The Physical AI Glossary

Software and Connectivity

5G: Fifth Generation Mobile Network

5G is a generation of cellular networking designed for higher capacity, lower delay, and support for many connected devices. Private 5G networks can connect robots and industrial equipment across facilities where cables or WiFi are impractical.

Sensors and Perception

Absolute Encoder

An absolute encoder reports a unique position value, including after a restart. A robot joint can therefore know its position without first moving to a reference point, depending on the encoder design.

Sensors and Perception

Accelerometer

An accelerometer measures changes in motion along one or more directions. It can detect vibration, impacts, tilt related to gravity, and movement, making it useful for navigation and equipment monitoring.

Industry and Safety

Acceptance Testing

Acceptance testing checks a delivered system against agreed requirements before the customer formally accepts it. Clear tests reduce arguments about what success means.

Robotics and Actuation

Accuracy

Accuracy measures how close a robot reaches the commanded position. It is different from repeatability, which measures how consistently it reaches the same position.

Sensors and Perception

Acoustic Sensing

Acoustic sensing uses sound to detect events, condition, distance, or location. Changes in a machine's sound can reveal wear before a visible failure appears.

Models and Learning

Action Token

An action token is a model output that represents part of a physical command. A sequence of tokens might describe gripper state, joint movement, direction, or when a task should stop.

Models and Learning

Active Learning

Active learning lets a model identify the examples from which it would learn the most. Human labeling effort can then focus on uncertain, rare, or especially valuable cases.

Models and Learning

Actor Critic

Actor critic methods use one model to choose actions and another to judge how useful those actions are. The critic's feedback helps the actor improve its policy.

Robotics and Actuation

Actuator

An actuator converts energy into physical movement. It is the part that lets a machine turn a wheel, move a joint, close a gripper, or raise an arm. Electric motors, hydraulic cylinders, and pneumatic systems are common actuators.

Foundations

Adaptation

Adaptation is the process of changing behavior when conditions, tasks, or equipment change. It can happen through learning, parameter adjustment, planning, or a person updating the system.

Navigation and Control

Admittance Control

Admittance control measures an outside force and commands motion in response. A person can push on the robot, and the controller moves it according to the chosen behavior.

Robotics and Actuation

AGV: Automated Guided Vehicle

An AGV transports material along defined routes using markers, wires, reflectors, or mapped paths. It is often highly reliable in structured facilities but less flexible than an AMR.

Compute and Memory

AI Accelerator

An AI accelerator is hardware designed to run machine learning calculations faster or more efficiently than a general purpose processor. GPUs, NPUs, TPUs, FPGAs, and specialized ASICs can all serve as AI accelerators.

Foundations

AI Agent

An AI agent observes information, pursues a goal, chooses actions, and responds to results. In physical AI, those actions may become robot commands, work orders, route changes, or other decisions that affect the real world.

Foundations

AI: Artificial Intelligence

AI is the broad field of building computer systems that can perform tasks associated with human intelligence, including recognizing patterns, understanding language, making predictions, and choosing actions. Not every AI system learns, and not every AI system controls a machine.

Compute and Memory

ALU: Arithmetic Logic Unit

An ALU is the part of a processor that performs basic arithmetic and logical operations. Large chips contain many ALUs working together on different pieces of a calculation.

Robotics and Actuation

AMR: Autonomous Mobile Robot

An AMR uses sensors, maps, and planning to navigate with limited human control. Unlike a machine following one fixed path, it can usually reroute around changes.

Software and Connectivity

API: Application Programming Interface

An API is a defined way for software systems to request data or functions from one another. A fleet application might use an API to assign missions, read robot status, or connect with warehouse software.

Compute and Memory

ASIC: Application Specific Integrated Circuit

An ASIC is a chip designed for a narrow purpose. Because its hardware is built around a specific job, it can be faster and more efficient than a general processor, but it is expensive to design and difficult to change.

Industry and Safety

Assembly Automation

Assembly automation joins parts using machines, robots, fixtures, tools, and controls. AI can help when part position, appearance, or process conditions vary.

Foundations

Automation

Automation uses machines or software to perform work with less direct human effort. It may rely on fixed rules, AI, or both. A highly automated machine is not necessarily autonomous.

Foundations

Autonomous System

An autonomous system can carry out a defined task with limited direct human control. Its independence always has boundaries. A delivery robot may navigate sidewalks on its own but still need a remote operator for unusual situations.

Foundations

Autonomy

Autonomy is the ability of a system to make and execute decisions within a defined task and environment. It is not an all or nothing property. A machine may be autonomous during normal operation and require help during exceptions.

Robotics and Actuation

Backdrivability

Backdrivability describes how easily an outside force can move an actuator or joint backward through its transmission. A backdrivable arm can feel more natural to guide by hand, but it may need power to hold a load in place.

Robotics and Actuation

Backlash

Backlash is unwanted free movement between mechanical parts when direction changes. In a robot gearbox, backlash can reduce positioning accuracy and create vibration.

Models and Learning

Batch

A batch is a group of examples processed together during training or inference. Larger batches can improve hardware use but require more memory and may change learning behavior.

Models and Learning

BC: Behavior Cloning

Behavior cloning is a direct form of imitation learning. A model studies demonstrations and learns to predict the action a skilled operator took in each situation. It is simple and useful, but it may struggle when the robot enters unfamiliar conditions.

Robotics and Actuation

Bearing

A bearing supports a moving shaft or joint while reducing friction. Its size, load rating, precision, and sealing affect the life and accuracy of a robot.

Robotics and Actuation

BLDC Motor: Brushless Direct Current Motor

A BLDC motor uses electronic switching instead of mechanical brushes. It can provide efficient, reliable motion with good power for its size, which makes it common in drones, mobile robots, tools, and joint actuators.

Software and Connectivity

BLE: Bluetooth Low Energy

BLE is a short range wireless technology designed for low power devices. It is common in beacons, wearables, sensors, setup tools, and nearby device communication.

Industry and Safety

Brownfield Deployment

A brownfield deployment adds technology to an existing facility or process. The new system must work with installed equipment, limited space, old interfaces, and ongoing production.

Compute and Memory

Cache

A cache is a small, fast memory that keeps frequently needed data close to a processor. Good caching reduces trips to slower memory and can improve the speed and energy efficiency of AI and control workloads.

