Glossary
Acronyms and Terms Used in This Research Corpus
Aviation / Airport
| Term | Definition |
|---|---|
| A-CDM | Airport Collaborative Decision Making — shared turnaround milestones |
| A-SMGCS | Advanced Surface Movement Guidance and Control Systems — surface radar + routing |
| ADS-B | Automatic Dependent Surveillance-Broadcast — aircraft transponder positions |
| AIXM | Aeronautical Information Exchange Model — airport geometry standard |
| AMDB/AMXM | Airport Mapping Database / Exchange Model — detailed surface features |
| AODB | Airport Operational Database — flight schedules, gate assignments |
| APU | Auxiliary Power Unit — aircraft onboard generator |
| ARP | Aerodrome Reference Point — airport coordinate origin |
| ATC | Air Traffic Control |
| CAAS | Civil Aviation Authority of Singapore |
| CBRS | Citizens Broadband Radio Service — 3.5 GHz shared spectrum (US) |
| CDM | Collaborative Decision Making |
| CFD | Computational Fluid Dynamics — for jet blast modeling |
| CTOT | Calculated Take-Off Time |
| DPI | Departure Planning Information — A-CDM message type |
| EASA | European Union Aviation Safety Agency |
| EIBT | Estimated In-Block Time |
| FAA | Federal Aviation Administration (US) |
| FOD | Foreign Object Debris — objects on runway/taxiway/apron |
| GSE | Ground Support Equipment — vehicles servicing aircraft |
| ICAO | International Civil Aviation Organization |
| IGOM | IATA Ground Operations Manual |
| MLAT | Multilateration — triangulating position from multiple receivers |
| NOTAM | Notice to Air Missions — temporary airspace/surface restrictions |
| ODD | Operational Design Domain — defined conditions for autonomous operation |
| TOBT | Target Off-Block Time — planned pushback time |
| TSAT | Target Start-up Approval Time — engine start clearance |
| ULD | Unit Load Device — standardized cargo/baggage container |
| URLLC | Ultra-Reliable Low-Latency Communication — 5G service class |
Autonomous Vehicles
| Term | Definition |
|---|---|
| ADAS | Advanced Driver Assistance Systems — L1-L2 automation |
| AGVS | Automated Guided Vehicle System — FAA's term for autonomous GSE |
| AV | Autonomous Vehicle |
| BEV | Bird's-Eye-View — top-down representation for driving |
| CAN | Controller Area Network — vehicle communication bus |
| DBW | Drive-by-Wire — electronic vehicle control |
| E2E | End-to-End — single model from sensors to controls |
| FSD | Full Self-Driving (Tesla) |
| HD Map | High-Definition Map — centimeter-accurate road map |
| L4 | SAE Level 4 — high driving automation (no human fallback in ODD) |
| LiDAR | Light Detection and Ranging — laser-based 3D scanning |
| MPC | Model Predictive Control — optimization-based control |
| OTA | Over-the-Air — remote software/model updates |
| PID | Proportional-Integral-Derivative — basic control algorithm |
| RSS | Responsibility-Sensitive Safety — Mobileye's formal safety model |
| RTK | Real-Time Kinematic — cm-level GPS correction |
| SLAM | Simultaneous Localization and Mapping |
| SOTIF | Safety of the Intended Functionality — ISO 21448 |
| TRL | Technology Readiness Level — 1 (concept) to 9 (proven) |
| V2X | Vehicle-to-Everything — communication standard |
| VRU | Vulnerable Road User — pedestrians, cyclists |
AI / ML
| Term | Definition |
|---|---|
| 3DGS | 3D Gaussian Splatting — neural scene representation |
| CFG | Classifier-Free Guidance — controlling generation quality |
| CoC | Chain-of-Causation — Alpamayo's reasoning format |
| DDIM | Denoising Diffusion Implicit Models — fast diffusion sampling |
| DDPM | Denoising Diffusion Probabilistic Models |
| DiT | Diffusion Transformer — transformer-based diffusion model |
| DLA | Deep Learning Accelerator — NVIDIA dedicated inference hardware |
| DRL/MARL | Deep/Multi-Agent Reinforcement Learning |
| EMA | Exponential Moving Average — codebook update method |
| FID | Fréchet Inception Distance — image quality metric |
| FSQ | Finite Scalar Quantization — codebook-free tokenization (used by Cosmos) |
| FVD | Fréchet Video Distance — video quality metric |
| GTSAM | Georgia Tech Smoothing and Mapping — factor graph library |
| IoU | Intersection over Union — overlap metric |
| ISAM2 | Incremental Smoothing and Mapping — fast factor graph optimization |
| JEPA | Joint Embedding Predictive Architecture — LeCun's world model paradigm |
| KV-cache | Key-Value cache — transformer inference optimization |
| LoRA | Low-Rank Adaptation — efficient fine-tuning method |
| mAP | Mean Average Precision — detection accuracy metric |
| MBRL | Model-Based Reinforcement Learning |
| MoE | Mixture of Experts — conditional computation |
| NDS | nuScenes Detection Score — composite metric |
| NeRF | Neural Radiance Fields — neural scene representation |
| ONNX | Open Neural Network Exchange — model interchange format |
| OOD | Out-of-Distribution — input outside training distribution |
| PAC | Probably Approximately Correct — learning theory bound |
| POMDP | Partially Observable Markov Decision Process |
| PTQ/QAT | Post-Training Quantization / Quantization-Aware Training |
| RoPE | Rotary Position Embedding |
| RSSM | Recurrent State-Space Model — Dreamer's dynamics model |
| SSM | State Space Model — O(n) sequence model (Mamba) |
| TOPS | Tera Operations Per Second — compute performance |
| TRT | TensorRT — NVIDIA inference optimization |
| VGICP | Voxelized Generalized Iterative Closest Point — scan matching |
