg | New robust point-cloud registration directions for cold start and loop verification. | Promising but too new for core method pages until SLAM integration evidence matures. | https://arxiv.org/abs/2601.02759, https://github.com/MIT-SPARK/BUFFER-X, https://openaccess.thecvf.com/content/ICCV2025/html/Wang_TurboReg_TurboClique_for_Robust_Fast_and_Accurate_Point_Cloud_Registration_ICCV_2025_paper.html |
| DeepPointMap / DeepPointMap2 | Neural descriptors for compact LiDAR maps, odometry, loop closure, and multi-agent SLAM. | Good direction but venue/code maturity unclear from primary sources checked. | https://arxiv.org/abs/2312.02684 |
| TinyDEVO, Edged USLAM, event teach-and-repeat | Event-camera SLAM is moving toward embedded and route-repeat use. | Fresh 2025-2026 preprints; promote only after code and hardware claims are verified. | https://arxiv.org/abs/2604.08060, https://arxiv.org/abs/2603.08150, https://arxiv.org/abs/2509.17287 |
| All-UWB SLAM, 5G/mmWave/RIS radio SLAM | Infrastructure-localization literature could become practical for indoor/outdoor industrial autonomy. | Many papers rely on idealized channels or fixed infrastructure; keep outside core AV SLAM for now. | https://arxiv.org/abs/2507.15474, https://arxiv.org/abs/2312.13741 |
| SKiD-SLAM, MCN-SLAM, DVM-SLAM, CoViS-Net | New distributed/collaborative SLAM and spatial-prior work. | Preprint/workshop/source maturity varies; compare against COVINS, D2SLAM, Swarm-SLAM, and Kimera-Multi before promotion. | https://sparolab.github.io/research/skid_slam/, https://arxiv.org/abs/2506.18678, https://arxiv.org/abs/2503.04126, https://proroklab.github.io/CoViS-Net/ |
P0 Dedicated Files
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
lvi-sam.md | LVI-SAM | LiDAR-visual-inertial factor-graph SLAM | Canonical bridge from LIO-SAM and VINS-Mono into LVIO; relevant for GNSS-denied terminal edges, covered roads, hangars, and weak LiDAR/vision handoff. | https://arxiv.org/abs/2104.10831, https://github.com/TixiaoShan/LVI-SAM |
fast-livo-fast-livo2.md | FAST-LIVO and FAST-LIVO2 | Direct LiDAR-inertial-visual odometry | Natural companion to FAST-LIO2; high-value open-source LIV baseline for spinning and solid-state LiDAR with camera/IMU fusion. | https://arxiv.org/abs/2203.00893, https://arxiv.org/abs/2408.14035, https://github.com/hku-mars/FAST-LIVO, https://github.com/hku-mars/FAST-LIVO2 |
r2live-r3live.md | R2LIVE and R3LIVE | LiDAR-inertial-visual fusion and RGB map reconstruction | Real-time tightly coupled LIV line from HKU-MARS; useful for colorized survey maps, digital twins, and robustness when either LiDAR geometry or visual texture is weak. | https://arxiv.org/abs/2102.12400, https://arxiv.org/abs/2109.07982, https://github.com/hku-mars/r2live, https://github.com/hku-mars/r3live |
ground-fusion-m2dgr-m3dgr.md | M2DGR, Ground-Fusion, Ground-Fusion++ / M3DGR | Ground-robot multi-sensor SLAM and benchmark lineage | Strong fit for airport GSE: RGB-D/camera, IMU, wheel, GNSS, LiDAR, indoor/outdoor transitions, wheel slip, GNSS denial, and LiDAR degeneracy. | https://arxiv.org/abs/2112.13659, https://github.com/SJTU-ViSYS/M2DGR, https://arxiv.org/abs/2402.14308, https://arxiv.org/abs/2507.08364 |
| Promoted: GVINS and GLIO Raw GNSS Factor Fusion | GVINS and GLIO | GNSS-visual-inertial and GNSS-LiDAR-IMU factor graphs | Core AV localization gap: raw GNSS pseudorange/Doppler factors, urban-canyon handling, GNSS reacquisition, and globally anchored local odometry. | https://arxiv.org/abs/2103.07899, https://github.com/HKUST-Aerial-Robotics/GVINS, https://ieeexplore.ieee.org/document/10285475, https://github.com/XikunLiu-huskit/GLIO |