Sensors and Perception

Camera

A camera captures light as images or video. In physical AI, cameras can support inspection, navigation, object recognition, tracking, and human interaction. Their usefulness depends on lighting, placement, lenses, resolution, and processing software.

Software and Connectivity

CAN: Controller Area Network

CAN is a durable communication network originally developed for vehicles. It lets controllers and devices exchange short messages without a central computer and is widely used in cars, robots, mobile equipment, and industrial machines.

Also called CAN bus

Robotics and Actuation

Cartesian Robot

A Cartesian robot moves along straight axes, usually arranged as X, Y, and Z. It is simple to program and common in gantries, printers, and material handling.

Industry and Safety

Changeover

Changeover is the work needed to switch a process from one product, format, tool, or recipe to another. Flexible automation aims to reduce this time and effort.

Compute and Memory

Chiplet

A chiplet is a smaller piece of silicon designed to be combined with other chiplets in one package. This lets manufacturers mix compute, memory, and communication functions without building one enormous chip.

Navigation and Control

Closed Loop Control

Closed loop control measures the result of an action and adjusts the command to reduce error. A motor controller may continuously compare actual speed with requested speed and increase or decrease power.

Software and Connectivity

Cloud Computing

Cloud computing provides shared computing, storage, and software through remote data centers. Physical AI systems often use the cloud for training, fleet analysis, updates, and heavy processing that does not need an immediate local response.

Industry and Safety

Cobot: Collaborative Robot

A cobot is a robot designed for applications where people and robots may share a workspace under defined safety measures. The label does not make every task automatically safe. The complete application still requires a risk assessment.

Foundations

Cognition

Cognition refers to the processes used to understand information, form an internal view of a situation, and decide what to do. In robotics, the word often covers reasoning, memory, planning, and learning rather than raw sensing or motor control.

Simulation and Data

Collision Detection

Collision detection determines whether virtual or physical shapes touch or intersect. Planners and simulators use it to prevent or model contact.

Industry and Safety

Commissioning

Commissioning is the process of testing, adjusting, and proving that an installed system works as intended. It includes equipment, software, safety, operators, and real production conditions.

Compute and Memory

Compiler

A compiler translates code or a model into instructions a target processor can execute. AI compilers may reorganize operations so the model runs faster on a chosen accelerator.

Robotics and Actuation

Compliance

Compliance is the ability of a mechanism or controller to yield when force is applied. A compliant robot can make contact more gently, tolerate small alignment errors, and interact more safely with uncertain objects.

Industry and Safety

Condition Monitoring

Condition monitoring tracks signals such as vibration, heat, current, pressure, and sound to understand equipment health and identify meaningful changes.

Simulation and Data

Contact Model

A contact model describes what happens when simulated objects touch, including friction, impact, sliding, and deformation. Small modeling errors can strongly affect manipulation results.

Software and Connectivity

Containerization

Containerization packages an application with the software it needs so it runs more consistently across computers. Edge deployments use containers to separate services and simplify updates.

Models and Learning

Context Window

A context window is the amount of information a model can consider at one time. For a robot, that context may include instructions, recent observations, previous actions, and task history.

Models and Learning

Continual Learning

Continual learning helps a model acquire new skills or knowledge over time while trying not to forget what it previously learned.

Robotics and Actuation

Continuum Robot

A continuum robot bends smoothly along its body instead of moving through a few rigid joints. It can reach through narrow or curved spaces in medicine and inspection.

Foundations

Control

Control is the process of directing a machine so its actual behavior matches the desired behavior. A controller might adjust motor power many times per second to keep a robot balanced or hold an arm at the correct position.

Navigation and Control

Control Gain

A control gain determines how strongly a controller responds to an input or error. Poor gain settings can make a system slow, unstable, noisy, or prone to overshoot.

Navigation and Control

Control Loop Frequency

Control loop frequency is how often a controller reads feedback and updates its command. Faster loops can respond sooner but require dependable sensors, communication, and computing.

Navigation and Control

Coordinate Frame

A coordinate frame is a reference system used to describe position and direction. A robot may track separate frames for the room, its base, its camera, its arm, and the object it wants to pick up.

Navigation and Control

Costmap

A costmap assigns a travel cost to areas around a robot. Obstacles receive very high cost, while narrow, uncertain, or undesirable areas can receive smaller penalties.

Foundations

CPS: Cyber Physical System

A cyber physical system tightly connects software, networks, sensors, and physical equipment. A smart factory line is one example because digital control decisions continuously affect machines, materials, and production in the real world.

Compute and Memory

CPU: Central Processing Unit

The CPU is the general purpose processor inside a computer. It runs applications and coordinates other components. In a robot, the CPU may manage navigation software, communications, and routine decisions while specialized chips handle demanding AI calculations.

Compute and Memory

CU: Compute Unit

A compute unit is a group of processing resources inside a chip that executes work. The exact design varies by manufacturer, so counts are most useful when comparing products built on the same architecture.

Compute and Memory

CUDA: Compute Unified Device Architecture

CUDA is NVIDIA's software platform for programming its GPUs. Many AI and robotics tools use CUDA libraries to run demanding calculations on NVIDIA hardware.

Sensors and Perception

Current Sensor

A current sensor measures the electrical current flowing through a circuit. Motor current can help estimate load, detect jams, and monitor energy use.

Models and Learning

Curriculum Learning

Curriculum learning presents training tasks in a purposeful order, often moving from easier situations to harder ones. This can help a robot learn complex behavior more reliably.

Sensors and Perception

CV: Computer Vision

Computer vision is the field of extracting useful information from images and video. It lets machines identify objects, measure positions, inspect products, read text, track motion, and understand scenes.

Compute and Memory

CXL: Compute Express Link

CXL is a connection standard that lets processors, accelerators, and memory share data more efficiently over PCIe hardware. In large AI systems, it can make memory easier to pool and access across different devices.

Industry and Safety

Cybersecurity

Cybersecurity protects systems, networks, software, and data from unauthorized access or harmful change. In physical AI, a digital attack can create physical consequences.

Industry and Safety

Cycle Time

Cycle time is the time required to complete one unit, operation, or repeated sequence. A faster robot does not improve total cycle time if another process remains the bottleneck.

Robotics and Actuation

Cycloidal Drive

A cycloidal drive is a gearbox that uses an offset rolling motion to create high reduction in a compact form. It can handle shock loads well and is used in industrial robots and heavy motion systems.

