Skip to content

Glossary

Acronyms and Terms Used in This Research Corpus


Aviation / Airport

TermDefinition
A-CDMAirport Collaborative Decision Making — shared turnaround milestones
A-SMGCSAdvanced Surface Movement Guidance and Control Systems — surface radar + routing
ADS-BAutomatic Dependent Surveillance-Broadcast — aircraft transponder positions
AIXMAeronautical Information Exchange Model — airport geometry standard
AMDB/AMXMAirport Mapping Database / Exchange Model — detailed surface features
AODBAirport Operational Database — flight schedules, gate assignments
APUAuxiliary Power Unit — aircraft onboard generator
ARPAerodrome Reference Point — airport coordinate origin
ATCAir Traffic Control
CAASCivil Aviation Authority of Singapore
CBRSCitizens Broadband Radio Service — 3.5 GHz shared spectrum (US)
CDMCollaborative Decision Making
CFDComputational Fluid Dynamics — for jet blast modeling
CTOTCalculated Take-Off Time
DPIDeparture Planning Information — A-CDM message type
EASAEuropean Union Aviation Safety Agency
EIBTEstimated In-Block Time
FAAFederal Aviation Administration (US)
FODForeign Object Debris — objects on runway/taxiway/apron
GSEGround Support Equipment — vehicles servicing aircraft
ICAOInternational Civil Aviation Organization
IGOMIATA Ground Operations Manual
MLATMultilateration — triangulating position from multiple receivers
NOTAMNotice to Air Missions — temporary airspace/surface restrictions
ODDOperational Design Domain — defined conditions for autonomous operation
TOBTTarget Off-Block Time — planned pushback time
TSATTarget Start-up Approval Time — engine start clearance
ULDUnit Load Device — standardized cargo/baggage container
URLLCUltra-Reliable Low-Latency Communication — 5G service class

Autonomous Vehicles

TermDefinition
ADASAdvanced Driver Assistance Systems — L1-L2 automation
AGVSAutomated Guided Vehicle System — FAA's term for autonomous GSE
AVAutonomous Vehicle
BEVBird's-Eye-View — top-down representation for driving
CANController Area Network — vehicle communication bus
DBWDrive-by-Wire — electronic vehicle control
E2EEnd-to-End — single model from sensors to controls
FSDFull Self-Driving (Tesla)
HD MapHigh-Definition Map — centimeter-accurate road map
L4SAE Level 4 — high driving automation (no human fallback in ODD)
LiDARLight Detection and Ranging — laser-based 3D scanning
MPCModel Predictive Control — optimization-based control
OTAOver-the-Air — remote software/model updates
PIDProportional-Integral-Derivative — basic control algorithm
RSSResponsibility-Sensitive Safety — Mobileye's formal safety model
RTKReal-Time Kinematic — cm-level GPS correction
SLAMSimultaneous Localization and Mapping
SOTIFSafety of the Intended Functionality — ISO 21448
TRLTechnology Readiness Level — 1 (concept) to 9 (proven)
V2XVehicle-to-Everything — communication standard
VRUVulnerable Road User — pedestrians, cyclists

AI / ML

TermDefinition
3DGS3D Gaussian Splatting — neural scene representation
CFGClassifier-Free Guidance — controlling generation quality
CoCChain-of-Causation — Alpamayo's reasoning format
DDIMDenoising Diffusion Implicit Models — fast diffusion sampling
DDPMDenoising Diffusion Probabilistic Models
DiTDiffusion Transformer — transformer-based diffusion model
DLADeep Learning Accelerator — NVIDIA dedicated inference hardware
DRL/MARLDeep/Multi-Agent Reinforcement Learning
EMAExponential Moving Average — codebook update method
FIDFréchet Inception Distance — image quality metric
FSQFinite Scalar Quantization — codebook-free tokenization (used by Cosmos)
FVDFréchet Video Distance — video quality metric
GTSAMGeorgia Tech Smoothing and Mapping — factor graph library
IoUIntersection over Union — overlap metric
ISAM2Incremental Smoothing and Mapping — fast factor graph optimization
JEPAJoint Embedding Predictive Architecture — LeCun's world model paradigm
KV-cacheKey-Value cache — transformer inference optimization
LoRALow-Rank Adaptation — efficient fine-tuning method
mAPMean Average Precision — detection accuracy metric
MBRLModel-Based Reinforcement Learning
MoEMixture of Experts — conditional computation
NDSnuScenes Detection Score — composite metric
NeRFNeural Radiance Fields — neural scene representation
ONNXOpen Neural Network Exchange — model interchange format
OODOut-of-Distribution — input outside training distribution
PACProbably Approximately Correct — learning theory bound
POMDPPartially Observable Markov Decision Process
PTQ/QATPost-Training Quantization / Quantization-Aware Training
RoPERotary Position Embedding
RSSMRecurrent State-Space Model — Dreamer's dynamics model
SSMState Space Model — O(n) sequence model (Mamba)
TOPSTera Operations Per Second — compute performance
TRTTensorRT — NVIDIA inference optimization
VGICPVoxelized Generalized Iterative Closest Point — scan matching
VLAVision-Language-Action model — unified perception+reasoning+control
VLMVision-Language Model — multimodal AI model
VQ-VAEVector Quantized Variational Autoencoder — discrete tokenization