| Promoted: Wheel Odometry and Vehicle Motion Factors | Wheel odometry, nonholonomic constraints, vehicle motion factors | Vehicle localization factors | Needed for low-speed airport vehicles, tug/dolly slip detection, 4WS/skid-steer constraints, and dead reckoning during GNSS/LiDAR degradation. | https://mars.cs.umn.edu/papers/KejianWu_VINSonWheels.pdf, https://woosiklee.com/downloads/papers/Lee2020IROS.pdf, https://arxiv.org/abs/2404.02515 |
| Promoted: 4D Imaging Radar RIO/SLAM | iRIOM, Go-RIO, x-RIO, multi-radar IO | 4D imaging radar-inertial SLAM | Direct all-weather localization relevance; Doppler/elevation improve observability over 2D spinning radar and sparse automotive radar. | https://arxiv.org/abs/2303.13962, https://github.com/wooseongY/Go-RIO, https://christopherdoer.github.io/publications/2022_02_JGN2022, https://www.cs.cmu.edu/~kaess/pub/Huang24icra.pdf |
robust-global-registration.md | TEASER++, Fast Global Registration, Super4PCS/4PCS, Go-ICP | Coarse/global point-cloud registration | Required for cold start, loop verification, map merging, and kidnapped-robot recovery when ICP/NDT initialization is weak. | https://github.com/MIT-SPARK/TEASER-plusplus, https://arxiv.org/abs/2001.07715, https://github.com/isl-org/FastGlobalRegistration, https://nmellado.github.io/Super4PCS/, https://jlyang.org/go-icp/ |
| Promoted: Scan Context Family | Scan Context, Scan Context++, ISC, FreSCo-style variants | LiDAR place recognition | Deterministic LiDAR loop/relocalization baseline; easier to validate for airside/warehouse use than learned descriptors. | https://github.com/SignalImageCV/scancontext, https://gisbi-kim.github.io/publications/gkim-2018-iros.pdf, https://arxiv.org/abs/2109.13494 |
robust-loop-closure-backends.md | Switchable constraints, DCS, max-mixtures, robust kernels, loop quarantine | Robust graph backend | False loop closure is one of the highest-severity SLAM failures; this should explain backend insertion, rejection, rollback, and robust factors. | https://doi.org/10.1109/IROS.2012.6385590, https://www.tu-chemnitz.de/etit/proaut/ICRAWorkshopFactorGraphs/ICRA_Workshop_on_Robust_and_Multimodal_Inference_in_Factor_Graphs/Program_files/4 - DCS.pdf, https://gtsam.org/docs/ |
| Promoted: LiDAR Bundle-Adjustment Factors | BALM, BALM 2.0, BA-CLM, LiDAR BA cost factors, gtsam_points factors | LiDAR mapping backend | Visual BA is already covered, but LiDAR BA optimizes poses against geometric edge/plane/voxel structure and matters for HD-map quality. | https://arxiv.org/abs/2010.08215, https://github.com/hku-mars/BALM, https://pmc.ncbi.nlm.nih.gov/articles/PMC11398242/, https://github.com/koide3/gtsam_points |
bad-slam.md | BAD SLAM | RGB-D dense SLAM | Strong missing dense RGB-D baseline: direct bundle adjustment over dense RGB-D maps with calibration/sync sensitivity and ETH3D relevance. | https://openaccess.thecvf.com/content_CVPR_2019/html/Schops_BAD_SLAM_Bundle_Adjusted_Direct_RGB-D_SLAM_CVPR_2019_paper.html, https://github.com/ETH3D/badslam |
okvis-rovio-basalt.md | OKVIS, ROVIO, BASALT, VI-DSO line | Classical VIO | Fills the gap between OpenVINS, VINS, Kimera, and ORB-SLAM3; key estimator tradeoffs across EKF, direct filtering, and nonlinear optimization. | https://github.com/ethz-asl/okvis, https://www.research-collection.ethz.ch/handle/20.500.11850/236658, https://www.research-collection.ethz.ch/handle/20.500.11850/187364, https://cvg.cit.tum.de/research/vslam/basalt, https://arxiv.org/abs/1904.06504 |
colmap-sfm-mvs.md | COLMAP / SfM + MVS | Offline visual mapping backend | Not online SLAM, but industry-standard for offline camera poses, 3D reconstruction, 3DGS initialization, dataset building, and map QA. | https://colmap.org/, https://www.cv-foundation.org/openaccess/content_cvpr_2016/app/S18-10.pdf |