Navigation and Control

Damping

Damping reduces oscillation and motion over time. Physical materials, mechanical components, and control software can all add damping to a robot.

Simulation and Data

Data Annotation

Data annotation adds useful labels to raw data. People or software might mark objects in images, identify robot actions, describe events, or flag failures so a model has clearer examples to learn from.

Simulation and Data

Data Augmentation

Data augmentation creates varied training examples by changing existing data. Images may be cropped, brightened, rotated, or blurred so a model becomes less sensitive to small differences.

Simulation and Data

Data Generation

Data generation is the process of creating or collecting examples for development and testing. Sources can include real machines, simulation, people, automated tools, and controlled experiments.

Simulation and Data

Data Governance

Data governance defines who may collect, access, change, retain, and use data. Physical AI programs need it because machines can capture sensitive operational and human information.

Simulation and Data

Data Lake

A data lake stores large amounts of raw or lightly processed data for later use. Robot fleets may send sensor recordings, events, and logs into a shared lake.

Simulation and Data

Data Pipeline

A data pipeline moves information through collection, storage, cleaning, labeling, training, evaluation, and deployment. Physical AI teams need reliable pipelines because field data is large, messy, and closely tied to model improvement.

Simulation and Data

Data Provenance

Data provenance records where data came from, how it was collected, and what changed along the way. It supports debugging, trust, licensing, and repeatable model development.

Simulation and Data

Data Quality

Data quality describes whether data is accurate, complete, consistent, timely, and useful for its purpose. More data does not help when the measurements or labels are unreliable.

Simulation and Data

Data Split

A data split divides examples into training, validation, and test groups. Poor splitting can accidentally place nearly identical examples in several groups and exaggerate performance.

Simulation and Data

Data Warehouse

A data warehouse organizes cleaned business and operational data for reporting and analysis. It is usually more structured than a data lake.

Simulation and Data

Dataset

A dataset is an organized collection of examples used to train, test, or study a system. For physical AI, it may contain images, video, sensor readings, actions, outcomes, failures, and information about the environment.

Software and Connectivity

DDS: Data Distribution Service

DDS is a communication standard for distributing data between programs and devices. ROS 2 commonly uses DDS implementations beneath its messaging layer.

Navigation and Control

Dead Reckoning

Dead reckoning estimates a new position from a known starting point and measured movement. Small errors accumulate over time, so another reference is usually needed for correction.

Models and Learning

Decoder

A decoder turns an internal model representation into an output such as text, an image, a prediction, or a sequence of robot actions.

Robotics and Actuation

Delta Robot

A delta robot uses several lightweight arms connected to one platform. Its low moving mass supports very fast picking and packaging of light products.

Industry and Safety

Depalletizing

Depalletizing removes items from a pallet. Mixed loads, damaged packaging, uncertain positions, and hidden objects make it harder than stacking a known pattern.

Sensors and Perception

Depth Camera

A depth camera estimates the distance from the camera to points in a scene. This helps a robot understand shape, size, and free space rather than seeing only a flat image.

Sensors and Perception

Depth Estimation

Depth estimation calculates how far parts of a scene are from a camera. It can use stereo cameras, active depth sensors, or AI models, and it helps robots judge shape, distance, and free space.

Navigation and Control

Deterministic Computing

Deterministic computing produces results within predictable timing limits. In control systems, dependable timing can matter more than the fastest average performance.

Software and Connectivity

Device Management

Device management tracks, configures, secures, and updates connected machines. It becomes essential when an organization operates hundreds or thousands of edge devices.

Industry and Safety

Diagnostics

Diagnostics are the tests and information used to identify the cause of a problem. Good diagnostics turn a vague robot failure into an actionable repair.

Robotics and Actuation

Differential Drive

Differential drive uses separately controlled wheels on opposite sides. Changing their relative speeds lets the robot move straight, curve, or rotate in place.

Models and Learning

Diffusion Model

A diffusion model learns to create useful outputs by gradually removing noise. Beyond image generation, diffusion methods can propose robot movements by refining a rough set of possible actions into a smoother and more useful plan.

Simulation and Data

Digital Thread

A digital thread connects information across the life of a product or asset, from design and manufacturing through operation, service, and retirement.

Simulation and Data

Digital Twin

A digital twin is a digital representation of a real asset, process, or facility that stays connected to relevant real data. Its value comes from supporting decisions, testing changes, monitoring condition, or predicting outcomes.

Robotics and Actuation

Direct Drive

Direct drive connects a motor to the load without a reduction gearbox. This can reduce backlash and friction, but the motor must provide enough torque directly and may be larger or more expensive.

Foundations

DL: Deep Learning

Deep learning is a form of machine learning that uses neural networks with many layers. It is especially useful for complex data such as images, speech, video, and sensor readings, where useful patterns are difficult to describe with fixed rules.

Compute and Memory

DMA: Direct Memory Access

DMA lets hardware move data to or from memory without making the CPU handle every step. This frees the CPU and helps cameras, networks, and accelerators transfer data efficiently.

Robotics and Actuation

DOF: Degrees of Freedom

Degrees of freedom count the independent ways a mechanism can move. A simple sliding axis has one. A six axis robot arm can position and orient its tool through six controlled movements.

Models and Learning

Domain Adaptation

Domain adaptation helps a model work when its operating data differs from its training data. A vision system trained in a bright laboratory may need adaptation before it performs reliably in a dim warehouse.

Simulation and Data

Domain Randomization

Domain randomization changes lighting, textures, object positions, friction, mass, and other simulated conditions during training. The goal is to prevent a model from depending too heavily on one perfect virtual environment.

Industry and Safety

Downtime

Downtime is time when equipment or a process is unavailable for expected production. Planned maintenance and unexpected failures should be tracked separately.

Compute and Memory

DPU: Data Processing Unit

A DPU handles data movement, networking, storage, and security work that would otherwise occupy a CPU. In an AI server, it can keep information flowing while the CPU and accelerators focus on models and applications.

Compute and Memory

DRAM: Dynamic Random Access Memory

DRAM is the common working memory used by computers and many embedded systems. It holds data while the machine is running, offers more capacity than on chip memory, and loses its contents when power is removed.

Compute and Memory

DSP: Digital Signal Processor

A DSP is a processor optimized for fast, repeated mathematical operations on signals. It is commonly used for audio, radar, motor control, communications, and sensor processing where data arrives continuously and must be handled quickly.

