GLIM and GTSAM Pipeline Hub

This is the cross-section hub for understanding GLIM as a SLAM pipeline and GTSAM as the mathematical backend behind that pipeline. It links the method pages to the knowledge-base pages that explain the probability, geometry, optimization, and sparse linear algebra layers.

Use this page when a question spans more than one file, for example: "how does a GLIM scan factor become a GTSAM solve?", "where do Bayes trees and Hessians enter the pipeline?", or "which KB page explains the failure I am seeing?"

Core Spine

sensor packets
  -> time sync, calibration, preprocessing, deskew
  -> range/IMU/GNSS/loop/custom residual factors
  -> GTSAM nonlinear factor graph over poses, velocities, biases, and submaps
  -> manifold linearization and whitening
  -> sparse Jacobian/Hessian or Bayes-tree solve
  -> marginals, diagnostics, trajectories, submaps, and map artifacts

The key point: GLIM is the SLAM/mapping framework; GTSAM is the graph optimization and inference machinery; the KB pages explain the math that makes the machinery inspectable.

Pipeline Crosswalk

GLIM pipeline stage	GTSAM object or operation	Mathematical topic	Diagnostic artifact
Sensor ingestion and calibration	measurements, timestamps, frame transforms	Lie groups, SE(3), SO(3), and Jacobians, Sensor Calibration and Time Synchronization	frame trace, TF chain, time-offset replay, lever-arm check
Deskew and inertial propagation	PreintegratedImuMeasurements, ImuFactor, NavState, bias variables	IMU Error Models and Preintegration	IMU residuals, bias trajectory, gravity alignment, deskew sharpness
Factor construction	NonlinearFactorGraph, Values, NoiseModelFactorN, custom factors	GTSAM Factor Graph Optimization, Objective and Residual Design Audit	factor list, residual units, connected keys, zero-residual synthetic case
Probability model	priors, likelihood factors, robust noise models	Probabilistic Graphical Models and Message Passing, Likelihood, MAP, MLE, and Least Squares	posterior factorization, prior policy, factor independence assumptions
Noise and whitening	noiseModel::Diagonal, Gaussian models, robust wrappers	Gaussian Noise, Covariance, Information, Whitening, and Uncertainty Ellipses, Robust Losses and M-Estimators	whitened residual histograms, per-factor chi-square, robust weights
Scan matching and submap matching	gtsam_points scan/VGICP factors, loop factors, plane factors	GICP and VGICP, LiDAR Bundle-Adjustment Factors, Point Cloud Registration Math	inlier count, overlap, voxel covariance, scan Hessian, weak eigenvectors
Nonlinear step	Gauss-Newton, Levenberg-Marquardt, Dogleg, iSAM2 update policy	Gauss-Newton, Levenberg-Marquardt, and Dogleg, Nonlinear Solver Diagnostics Crosswalk	cost trace, damping/radius, predicted-vs-actual reduction, accepted/rejected steps
Linearization and Hessian	GaussianFactorGraph, JacobianFactor, HessianFactor, H = J^T J	Jacobians, Autodiff, and Manifold Linearization, Eigenvalues, Hessian Conditioning, and Observability	finite-difference checks, Hessian spectrum, nullspace, local observability
Sparse backend	ordered elimination, Cholesky, QR, Bayes net, Bayes tree	Sparse Matrices, Fill-In, and Ordering, Cholesky, LDLT, and Normal Equations, QR, SVD, and Rank-Revealing Solvers	fill report, clique size, pivot warnings, QR/SVD rank snapshot
Fixed-lag and map refinement	fixed-lag smoother, marginal factors, global submap graph	Schur Complement, Marginalization, and PCG, Square-Root Information and Covariance Recovery	dense prior rank, separator variables, selected marginal covariance
Pipeline-level SLAM method	GLIM odometry, global mapping, offline correction, multi-session merge	GLIM, Factor Graph SLAM with iSAM2 and GTSAM	odom_*.txt, traj_*.txt, submap graph, exported PLY/map artifacts

Failure Routing

Symptom	First route
Low scalar cost but wrong map	Objective and Residual Design Audit, then scan/loop residual pages
Cholesky or indeterminate linear-system failure	Cholesky, LDLT, and Normal Equations, then Eigenvalues, Hessian Conditioning, and Observability
iSAM2 update-time spike	Sparse Matrices, Fill-In, and Ordering, then Factor Graph SLAM with iSAM2 and GTSAM
Covariance looks too confident	Gaussian Noise, Covariance, Information, Whitening, and Uncertainty Ellipses, then SLAM/VIO Observability, FEJ, Nullspace, and Consistency
Loop closure bends a map	Robust Losses and M-Estimators, Loop Closure and Place Recognition, and Nonlinear Solver Diagnostics Crosswalk
Open-area drift or repeated-structure ambiguity	Eigenvalues, Hessian Conditioning, and Observability, GICP and VGICP, and GLIM
Custom factor behaves unexpectedly	Jacobians, Autodiff, and Manifold Linearization, Lie Groups SE(3), SO(3), Adjoints, and Jacobians, and GTSAM Factor Graph Optimization

Reading Paths

For the full GLIM pipeline, read GLIM, then GTSAM Factor Graph Optimization, then Factor Graph SLAM with iSAM2 and GTSAM.

For the math behind the solve, read Likelihood, MAP, MLE, and Least Squares, Nonlinear Least Squares from First Principles, Jacobians, Autodiff, and Manifold Linearization, and Gauss-Newton, Levenberg-Marquardt, and Dogleg.

For sparse backend behavior, read Sparse Estimation Backend Crosswalk, then Sparse Matrices, Fill-In, and Ordering, Cholesky, LDLT, and Normal Equations, QR, SVD, and Rank-Revealing Solvers, and Schur Complement, Marginalization, and PCG.

For production-style debugging, start with Nonlinear Solver Diagnostics Crosswalk, then route to objective design, noise whitening, Jacobians, rank/conditioning, sparse backend, or state-estimation observability.

Boundary Notes

• GLIM owns the concrete range-inertial mapping workflow: odometry, submaps, global mapping, offline correction, multi-session merge, extension modules, and exported artifacts.
• GTSAM owns the graph abstraction and inference machinery: factor graphs, values, noise models, nonlinear optimization, elimination, Bayes tree/iSAM2, marginals, and fixed-lag smoothing.
• The KB pages own reusable theory. They should explain the math so that a GLIM/GTSAM issue can be debugged without treating either codebase as a black box.