slam3r-vggt-foundation-slam.md | SLAM3R, VGGT, CUT3R, AIM-SLAM, InfiniteVGGT | Foundation-model dense SLAM/reconstruction | MASt3R-SLAM is covered, but feed-forward pointmap/geometry-transformer SLAM is now a distinct 2024-2026 lineage. | https://arxiv.org/abs/2412.09401, https://github.com/PKU-VCL-3DV/SLAM3R, https://arxiv.org/abs/2503.11651, https://cut3r.github.io/, https://arxiv.org/abs/2603.05097, https://arxiv.org/abs/2601.02281 |
| Promoted: KISS-SLAM | KISS-SLAM | Full LiDAR-only SLAM | KISS-ICP covers odometry; KISS-SLAM adds loop closure and full 3D LiDAR SLAM with a simple baseline philosophy. | https://arxiv.org/abs/2503.12660, https://github.com/PRBonn/kiss-slam |
P1 Dedicated Files
| Suggested file | Method or technique | Category | Why it matters | Primary sources |
|---|---|---|---|---|
faster-lio-ivox.md | Faster-LIO and iVox | Fast tightly coupled LIO | Complements FAST-LIO2 with parallel sparse incremental voxels for throughput and resource-constrained robots. | https://github.com/gaoxiang12/faster-lio, https://raw.githubusercontent.com/gaoxiang12/faster-lio/main/doc/faster-lio.pdf |
direct-lidar-odometry-dlio-dliom.md | DLO, DLIO, DLIOM | Direct dense LiDAR odometry/LIO/SLAM | NASA/JPL field-robotics lineage with minimally preprocessed dense clouds, continuous-time correction, and degraded-environment robustness. | https://arxiv.org/abs/2110.00605, https://github.com/vectr-ucla/direct_lidar_odometry, https://arxiv.org/abs/2203.03749, https://github.com/vectr-ucla/direct_lidar_inertial_odometry, https://arxiv.org/abs/2305.01843 |
voxelmap-lio.md | VoxelMap, VoxelMap++, CT-VoxelMap | Probabilistic/adaptive voxel map representation | Important for scan-to-map registration, memory, solid-state LiDAR support, and modern LIO/LVIO map management. | https://arxiv.org/abs/2109.07082, https://github.com/hku-mars/VoxelMap, https://arxiv.org/abs/2308.02799, https://arxiv.org/abs/2604.03747 |
slict-clins-continuous-time-lio.md | SLICT and CLINS | Continuous-time LiDAR-inertial odometry/mapping | Important for aggressive motion, scan distortion, asynchronous sensors, multi-LiDAR input, and high-quality mapping on rough terrain. | https://arxiv.org/abs/2211.03900, https://arxiv.org/abs/2109.04687, https://github.com/APRIL-ZJU |
locus-lamp-nebula.md | LOCUS, LOCUS 2.0, LAMP | Field LiDAR odometry and multi-robot SLAM | Relevant for tunnels, mines, SubT-style operation, uncertainty-aware fusion, health-aware mapping, and multi-robot pose graphs. | https://arxiv.org/abs/2012.14447, https://robotics.jpl.nasa.gov/media/documents/2205.11784.pdf, https://arxiv.org/abs/2003.01744, https://software.nasa.gov/software/NPO-51451-1 |
mola.md | MOLA / MOLA-LO/LIO | Modular ROS 2 LiDAR odometry, mapping, localization | Production localization relevance: ROS 2-ready, modular ICP/SLAM pipelines, map manipulation, and localization-only modes. | https://github.com/MOLAorg/mola, https://arxiv.org/abs/2407.20465, https://docs.mola-slam.org/latest/mola_lidar_odometry.html, https://docs.mola-slam.org/latest/localization.html |
mulls.md | MULLS | LiDAR-only SLAM / multi-metric ICP | Useful for AV and survey mapping because it targets indoor/outdoor complex scenes and multiple LiDAR specifications. | https://arxiv.org/abs/2102.03771, https://yujie-he.github.io/publication/2021_mulls_icra/, https://github.com/YuePanEdward/MULLS |
| Promoted: RKO-LIO | RKO-LIO | Robust LiDAR-inertial odometry | PRBonn 2025/RA-L 2026 line aimed at sensor-agnostic LIO and single-configuration multi-platform use. | https://arxiv.org/abs/2509.06593, https://github.com/PRBonn/rko_lio, https://docs.ros.org/en/jazzy/p/rko_lio/pages/quickstart.html |