Sensors and Perception

Dynamic Range

Dynamic range is the span between the darkest and brightest signals a sensor can capture usefully. High dynamic range helps a camera see detail in shadows and bright areas together.

Robotics and Actuation

Dynamics

Dynamics studies how forces, mass, and motion affect one another. Robot control uses dynamics to estimate how much torque is needed to accelerate a joint, carry a payload, or resist gravity.

Industry and Safety

E Stop: Emergency Stop

An emergency stop is a manually operated control intended to stop a hazardous situation quickly. It supports other safeguards and is not a substitute for proper guarding.

Software and Connectivity

Edge AI

Edge AI means running AI near the source of the data, such as inside a robot, vehicle, camera, or factory computer. It can improve response time, privacy, reliability, and control over operating costs.

Software and Connectivity

Edge Computing

Edge computing processes data near the machine, sensor, or location where it is created. This can reduce network delays and cloud traffic while allowing important functions to continue when connectivity is limited.

Simulation and Data

Edge Data

Edge data is information created or stored near machines and sensors. Teams must decide what to process locally, what to retain, and what is valuable enough to send elsewhere.

Software and Connectivity

Edge Gateway

An edge gateway connects local machines and sensors with wider networks or cloud systems. It may translate protocols, filter data, run applications, and enforce security.

Compute and Memory

Edge Processor

An edge processor runs computing tasks near the machine or sensor that creates the data. It is usually chosen for a balance of speed, power use, size, cost, and the models it can support.

Navigation and Control

EKF: Extended Kalman Filter

An EKF adapts the Kalman filter for systems whose motion or measurements are not purely linear. Robots commonly use it to combine odometry, IMU, and positioning data.

Robotics and Actuation

Electric Actuator

An electric actuator uses electrical energy to create movement, usually through a motor and mechanical transmission. It is common in robots because it is relatively easy to control, power, and integrate with electronics.

Models and Learning

Embedding

An embedding represents information as a list of numbers that captures useful relationships. Similar objects, words, images, or actions often receive embeddings that are close together.

Foundations

Embodied AI

Embodied AI is artificial intelligence that learns or operates through a body, such as a robot, vehicle, or smart machine. Its intelligence is shaped by what the body can sense, how it can move, and the environment around it.

Foundations

Emergent Behavior

Emergent behavior is an ability or pattern that appears from interactions inside a complex system without being programmed as one explicit rule. It can be useful, surprising, or difficult to predict.

Sensors and Perception

Encoder

An encoder measures the position or movement of a rotating or linear part. Robots use encoders to know how far a wheel turned, where a joint is, or how quickly a motor is moving.

Models and Learning

Encoder

An encoder turns raw input into a more useful internal representation. A vision encoder may convert camera pixels into features describing objects, shapes, and spatial relationships.

Robotics and Actuation

EOAT: End of Arm Tooling

An end effector is the tool attached to the working end of a robot arm. It may be a gripper, welder, suction cup, camera, screwdriver, spray nozzle, or another device chosen for the task.

Also called End effector

Models and Learning

Epoch

An epoch is one complete pass through a training dataset. Models usually train for several epochs, but too many can cause them to memorize the training examples.

Navigation and Control

Error Signal

An error signal is the difference between the desired value and the measured value. Controllers use that difference to decide how strongly to correct the system.

Software and Connectivity

EtherCAT: Ethernet for Control Automation Technology

EtherCAT is an industrial Ethernet system designed for fast, predictable communication with motors, sensors, and controllers. It is common in machines that need many devices to update with tight timing.

Software and Connectivity

EtherNet IP

EtherNet IP is an industrial network that carries the Common Industrial Protocol over standard Ethernet technology. It connects controllers, machines, drives, and safety devices.

Sensors and Perception

Event Camera

An event camera reports changes in brightness at individual pixels instead of capturing full images at fixed intervals. It can react quickly to motion and use less data in scenes where most of the view is not changing.

Sensors and Perception

Exteroception

Exteroception is sensing the world outside the machine. Cameras, LiDAR, radar, microphones, and proximity sensors provide external information about objects, people, terrain, and other conditions.

Sensors and Perception

Extrinsic Parameters

Extrinsic parameters describe a sensor's position and orientation relative to another sensor or coordinate frame. They let a robot combine measurements from different viewpoints.

Industry and Safety

Fail Safe System

A fail safe system moves toward a safer condition when a fault occurs. That might mean removing motor power, applying a brake, closing a valve, or handing control to a person. The safest response depends on the machine.

Industry and Safety

FAT: Factory Acceptance Test

A FAT tests equipment at the supplier's facility before shipment. It can confirm major functions early, but it cannot reproduce every condition at the final site.

Industry and Safety

Fault Detection

Fault detection identifies when equipment, software, or a sensor is no longer behaving as expected. Early detection supports safe response and faster repair.

Industry and Safety

Fault Tolerance

Fault tolerance is the ability to continue an important function when part of a system fails. The allowed behavior and duration must be clearly defined.

Models and Learning

Federated Learning

Federated learning trains across many devices without collecting all raw data in one place. Devices share model updates, which can improve privacy but adds coordination and security challenges.

Foundations

Feedback Loop

A feedback loop measures what happened after an action and uses that result to guide the next action. A robot arm can keep checking its joint position and correcting the motor command until it reaches the target.

Navigation and Control

Feedforward Control

Feedforward control predicts the command needed before an error appears. It often works with feedback control, which then corrects anything the prediction missed.

Models and Learning

Fine Tuning

Fine tuning continues training an existing model on a smaller, more focused dataset. A general vision model can be fine tuned on images from one factory so it becomes better at recognizing that factory's parts and defects.

Software and Connectivity

Firmware

Firmware is low level software stored on a device that controls its basic hardware behavior. Motors, sensors, batteries, cameras, and controllers often have their own firmware.

Sensors and Perception

Fisheye Camera

A fisheye camera uses a very wide angle lens to capture more of the surroundings. The image appears curved, so software must account for distortion when measuring geometry.

Robotics and Actuation

FK: Forward Kinematics

Forward kinematics calculates where a robot's tool will be when its joint positions are known. It answers the question: given these joint angles, where is the end effector?