Optimization and Numerical Linear Algebra

TermDefinition
ObjectiveFunction minimized by a solver; in autonomy it combines residuals, priors, weights, and constraints. See Objective and Residual Design Audit.
ResidualDifference between a predicted measurement and an observed measurement, expressed in the correct frame and units. See Nonlinear Least Squares.
Whitened residualResidual premultiplied by square-root information so its components are in normalized noise units. See Objective and Residual Design Audit.
JacobianDerivative of a residual with respect to a local state perturbation. See Jacobians, Autodiff, and Manifold Linearization.
Manifold updateTangent-space update retracted back to a constrained state such as SO(3), SE(3), or a unit quaternion. See Jacobians, Autodiff, and Manifold Linearization.
Rank deficiencyCondition where a Jacobian or Hessian has unobservable or redundant directions. See Eigenvalues, Hessian Conditioning, and Observability.
Schur complementBlock elimination algebra used to solve reduced systems or form marginalization priors, with different interpretation in each use. See Sparse Estimation Backend Crosswalk.
Marginalization priorPrior produced by eliminating old variables from a fixed-lag or reduced estimator while preserving their linearized information on remaining variables. See Schur Complement, Marginalization, and PCG.
Covariance recoveryProcess of extracting selected uncertainty blocks from a solved information or square-root system. See Square-Root Information and Covariance Recovery.
PCGPreconditioned conjugate gradients, an iterative method for large symmetric positive definite systems. See Sparse Estimation Backend Crosswalk.
LinearizationLocal first-order approximation of residuals around the current estimate. See Nonlinear Least Squares.
Local coordinatesTangent-space coordinates used to perturb manifold states during linearization. See Jacobians, Autodiff, and Manifold Linearization.
Normal equationsLinear system J^T J delta = -J^T r formed from a least-squares linearization; fast but can square conditioning. See Cholesky, LDLT, and Normal Equations.
DampingNumerical regularization that changes a nonlinear step to improve local stability; it is not a physical prior. See Solver Selection and Convergence Diagnosis.
Trust-region ratioActual reduction divided by predicted reduction, used to accept or reject trial steps and update the trust region. See Trust Region and Line Search Globalization.
Line-search step lengthScalar step multiplier selected to reduce the objective along a chosen direction. See Trust Region and Line Search Globalization.
Convergence criterionStopping rule based on cost change, gradient norm, step norm, solver status, or iteration budget. See Solver Selection and Convergence Diagnosis.
NullspaceState direction that does not change the linearized residual. See Eigenvalues, Hessian Conditioning, and Observability.
Gauge freedomModel symmetry such as global pose or scale that measurements cannot determine without a chosen gauge or prior. See Sparse Estimation Backend Crosswalk.
Condition numberRatio describing how sensitive a linear solve is to perturbations. See Sparse Estimation Backend Crosswalk.
SparsityMatrix structure where most entries are zero because factors touch only a few variables. See Sparse Matrices, Fill-In, and Ordering.
Fill-inNew nonzero entries created during sparse elimination. See Sparse Matrices, Fill-In, and Ordering.
OrderingVariable elimination order that changes fill-in, runtime, memory, and sometimes diagnostic visibility. See Sparse Estimation Backend Crosswalk.
CholeskyFactorization for symmetric positive definite systems, often used on normal equations. See Cholesky, LDLT, and Normal Equations.
LDLTSymmetric factorization that can expose indefinite or semidefinite behavior more directly than plain Cholesky. See Cholesky, LDLT, and Normal Equations.
QRLeast-squares factorization that works directly on J and avoids explicitly forming J^T J. See QR, SVD, and Rank-Revealing Solvers.
SVDSingular value decomposition used to expose rank, weak modes, and nullspace directions. See QR, SVD, and Rank-Revealing Solvers.
Square-root informationFactor whose transpose times itself is the information matrix, commonly used for stable residual whitening and priors. See Square-Root Information and Covariance Recovery.
Marginal covarianceUncertainty block for selected variables after accounting for eliminated or unqueried variables. See Square-Root Information and Covariance Recovery.
PreconditionerApproximate inverse or scaling that improves PCG convergence. See Schur Complement, Marginalization, and PCG.

Safety / Certification

TermDefinition
AMLASAssurance of Machine Learning for Autonomous Systems — safety methodology
ASILAutomotive Safety Integrity Level — ISO 26262 (A-D)
CEConformité Européenne — EU product certification mark
GSNGoal Structuring Notation — safety case diagram notation
PLPerformance Level — ISO 13849 safety rating (a-e)
SILSafety Integrity Level — IEC 61508
SimplexDual-controller architecture — high-performance + verified fallback
UL 4600Standard for evaluation of autonomous products

Companies

AbbreviationFull Name
IAGInternational Airlines Group (British Airways parent)
SATSSingapore Airport Terminal Services
TLDTracteurs et Lourds de Distribution (GSE manufacturer)

160+ terms defined. Updated as corpus grows.

Public research notes collected from public sources.