lic-fusion-lvio-fusion.md | LIC-Fusion, LIC-Fusion 2.0, LVIO-Fusion | LiDAR-inertial-camera odometry | Earlier tightly coupled LIC/LVIO line with online calibration and plane tracking; complements HKU-MARS direct methods. | https://arxiv.org/abs/1909.04102, https://arxiv.org/abs/2008.07196, https://researchportal.hkust.edu.hk/en/publications/lvio-fusiontightly-coupled-lidar-visual-inertial-odometry-and-map/ |
vilens-and-multimodal-landmark-tracking.md | VILENS and unified multi-modal landmark tracking | Multi-modal factor graph odometry | Legged-focused but useful for weak/degenerate modality fusion in degraded apron, tunnel, hangar, and underground-like transitions. | https://arxiv.org/abs/2107.07243, https://robots.ox.ac.uk/~mfallon/publications/2022TRO_wisth.pdf, https://arxiv.org/abs/2011.06838 |
cfear-radarodometry.md | CFEAR Radarodometry | Radar odometry | Strong learning-free spinning-radar baseline for adverse weather and indoor/outdoor generalization. | https://arxiv.org/abs/2105.01457, https://arxiv.org/abs/2211.02445, https://github.com/dan11003/CFEAR_Radarodometry |
under-the-radar-hero-radar-odometry.md | Under the Radar and HERO | Learned radar keypoints / radar odometry | Important learned radar baselines on Oxford/Boreas; HERO combines learned radar features with classical probabilistic estimation. | https://arxiv.org/abs/2001.10789, https://arxiv.org/abs/2105.14152, https://github.com/utiasASRL/hero_radar_odometry |
steam-icp-continuous-time-radar-lidar-odometry.md | STEAM-ICP, STEAM-RO/RIO/LIO | Continuous-time radar/LiDAR/inertial odometry | Important for spinning radar motion distortion, asynchronous sensors, and smooth trajectory factors. | https://github.com/utiasASRL/steam_icp, https://www.roboticsproceedings.org/rss17/p029.pdf |
| Promoted: Doppler Radar-LiDAR SLAM | Doppler-SLAM, Radarize, DRO | Doppler radar and radar-LiDAR SLAM | Bridge between radar-only, radar-inertial, and radar-LiDAR fusion pages. | https://arxiv.org/abs/2504.11634, https://github.com/Wayne-DWA/Doppler-SLAM, https://arxiv.org/abs/2311.11260, https://radarize.github.io/, https://arxiv.org/abs/2504.20339, https://github.com/utiasASRL/dro |
| Promoted: Radar-to-LiDAR Map Localization | Radar-on-LiDAR, RaLL, UnLoc, RLPR | Cross-modal all-weather localization | AV relevance: localize radar/camera/LiDAR observations against existing LiDAR maps when LiDAR/camera degrade. | https://arxiv.org/abs/2005.04644, https://arxiv.org/abs/2009.07061, https://arxiv.org/abs/2307.00741, https://arxiv.org/abs/2603.07920 |
learned-lidar-place-recognition.md | PointNetVLAD, MinkLoc3D, LoGG3D-Net, LCDNet, OverlapNet, BEVPlace | Learned loop/place recognition | Existing survey is rich but broad; this page should focus on training data, domain shift, descriptor indexing, and retrieval-vs-pose estimation. | https://github.com/mikacuy/pointnetvlad, https://github.com/jac99/MinkLoc3D, https://github.com/csiro-robotics/LoGG3D-Net, https://lcdnet.cs.uni-freiburg.de/, https://github.com/PRBonn/OverlapNet, https://arxiv.org/abs/2302.14325 |
| Promoted: NDT Variants and NDT Maps | NDT-D2D, NDT-OM, NDT-MCL, multi-resolution NDT maps | Registration/map representation | Split NDT map representations and deployment variants from the core NDT scan-matching page. | https://ieeexplore.ieee.org/document/1249285/, https://journals.sagepub.com/doi/10.1177/0278364913499415 |