Compute and Memory

Flash Memory

Flash memory stores data even when power is off. Robots and edge devices use it for operating systems, model files, logs, and application software. It is persistent but usually slower than working memory such as DRAM.

Compute and Memory

FLOPS: Floating Point Operations per Second

FLOPS measures how many floating point calculations a processor can perform each second. It is useful for comparing certain workloads, but real performance also depends on memory, software, precision, cooling, and how well the task uses the hardware.

Software and Connectivity

FMS: Fleet Management System

A fleet management system coordinates many robots or vehicles. It assigns work, tracks status, manages traffic, monitors health, and connects the fleet with business operations.

Navigation and Control

Force Control

Force control regulates how strongly a robot pushes or pulls instead of controlling only position. It is important for polishing, assembly, testing, and delicate handling.

Models and Learning

Foundation Model

A foundation model is trained on a broad collection of data and can later be adapted to many tasks. Instead of building every robot skill from the beginning, developers can start with a general model and specialize it for a machine or workplace.

Sensors and Perception

FOV: Field of View

Field of view is the area a sensor can observe, usually expressed as an angle. A wide view sees more surroundings while a narrow view can provide more detail at distance.

Compute and Memory

FP16: Sixteen Bit Floating Point

FP16 represents decimal values with sixteen bits. It uses half the storage of standard FP32 and is widely used to accelerate neural network training and inference.

Compute and Memory

FP8: Eight Bit Floating Point

FP8 is a compact floating point format used by newer AI accelerators. It can increase speed and reduce memory use, but models need careful scaling to preserve useful accuracy.

Compute and Memory

FPGA: Field Programmable Gate Array

An FPGA is a chip whose internal logic can be reconfigured after manufacturing. It offers more hardware flexibility than an ASIC and can provide very fast, predictable processing for cameras, communications, control, and specialized AI workloads.

Sensors and Perception

FPS: Frame Rate

Frame rate measures how many images a camera captures each second. More frames can track fast motion better, but they also create more data to move and process.

Sensors and Perception

FT Sensor: Force Torque Sensor

A force torque sensor measures pushes, pulls, and twisting forces. It is often mounted near a robot wrist so the machine can control contact during assembly, polishing, testing, or delicate manipulation.

Industry and Safety

Functional Safety

Functional safety is the part of safety that depends on a control system responding correctly to dangerous conditions or failures. Examples include safely stopping motion when a guard opens or when a safety sensor detects a person.

Robotics and Actuation

Gantry Robot

A gantry robot moves a tool along rails mounted over or beside a work area. It can cover a large rectangular workspace and carry loads with strong support.

Robotics and Actuation

Gear Ratio

Gear ratio compares the rotation of a motor with the rotation of its output. Gearing can trade speed for greater torque, but it also affects efficiency, precision, noise, and how easily outside forces can move the joint.

Robotics and Actuation

Gearbox

A gearbox uses gears to change speed, torque, or direction between a motor and a load. Robot gearboxes must often provide high precision and stiffness in a small, durable package.

Foundations

Generalization

Generalization is a system's ability to perform well on situations it did not see exactly during training. A picking model generalizes when it can handle new objects without being retrained for each one.

Industry and Safety

Geofencing

Geofencing creates a virtual boundary around a real place. A drone, vehicle, or mobile robot can use that boundary to restrict where it may travel, change its behavior, or alert an operator.

Navigation and Control

Geolocation

Geolocation determines where a device or observation is located on Earth or within a known area. It can use satellites, networks, maps, visual landmarks, or several methods together.

Navigation and Control

Global Planner

A global planner chooses an overall route using a map and a distant goal. It usually works with a local planner that reacts to nearby changes.

Sensors and Perception

Global Shutter

A global shutter exposes all image pixels at the same time. It helps preserve the shape of fast moving objects and is useful for robotics, inspection, and machine vision.

Navigation and Control

GNSS: Global Navigation Satellite System

GNSS is the general name for satellite positioning systems such as GPS, Galileo, and others. Robots, drones, farm equipment, and vehicles use it outdoors, but buildings, trees, and reflections can reduce accuracy.

Compute and Memory

GPU: Graphics Processing Unit

A GPU performs many similar calculations at the same time. It was designed for graphics, but that parallel structure also suits AI. Robots use GPUs to process camera feeds, run large models, and train systems from data.

Industry and Safety

Greenfield Deployment

A greenfield deployment begins with a new facility or system. Designers have more freedom, but they must make important assumptions before real operating data exists.

Robotics and Actuation

Gripper

A gripper is an end effector that holds or moves objects. Designs may use fingers, suction, magnets, soft materials, or specialized fixtures depending on the object's shape, weight, surface, and fragility.

Simulation and Data

Ground Truth

Ground truth is the best available reference for what actually happened or what the correct label should be. It is used to train and evaluate models.

Sensors and Perception

Gyroscope

A gyroscope measures how quickly a device rotates. Combined with other sensors, it helps a drone stay level, a mobile robot track turns, or a humanoid estimate the motion of its body.

Sensors and Perception

Hall Effect Sensor

A Hall effect sensor detects magnetic fields. It can measure motor position, wheel speed, electrical current, or whether a magnet attached to a moving part is nearby.

Robotics and Actuation

Harmonic Drive

A harmonic drive is a compact precision gearbox that uses a flexible gear element to produce a large reduction ratio. It is common in robot joints because it offers low backlash, though it can be costly and mechanically delicate.

Industry and Safety

Hazard

A hazard is a source or situation with the potential to cause harm. Motion, heat, electricity, pressure, sharp tools, falling loads, and incorrect decisions can all be hazards.

Industry and Safety

Hazard Analysis

Hazard analysis systematically examines how harm could occur. It looks at normal work, faults, maintenance, foreseeable misuse, unusual environments, and interactions between components.

Compute and Memory

HBM: High Bandwidth Memory

HBM stacks memory close to a processor so large amounts of data can move quickly. This matters because a powerful AI chip can waste time waiting for information. HBM helps keep the processor supplied with model data.

Compute and Memory

Heat Sink

A heat sink draws heat away from a chip and spreads it over a larger surface. Air or liquid then carries that heat away so the processor can operate safely.

Compute and Memory

Heterogeneous Computing

Heterogeneous computing uses different kinds of processors in one system, with each handling the work it does best. A robot might use a CPU for coordination, a GPU for vision, and a microcontroller for motor control.