| VLA | Vision-Language-Action model — unified perception+reasoning+control |
| VLM | Vision-Language Model — multimodal AI model |
| VQ-VAE | Vector Quantized Variational Autoencoder — discrete tokenization |
Optimization and Numerical Linear Algebra
| Term | Definition |
|---|---|
| Objective | Function minimized by a solver; in autonomy it combines residuals, priors, weights, and constraints. See Objective and Residual Design Audit. |
| Residual | Difference between a predicted measurement and an observed measurement, expressed in the correct frame and units. See Nonlinear Least Squares. |
| Whitened residual | Residual premultiplied by square-root information so its components are in normalized noise units. See Objective and Residual Design Audit. |
| Jacobian | Derivative of a residual with respect to a local state perturbation. See Jacobians, Autodiff, and Manifold Linearization. |
| Manifold update | Tangent-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 deficiency | Condition where a Jacobian or Hessian has unobservable or redundant directions. See Eigenvalues, Hessian Conditioning, and Observability. |
| Schur complement | Block elimination algebra used to solve reduced systems or form marginalization priors, with different interpretation in each use. See Sparse Estimation Backend Crosswalk. |
| Marginalization prior | Prior 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 recovery | Process of extracting selected uncertainty blocks from a solved information or square-root system. See Square-Root Information and Covariance Recovery. |
| PCG | Preconditioned conjugate gradients, an iterative method for large symmetric positive definite systems. See Sparse Estimation Backend Crosswalk. |
| Linearization | Local first-order approximation of residuals around the current estimate. See Nonlinear Least Squares. |
| Local coordinates | Tangent-space coordinates used to perturb manifold states during linearization. See Jacobians, Autodiff, and Manifold Linearization. |
| Normal equations | Linear 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. |
| Damping | Numerical regularization that changes a nonlinear step to improve local stability; it is not a physical prior. See Solver Selection and Convergence Diagnosis. |
| Trust-region ratio | Actual 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 length | Scalar step multiplier selected to reduce the objective along a chosen direction. See Trust Region and Line Search Globalization. |
| Convergence criterion | Stopping rule based on cost change, gradient norm, step norm, solver status, or iteration budget. See Solver Selection and Convergence Diagnosis. |
| Nullspace | State direction that does not change the linearized residual. See Eigenvalues, Hessian Conditioning, and Observability. |
| Gauge freedom | Model symmetry such as global pose or scale that measurements cannot determine without a chosen gauge or prior. See Sparse Estimation Backend Crosswalk. |
| Condition number | Ratio describing how sensitive a linear solve is to perturbations. See Sparse Estimation Backend Crosswalk. |
| Sparsity | Matrix structure where most entries are zero because factors touch only a few variables. See Sparse Matrices, Fill-In, and Ordering. |
| Fill-in | New nonzero entries created during sparse elimination. See Sparse Matrices, Fill-In, and Ordering. |
| Ordering | Variable elimination order that changes fill-in, runtime, memory, and sometimes diagnostic visibility. See Sparse Estimation Backend Crosswalk. |
| Cholesky | Factorization for symmetric positive definite systems, often used on normal equations. See Cholesky, LDLT, and Normal Equations. |
| LDLT | Symmetric factorization that can expose indefinite or semidefinite behavior more directly than plain Cholesky. See Cholesky, LDLT, and Normal Equations. |
| QR | Least-squares factorization that works directly on J and avoids explicitly forming J^T J. See QR, SVD, and Rank-Revealing Solvers. |
| SVD | Singular value decomposition used to expose rank, weak modes, and nullspace directions. See QR, SVD, and Rank-Revealing Solvers. |
| Square-root information | Factor whose transpose times itself is the information matrix, commonly used for stable residual whitening and priors. See Square-Root Information and Covariance Recovery. |
| Marginal covariance | Uncertainty block for selected variables after accounting for eliminated or unqueried variables. See Square-Root Information and Covariance Recovery. |
| Preconditioner | Approximate inverse or scaling that improves PCG convergence. See Schur Complement, Marginalization, and PCG. |
Safety / Certification
| Term | Definition |
|---|---|
| AMLAS | Assurance of Machine Learning for Autonomous Systems — safety methodology |
| ASIL | Automotive Safety Integrity Level — ISO 26262 (A-D) |
| CE | Conformité Européenne — EU product certification mark |
| GSN | Goal Structuring Notation — safety case diagram notation |
| PL | Performance Level — ISO 13849 safety rating (a-e) |
| SIL | Safety Integrity Level — IEC 61508 |
| Simplex | Dual-controller architecture — high-performance + verified fallback |
| UL 4600 | Standard for evaluation of autonomous products |
Companies
| Abbreviation | Full Name |
|---|---|
| IAG | International Airlines Group (British Airways parent) |
| SATS | Singapore Airport Terminal Services |
| TLD | Tracteurs et Lourds de Distribution (GSE manufacturer) |
160+ terms defined. Updated as corpus grows.