continuous-time-factor-graphs-steam.md | STEAM, GP motion priors, exactly sparse continuous-time trajectories | Continuous-time backend | Explains GP priors, asynchronous sensors, rolling-shutter LiDAR/cameras, and continuous-time smoothing beyond registration. | https://github.com/utiasASRL/steam, https://arxiv.org/abs/1412.0630, https://journals.sagepub.com/doi/10.1177/0278364915585860 |
surfel-mapping-and-registration.md | Surfels across ElasticFusion, SuMa, semantic surfels | Map representation/registration | Surfels are spread across method pages; representation-level treatment should cover primitives, render-based association, semantics, and dynamic filtering. | https://journals.sagepub.com/doi/abs/10.1177/0278364916669237, https://github.com/jbehley/SuMa, https://arxiv.org/abs/2105.11320 |
incremental-lidar-map-data-structures.md | ikd-tree, iVox, voxel hashes, incremental voxel maps | Map data structures | Determines whether real-time LIO and scan-to-map can run on embedded hardware. | https://arxiv.org/abs/2102.10808, https://arxiv.org/abs/2107.06829, https://github.com/gaoxiang12/faster-lio |
openvslam-ov2slam.md | OpenVSLAM and OV2SLAM | Open-source visual SLAM frameworks | Practical feature-based frameworks beyond ORB-SLAM: camera-model flexibility, online BoW, API concerns, ROS integration. | https://arxiv.org/abs/1910.01122, https://github.com/lp-research/openvslam, https://arxiv.org/abs/2102.04060, https://github.com/ov2slam/ov2slam |
infinitam-voxel-hashing-kintinuous.md | InfiniTAM, voxel hashing, Kintinuous | RGB-D dense mapping lineage | Core large-scale TSDF engineering patterns for indoor mapping and robotics. | https://www.robots.ox.ac.uk/~victor/infinitam/, https://arxiv.org/abs/1708.00783, https://www.graphics.stanford.edu/~niessner/niessner2013hashing.html |
vox-fusion-go-slam.md | Vox-Fusion, Vox-Fusion++, GO-SLAM | Neural implicit dense RGB-D/monocular SLAM | Complements iMAP/NICE/Co-SLAM/ESLAM with scalable voxelized neural fields and global optimization. | https://arxiv.org/abs/2210.15858, https://github.com/zju3dv/Vox-Fusion, https://arxiv.org/abs/2403.12536, https://arxiv.org/abs/2309.02436, https://github.com/youmi-zym/GO-SLAM |
codeslam-deepfactors.md | CodeSLAM and DeepFactors | Learned compact dense monocular SLAM | Bridge from learned depth priors to iMAP/NeRF-SLAM through compact optimizable depth codes inside SLAM graphs. | https://arxiv.org/abs/1804.00874, https://openaccess.thecvf.com/content_cvpr_2018/papers/Bloesch_CodeSLAM_--_Learning_CVPR_2018_paper.pdf, https://arxiv.org/abs/2001.05049, https://github.com/jczarnowski/DeepFactors |
cuvslam-isaac-ros-visual-slam.md | NVIDIA cuVSLAM / Isaac ROS Visual SLAM | Production visual-inertial SLAM | Practical ROS 2 / Jetson stack for stereo, RGB-D, multi-camera, and IMU robots. | https://nvidia-isaac-ros.github.io/v/release-3.1/concepts/visual_slam/cuvslam/index.html, https://arxiv.org/abs/2506.04359, https://developer.nvidia.com/isaac/ros |
gaussian-lic.md | Gaussian-LIC / Gaussian-LIC2 | LiDAR-inertial-camera Gaussian SLAM | Bridges metric LIV odometry and photorealistic 3DGS maps for AV map QA, simulation, and digital twins. | https://github.com/APRIL-ZJU/Gaussian-LIC, https://arxiv.org/abs/2507.04004 |
vings-mono.md | VINGS-Mono | Visual-inertial Gaussian monocular SLAM | Large-scene monocular/VI 3DGS SLAM with kilometer-scale demos and loop closure via novel-view synthesis. | https://arxiv.org/abs/2501.08286, https://vings-mono.github.io/ |
| Promoted: Neural/Gaussian SLAM Surveys | 2024-2026 NeRF/3DGS SLAM surveys | Survey/taxonomy | Fast-moving neural/Gaussian SLAM needs a periodic taxonomy page so individual method pages stay organized. | https://arxiv.org/abs/2402.13255, https://arxiv.org/abs/2602.04251, https://arxiv.org/abs/2510.23988 |
P2 or Mention-Only Queue