Simulation and Data

HIL: Hardware in the Loop

Hardware in the loop testing connects real controllers or components to a simulated machine or environment. It tests actual hardware without requiring the complete physical system.

Foundations

HITL: Human in the Loop

Human in the loop means a person actively contributes to a system's decisions, training, or operation. A worker may approve a robot's unusual action before it proceeds.

Industry and Safety

HMI: Human Machine Interface

An HMI is the screen or control panel through which a person monitors and operates a machine. A good HMI shows status, alarms, settings, and actions clearly enough for an operator to respond correctly.

Foundations

HOTL: Human on the Loop

Human on the loop means the system operates by itself while a person supervises and can intervene. The human is not approving every action but remains responsible for defined exceptions.

Foundations

HRI: Human Robot Interaction

Human robot interaction studies how people and robots communicate, share space, divide work, and build trust. Good interaction design helps a person understand what a robot is doing, what it needs, and when intervention is required.

Robotics and Actuation

Humanoid Robot

A humanoid robot has a body arranged around humanlike features such as a torso, arms, hands, or legs. The form can help it use spaces and tools designed for people.

Robotics and Actuation

Hydraulic Actuator

A hydraulic actuator uses pressurized liquid to produce movement and high force. It is common in heavy machinery, but pumps, valves, fluid lines, leaks, noise, and maintenance add complexity.

Robotics and Actuation

IK: Inverse Kinematics

Inverse kinematics calculates which joint positions can place a robot's tool at a desired pose. There may be several answers, one answer, or no possible answer because of reach and joint limits.

Sensors and Perception

Image Classification

Image classification assigns a label to an entire image or selected crop. It can answer questions such as whether a part is acceptable or defective, but it does not necessarily show where the defect is located.

Sensors and Perception

Image Resolution

Image resolution describes the number of pixels in an image. Higher resolution can reveal smaller details, but it increases storage, bandwidth, and processing demands.

Models and Learning

Imitation Learning

Imitation learning teaches a machine from demonstrations of the desired behavior. A person might guide a robot arm through a task while the system records the observations and actions needed to reproduce it.

Navigation and Control

Impedance Control

Impedance control makes a robot behave like a chosen combination of spring, mass, and damping. It controls how the machine responds when the environment pushes back.

Sensors and Perception

IMU: Inertial Measurement Unit

An IMU combines motion sensors, usually accelerometers and gyroscopes, to estimate how a device is moving and rotating. Robots, drones, vehicles, and wearable devices use IMUs when external position signals are limited or delayed.

Sensors and Perception

Incremental Encoder

An incremental encoder reports movement as a series of pulses. The controller counts those pulses to track distance and direction, but it may need a known reference position after losing power.

Foundations

Industrial Automation

Industrial automation applies control systems, machines, and software to production and industrial processes. It includes everything from fixed conveyor logic to adaptive robot cells and intelligent inspection.

Models and Learning

Inference

Inference happens when a trained model uses what it learned to produce an answer, prediction, or action. When a warehouse robot identifies a box or predicts where a person is moving, it is performing inference.

Compute and Memory

Inference Engine

An inference engine is software that loads a trained model, prepares inputs, runs the calculations, and returns outputs. It often optimizes the model for a particular processor.

Compute and Memory

InfiniBand

InfiniBand is a high speed network used in supercomputers and large AI clusters. It is designed to move data between servers with high bandwidth and low delay.

Compute and Memory

INT8: Eight Bit Integer

INT8 stores numbers as eight bit integers. Many inference accelerators process INT8 quickly and efficiently, though a model must be converted and checked to make sure accuracy remains acceptable.

Foundations

Intelligent Machine

An intelligent machine uses sensing and computation to adapt its behavior rather than following only one fixed sequence. The term is broad and does not specify how capable or autonomous the machine is.

Compute and Memory

Interconnect

An interconnect is the pathway that moves data between chips, memory, sensors, or computers. Its speed and design can limit the performance of a system even when the individual processors are very powerful.

Software and Connectivity

Interoperability

Interoperability is the ability of different products and systems to exchange information and work together. Shared standards help, but useful integration still depends on compatible meanings and behavior.

Sensors and Perception

Intrinsic Parameters

Intrinsic parameters describe a camera's internal geometry, including focal length, image center, and lens distortion. They are needed to turn pixels into useful measurements.

Compute and Memory

Jitter

Jitter is variation in timing. If a control message normally arrives every millisecond but occasionally arrives much later, that variation is jitter. Too much jitter can make robot motion less smooth or predictable.

Navigation and Control

Kalman Filter

A Kalman filter combines noisy measurements with a model to estimate a changing state. It is widely used to track position, speed, and orientation from several sensors.

Robotics and Actuation

Kinematics

Kinematics describes motion using position, speed, and geometry without focusing on the forces that create it. It connects robot joint movements to the location and orientation of the tool.

Models and Learning

Knowledge Distillation

Knowledge distillation trains a smaller model to imitate a larger one. The smaller model may not match every capability, but it can be faster and easier to run inside a robot, camera, vehicle, or other power limited device.

Compute and Memory

Latency

Latency is the delay between an input and the system's response. In physical AI, even a highly accurate model can be unusable if the robot sees a hazard but reacts too late.

Models and Learning

Latent Space

Latent space is the model's compressed internal representation of patterns in data. Nearby points may represent similar objects, scenes, motions, or concepts even when the original inputs look different.

Models and Learning

Learning Rate

The learning rate controls how much a model changes after each training step. Too high can make learning unstable, while too low can make progress painfully slow.

Robotics and Actuation

Legged Robot

A legged robot walks using two or more legs. Legs can cross stairs and rough ground but require complex balance, control, power, and mechanical design.

Foundations

Level of Autonomy

A level of autonomy describes how much of a particular task a machine can perform without help. The level only makes sense when the task, environment, and required human role are clearly defined.

Sensors and Perception

LiDAR: Light Detection and Ranging

LiDAR sends laser light into the environment and measures how long reflections take to return. It can create accurate distance measurements and three dimensional maps for robots, vehicles, surveying, and industrial inspection.

Industry and Safety

Light Curtain

A light curtain creates an invisible field of beams across an opening. Breaking the field can command machinery to stop or prevent hazardous motion from starting.

Robotics and Actuation

Linear Actuator

A linear actuator creates movement along a straight line. It can push, pull, raise, or position a load using an electric screw, hydraulic cylinder, pneumatic cylinder, or another mechanism.