| Method or technique | Recommended handling | Sources |
|---|---|---|
| MonoSLAM and PTAM | P1/P2 if the library needs historical foundations; otherwise alias from Bundle Adjustment SLAM. | https://www.robots.ox.ac.uk/~lav/Papers/davison_etal_pami2007/davison_etal_pami2007.html, https://www.robots.ox.ac.uk/~dwm/Publications/Papers/klein_murray_ismar2007/klein_murray_ismar2007.pdf |
| LOAM implementation lineage: A-LOAM, F-LOAM | P2 short file or add to LOAM; useful for implementation history. | https://github.com/HKUST-Aerial-Robotics/A-LOAM, https://arxiv.org/abs/2107.00822, https://sairlab.org/floam/ |
| LOAM-Livox and LIO-Livox | P2 if solid-state/Livox coverage matters; otherwise mention in LOAM/LIO pages. | https://arxiv.org/abs/1909.06700, https://github.com/hku-mars/loam_livox, https://github.com/Livox-SDK/LIO-Livox |
| M-LOAM / older multi-LiDAR SLAM lineage | Promote the multi-LiDAR LIO family through multi-lio.md and related third-sweep pages; keep M-LOAM as historical implementation lineage. | https://arxiv.org/abs/2010.14294, https://github.com/gogojjh/M-LOAM |
| LINS and LIO-mapping | P2 historical transition from LOAM-style odometry to tightly coupled LiDAR-IMU filtering. | https://arxiv.org/abs/1907.02233, https://arxiv.org/abs/1904.06993, https://github.com/hyye/lio-mapping |
| NeuralRecon, Atlas, SimpleRecon | P2 because these usually consume poses rather than solve SLAM, but they matter for learned dense mapping. | https://arxiv.org/abs/2104.00681, https://arxiv.org/abs/2003.10432, https://github.com/magicleap/Atlas, https://arxiv.org/abs/2208.14743 |
| GROUNDED and learned GPR representations | Promote Localizing GPR and Ground Encoding through the third-sweep GPR pages; keep GROUNDED as a follow-on learned representation/watch item. | https://tisl.cs.toronto.edu/publication/202005-ral-lgpr/ral20-lgpr.pdf, https://journals.sagepub.com/doi/10.1177/02783649231183460 |
| Optimization solver comparison: Ceres, g2o, GTSAM | P2, because GTSAM already exists; add only if readers need solver selection. | https://ceres-solver.org/, https://github.com/RainerKuemmerle/g2o, https://gtsam.org/docs/ |
| Older submap-graph design patterns | Promote map-centric SLAM through Voxgraph, LT-mapper, Khronos, and related third-sweep pages; keep Cartographer 3D as the canonical classical implementation. | https://google-cartographer.readthedocs.io/ |
| CPD, colored ICP, SegMatch, M2DP, ISC, OverlapNet | Mention under registration/place-recognition pages unless the repo expands those sublibraries. | https://arxiv.org/abs/0905.2635, https://www.open3d.org/docs/0.9.0/tutorial/Advanced/colored_pointcloud_registration.html, https://github.com/PRBonn/OverlapNet |
| PINGS, MegaSaM, VGGT/SwiftVGGT/Reloc-VGGT, QLIO, Dy3DGS-SLAM, Super4DR | Track as 2024-2026 emerging methods; most should live in a survey page until code/adoption stabilizes. RMGS-SLAM, VIGS-SLAM, and VGGT-SLAM++ were promoted into dedicated coverage. | https://www.roboticsproceedings.org/rss21/p040.pdf, https://mega-sam.github.io/, https://arxiv.org/abs/2503.11651, https://arxiv.org/abs/2512.02293, https://arxiv.org/abs/2604.06830 |
Benchmark and Dataset Gaps
| Dataset or benchmark | Why it should be added to SLAM Benchmarking Metrics and Datasets | Sources |
|---|---|---|
| Promoted: LaMAria / city-scale egocentric VI SLAM | Current hard visual-inertial benchmark with low light, moving platforms, long routes, and time-varying calibration. | https://lamaria.ethz.ch/, https://openaccess.thecvf.com/content/ICCV2025/papers/Krishnan_Benchmarking_Egocentric_Visual-Inertial_SLAM_at_City_Scale_ICCV_2025_paper.pdf |