Robotics and Actuation

Linear Motor

A linear motor produces straight motion directly from electromagnetic force. It can move quickly and precisely without a screw or belt converting rotary motion.

Models and Learning

LLM: Large Language Model

An LLM is trained on large amounts of text to understand and generate language. In robotics, an LLM can help interpret instructions or plan tasks, but it usually needs perception, control, and safety systems before its output can affect a machine.

Navigation and Control

Local Planner

A local planner selects short term motion using current sensor data. It avoids people and new obstacles while trying to stay aligned with the global route.

Navigation and Control

Localization

Localization is the process of estimating where a robot or vehicle is located. It can combine maps, cameras, LiDAR, GPS, wheel movement, and inertial sensors because no single source is reliable in every environment.

Navigation and Control

Loop Closure

Loop closure happens when a mapping system recognizes a place it visited before. That recognition helps correct accumulated position error and align the map.

Models and Learning

Loss Function

A loss function measures how far a model's output is from the desired result during training. The training process adjusts the model to reduce this value.

Compute and Memory

MAC: Multiply Accumulate

A MAC operation multiplies two numbers and adds the result to a running total. Neural networks perform enormous numbers of these operations, so AI chips are designed to execute them efficiently.

Industry and Safety

Machine Tending

Machine tending uses a robot to load, unload, or support equipment such as machine tools, presses, and test stations. The complete task includes doors, fixtures, parts, and process signals.

Sensors and Perception

Machine Vision

Machine vision uses cameras, lighting, optics, and software to inspect or guide industrial processes. It is usually engineered for a specific task, such as checking labels or aligning parts, and it may use traditional algorithms, AI, or both.

Sensors and Perception

Magnetometer

A magnetometer measures magnetic fields and can act like a digital compass. It can help estimate heading, but nearby motors, steel structures, and electrical equipment may disturb its readings.

Robotics and Actuation

Manipulator

A manipulator is a robot mechanism designed to position and move objects or tools. A typical industrial robot arm is a manipulator made from linked joints with an end effector attached at the end.

Navigation and Control

Map Matching

Map matching aligns an estimated position or observed features with a known map. It can keep a vehicle on the correct road or a robot aligned with a facility layout.

Navigation and Control

Mapping

Mapping creates a useful representation of an environment. A map may show walls and obstacles, road features, shelf locations, or three dimensional surfaces, depending on what the machine needs to do.

Compute and Memory

Matrix Multiplication

Matrix multiplication combines grids of numbers according to a defined rule. It is one of the main calculations behind neural networks, graphics, simulation, and many robotics algorithms.

Compute and Memory

MCU: Microcontroller Unit

An MCU is a small computer on a single chip that controls a focused device or function. A robot may use several microcontrollers to read sensors, drive motors, monitor batteries, and perform reliable low level tasks.

Compute and Memory

Memory Bandwidth

Memory bandwidth measures how much data can move between memory and a processor in a given time. High bandwidth helps data heavy AI workloads keep processing units busy instead of waiting for model weights or sensor data.

Compute and Memory

Memory Capacity

Memory capacity is the amount of data a memory system can hold at one time. For AI hardware, capacity often determines whether a model fits on the device at all. It is different from how quickly the data can move.

Compute and Memory

Memory Controller

A memory controller manages how a processor reads from and writes to memory. Its design affects usable bandwidth, delay, reliability, and how several processing units share data.

Compute and Memory

Memory Hierarchy

A memory hierarchy organizes storage by speed, size, and cost. Tiny caches sit closest to the processor, working memory sits farther away, and larger storage is slower. Software performs best when frequently used data stays near the compute.

Industry and Safety

MES: Manufacturing Execution System

An MES tracks and coordinates production work on the factory floor. It connects schedules, materials, machines, quality records, and operators so a company can see what was made, how, and when.

Software and Connectivity

Message Broker

A message broker receives, routes, and sometimes stores messages between applications or devices. It helps separate data producers from the systems that consume their data.

Sensors and Perception

Microphone

A microphone converts sound into an electrical signal. Robots can use microphones for speech, alarms, equipment monitoring, impact detection, and awareness of nearby activity.

Sensors and Perception

Microphone Array

A microphone array combines several microphones to estimate where sound comes from, focus on a speaker, or reduce unwanted noise.

Software and Connectivity

Middleware

Middleware is software that helps different programs, devices, or services communicate. In a robot, it can move camera data, motor commands, maps, and status messages between components built by different teams.

Compute and Memory

Mixed Precision

Mixed precision uses different number formats within the same workload. Less precise formats can run faster and use less memory, while higher precision is kept where accuracy or stability needs it.

Foundations

ML: Machine Learning

Machine learning is a way of building software by learning patterns from data instead of writing a rule for every situation. A quality inspection model can learn what defects look like from examples of acceptable and defective parts.

Robotics and Actuation

Mobile Manipulator

A mobile manipulator combines a moving base with a robot arm. It can travel between work areas and physically interact with objects at each location.

Robotics and Actuation

Mobile Robot

A mobile robot moves through an environment instead of remaining fixed in one place. It may use wheels, tracks, legs, flight, or another form of locomotion.

Software and Connectivity

Modbus

Modbus is a long established industrial protocol for reading and writing device data. It is simple and common, though modern security usually must be added around it.

Models and Learning

Model Based Reinforcement Learning

Model based reinforcement learning uses an internal model to predict what may happen after an action. The agent can consider possible futures before acting in the real environment.

Models and Learning

Model Checkpoint

A model checkpoint is a saved copy of a model's learned state at a particular point in training. Teams use checkpoints to resume work, compare versions, or deploy a chosen model.

Models and Learning

Model Compression

Model compression reduces the storage, memory, or compute needed by a model. Techniques include quantization, pruning, and distillation, often with some tradeoff in accuracy or flexibility.

Models and Learning

Model Free Reinforcement Learning

Model free reinforcement learning learns useful behavior directly from experience without first learning an explicit model of how the environment changes.

Models and Learning

Model Parameter

A model parameter is a number adjusted during training. Together, millions or billions of parameters store the patterns a model uses to produce predictions and actions.

Models and Learning

Model Weights

Model weights are the learned parameter values inside a neural network. Saving the weights preserves what the model learned so it can later run inference or continue training.