| M3DGR / Ground-Fusion++ | Ground-robot sensor-fusion benchmark for visual degradation, LiDAR degeneracy, wheel slip, and GNSS denial, with broad baseline evaluation. | https://arxiv.org/abs/2507.08364, https://github.com/SJTU-ViSYS/Ground-Fusion |
| Promoted: SNAIL Radar Benchmark | 4D radar SLAM/localization benchmark with stereo, LiDAR, IMU, GNSS/INS, and difficult rain/night/tunnel conditions. | https://snail-radar.github.io/, https://arxiv.org/abs/2407.11705 |
| Promoted: HeRCULES Radar Benchmark | Heterogeneous radar benchmark combining 4D radar, spinning radar, FMCW LiDAR, IMU, GPS, and cameras for multi-session radar SLAM and place recognition. | https://sites.google.com/view/herculesdataset, https://arxiv.org/abs/2502.01946 |
| Promoted: GEODE Degenerate LiDAR Benchmark | Large degenerate-scene LiDAR benchmark for stress-testing LIO and multi-LiDAR pipelines in weak geometry. | https://thisparticle.github.io/geode/, https://arxiv.org/abs/2409.04961 |
| Promoted: ScaleMaster | Learned monocular SLAM benchmark for scale consistency and map quality in large indoor and multi-floor environments. | https://arxiv.org/abs/2602.18174 |
| Promoted: COSMO-Bench | Distributed C-SLAM optimization benchmark with communication models, LiDAR-derived inputs, and loop outlier labels. | https://www.cosmobench.com/, https://arxiv.org/abs/2508.16731 |
| Matrix-covered: S3E | Multi-robot multimodal C-SLAM dataset with indoor/outdoor UGV runs, LiDAR, stereo, IMU, UWB, and RTK; keep as a dedicated-page candidate only if raw-sensor C-SLAM evaluation becomes a focused comparison. | https://pengyu-team.github.io/S3E/, https://arxiv.org/abs/2210.13723 |
| Event and thermal VIO benchmarks | Evaluation support for event-camera and thermal-inertial SLAM where EuRoC/KITTI-style RGB assumptions break down. | https://www.ifi.uzh.ch/en/rpg/software_datasets/davis_datasets.html, https://cvg.cit.tum.de/data/datasets/visual-inertial-event-dataset, https://arxiv.org/abs/2403.08997, https://data.caltech.edu/records/cks6g-ps927 |
| Collaborative and degraded-environment SLAM benchmarks | CU-Multi, Open MARS, CSE C-SLAM, and SubT-MRS cover multi-agent, service-environment, and subterranean evaluation gaps. | https://arxiv.org/abs/2509.19463, https://ai4ce.github.io/MARS/, https://arxiv.org/abs/2406.09383, https://arxiv.org/abs/2411.14775, https://arxiv.org/abs/2307.07607 |
| Matrix-covered: IILABS 3D | Indoor 3D LiDAR SLAM benchmark with wheeled robot, IMU, wheel odometry, MoCap ground truth, and multiple LiDAR types. | https://jorgedfr.github.io/3d_lidar_slam_benchmark_at_iilab/ |
| Matrix-covered: SMapper-light | Public indoor/outdoor multimodal sequences from an open-hardware SLAM platform; useful for reproducible LiDAR, visual, VI, and multimodal SLAM checks, with offline LiDAR-SLAM-derived ground truth. | https://snt-arg.github.io/smapper_docs/datasets/smapper-light/, https://arxiv.org/abs/2509.09509 |
| HeLiPR | Heterogeneous LiDAR and long-term place recognition, important for sensors and route revisits. | https://sites.google.com/view/heliprdataset, https://journals.sagepub.com/doi/10.1177/02783649241242136 |
| Matrix-covered: FusionPortableV2 | Generalized multi-platform SLAM evaluation: handheld, legged robot, UGV, vehicle, 27 sequences, 38.7 km; use map-quality scores carefully because some dynamic objects remain in ground-truth maps. | https://fusionportable.github.io/dataset/fusionportable_v2/, https://arxiv.org/abs/2404.08563 |
| Matrix-covered: Oxford Spires | LiDAR, cameras, IMU, TLS reference maps, and large landmark-scale reconstruction for radiance-field/SLAM/localization work; trajectory ground truth applies to core sequences. | https://dynamic.robots.ox.ac.uk/datasets/oxford-spires/, https://arxiv.org/abs/2411.10546 |
| Promoted: Hilti x Trimble 360 Visual-Inertial SLAM Challenge 2026 | Adds 360 visual-inertial data and floor-plan priors beyond Hilti 2023. | https://github.com/Hilti-Research/hilti-trimble-slam-challenge-2026 |
| ETH3D SLAM benchmark | Visual-inertial mono/stereo/RGB-D evaluation; useful for dense visual/neural SLAM sanity checks. | https://eth3d.ethz.ch/slam_benchmark |
| VBR: Vision Benchmark in Rome | Urban outdoor visual odometry/SLAM with RGB, point clouds, IMU, and GPS. | https://arxiv.org/abs/2404.11322 |
| M2DGR, NTU VIRAL, UrbanLoco | Older but important multimodal ground/aerial SLAM benchmarks that should be explicit rows. | https://github.com/SJTU-ViSYS/M2DGR, https://arxiv.org/abs/2112.13659, https://ntu-aris.github.io/ntu_viral_dataset/ |
Already Covered, But Needs Better Discoverability
| Existing file | Add aliases or cross-links for |
|---|---|
| FAST-LIO and FAST-LIO2 | Faster-LIO, iVox, VoxelMap, Point-LIO, Livox-specific methods, and a clear note that FAST-LIVO is a different LIV family. |
| KISS-ICP | KISS-SLAM as the full SLAM stack rather than odometry. |
| Continuous-Time Registration and CT-ICP | SLICT, CLINS, DLIO continuous-time motion correction, CT-VoxelMap, and STEAM/GP priors. |
| Bundle Adjustment SLAM | BALM/BALM2 as LiDAR-specific BA, plus COLMAP, MonoSLAM, and PTAM aliases. |
| Loop Closure and Place Recognition | Scan Context, learned LiDAR descriptors, switchable constraints, DCS, loop quarantine. |
| HDL Graph SLAM | GLIM and gtsam_points as the modern Koide ecosystem continuation. |
| LOAM and LeGO-LOAM | A-LOAM, F-LOAM, LOAM-Livox, MULLS, BALM, LINS, and LIO-mapping. |
| Radar Odometry and Radar SLAM | CFEAR, HERO, Under the Radar, Navtech/Oxford Radar RobotCar, Boreas, MulRan, RADIATE, K-Radar. |
| Radar-Inertial Odometry | iRIOM, Go-RIO, x-RIO, multi-radar IO, EKF-RIO-TC, DeRO. |
| Radar-LiDAR-Inertial Fusion | Doppler-SLAM, Radarize, DRO, GaRLIO, DR-LRIO, radar-to-LiDAR map localization. |
| OpenVINS | MSCKF family alias and explicit contrast to OKVIS/ROVIO/BASALT. |
| MASt3R-SLAM | SLAM3R, VGGT, CUT3R, MegaSaM, MonST3R, and feed-forward geometry-model SLAM. |
| KinectFusion and ElasticFusion | InfiniTAM, voxel hashing, Kintinuous, BAD SLAM, surfel maps. |
| iMAP, NICE-SLAM, Co-SLAM and ESLAM, NeRF-SLAM | CodeSLAM, DeepFactors, Vox-Fusion, GO-SLAM, and neural dense SLAM survey. |
| Open-Source SLAM Stack Comparison | MOLA, GLIM, KISS-SLAM, LOCUS/LAMP, FAST-LIVO2/R3LIVE, DLIO/DLIOM, cuVSLAM/Isaac ROS. |
| Splat-LOAM | Already covers LiDAR Gaussian-splatting odometry and mapping; cross-link it with GS-LIVM, Gaussian-LIC, RadarSplat-RIO, S3PO-GS, and the neural/Gaussian survey page. |
| WildGS-SLAM | Already covers dynamic monocular Gaussian SLAM; add aliases from the latest sweep so WildGS is discoverable from Gaussian, dynamic-scene, and visual SLAM paths. |
| LiDAR SLAM Algorithms | GenZ-ICP is already mentioned there; promote to a dedicated file if degeneracy-robust registration becomes a first-class subsection. |
Guardrail Process
When adding or revising SLAM research:
- Check whether the method is already a dedicated file, a mention-only item, or a backlog item in this audit.
- If it is P0, create a dedicated file before expanding lower-priority coverage.
- If it is P1, either create a dedicated file or add a clear cross-link from the closest existing method page.
- If it is P2 or mention-only, add aliases in INDEX, overview text, or the relevant method file so search catches the method name.
- After every SLAM expansion, update this audit, SLAM Method Library Overview, Open-Source SLAM Stack Comparison, README, and Research Index.