Sensors and Perception

Monocular Camera

A monocular camera is a single camera viewing a scene from one position. It captures appearance well, while distance must be inferred from motion, known size, AI, or another sensor.

Navigation and Control

Motion Planning

Motion planning finds a feasible way for a robot to move while respecting its shape, joints, obstacles, and limits. For an arm, it helps ensure the elbow and links avoid collisions, not just the gripper.

Robotics and Actuation

Motor Controller

A motor controller decides how a motor should be driven based on commands and sensor feedback. It may regulate position, speed, torque, current, and protective limits.

Robotics and Actuation

Motor Driver

A motor driver is the power electronics that deliver controlled electrical current to a motor. The controller decides what should happen, while the driver converts that decision into the voltage and current the motor needs.

Navigation and Control

MPC: Model Predictive Control

MPC uses a model of the system to predict several possible futures and choose a useful control action while respecting limits. It can coordinate complex machines, but it requires enough computing power to solve the problem quickly.

Software and Connectivity

MQTT: Message Queuing Telemetry Transport

MQTT is a lightweight messaging protocol built around publishers, subscribers, and a broker. It is widely used to move device and sensor data across unreliable or limited networks.

Models and Learning

Multimodal AI

Multimodal AI can work with more than one kind of information, such as images, text, audio, video, or sensor data. A service robot may combine a spoken request with camera input to understand which object a person means.

Navigation and Control

Navigation Stack

A navigation stack is the collection of software that handles localization, maps, planning, obstacle avoidance, and motion commands for a mobile machine.

Software and Connectivity

Network Bandwidth

Network bandwidth is the amount of data a connection can carry over time. Video and point clouds need far more bandwidth than simple status messages.

Software and Connectivity

Network Topology

Network topology describes how devices and connections are arranged. The structure affects performance, resilience, troubleshooting, and what happens when a link fails.

Models and Learning

Neural Network

A neural network is a machine learning model built from connected layers of mathematical units. During training, those connections are adjusted so the model becomes better at tasks such as recognizing objects, predicting motion, or interpreting speech.

Compute and Memory

NPU: Neural Processing Unit

An NPU is a processor specialized for neural network workloads. It can run common AI tasks with less power than a general purpose processor, which makes it useful in phones, cameras, laptops, vehicles, and compact robots.

Compute and Memory

NVLink

NVLink is NVIDIA's high speed connection for moving data between supported GPUs, CPUs, and systems. It can provide more direct bandwidth than standard peripheral connections for large AI workloads.

Sensors and Perception

Object Detection

Object detection identifies which objects appear in an image and where they are located, usually with labeled boxes. A warehouse robot might detect people, pallets, forklifts, and open pathways in the same camera frame.

Navigation and Control

Obstacle Avoidance

Obstacle avoidance detects hazards and adjusts motion to prevent a collision. It often works locally and quickly, while a broader planning system decides the overall route or task.

Navigation and Control

Occupancy Grid

An occupancy grid divides an area into small cells and estimates whether each cell is free, blocked, or unknown. Mobile robots use these grids to represent navigable space and plan routes around obstacles.

Industry and Safety

ODD: Operational Design Domain

An ODD defines the conditions in which an automated system is designed to operate. It can include location, weather, lighting, speed, road type, floor condition, and the people or objects expected nearby.

Industry and Safety

OEE: Overall Equipment Effectiveness

OEE combines availability, performance, and quality into one manufacturing measure. It helps identify whether losses come from downtime, slower operation, or defective output.

Robotics and Actuation

Omnidirectional Drive

An omnidirectional drive lets a mobile robot move sideways as well as forward, backward, and around. Special wheel arrangements make this possible on relatively smooth floors.

Software and Connectivity

On Device AI

On device AI runs directly on the product rather than depending on a remote server. A smart camera can detect an event locally and send only the result instead of streaming every video frame to the cloud.

Models and Learning

Online Learning

Online learning updates a model as new data arrives rather than waiting for one large training cycle. In physical systems, updates require careful controls so field data does not create unsafe behavior.

Software and Connectivity

OPC UA: Open Platform Communications Unified Architecture

OPC UA is an industrial communication standard for sharing structured machine data securely across vendors and systems. It helps connect equipment, control software, and business applications.

Navigation and Control

Open Loop Control

Open loop control sends a command without measuring whether the desired result occurred. It is simple and useful when conditions are predictable, but it cannot automatically correct for changing loads, wear, or disturbances.

Compute and Memory

OpenCL: Open Computing Language

OpenCL is a standard for writing programs that can run across different processors, including CPUs, GPUs, and FPGAs. It offers portability, though support and performance vary by device.

Industry and Safety

OTA: Over the Air Update

An OTA update delivers software or model changes through a network instead of a physical service visit. Secure updates can improve a fleet quickly, but teams need testing, staged rollout, authentication, and a recovery plan if an update fails.

Navigation and Control

Overshoot

Overshoot occurs when a system moves past its target before settling. Too much overshoot can reduce precision, spill material, or create unsafe contact.

Software and Connectivity

Packet Loss

Packet loss occurs when transmitted pieces of data never reach their destination. A system may recover, reduce quality, delay action, or fail depending on the application.

Industry and Safety

Palletizing

Palletizing stacks cases, bags, or products onto a pallet according to a planned pattern. Robots can reduce repetitive lifting and adapt patterns when product data is available.

Compute and Memory

Parallel Computing

Parallel computing divides work so many calculations can happen at once. This is useful for AI because training and inference repeatedly apply similar math across large groups of numbers, images, or sensor readings.

Navigation and Control

Particle Filter

A particle filter tracks many possible estimates of a system's state and gives more weight to those that match new measurements. It is useful when uncertainty has several possible answers.

Navigation and Control

Path Planning

Path planning finds a route from a starting point to a goal while avoiding blocked or unsafe areas. It usually describes where to go, not the exact motor commands or timing needed to get there.

Industry and Safety

Payback Period

Payback period estimates how long it takes for accumulated savings or profit to recover the initial investment. It is simple but does not capture every risk or later benefit.

Robotics and Actuation

Payload

Payload is the weight or load a robot is designed to carry, often including the end effector. The rated maximum may only apply under certain positions, speeds, and distances from the robot's wrist.

Compute and Memory

PCIe: Peripheral Component Interconnect Express

PCIe is a common high speed connection used to attach GPUs, accelerators, storage, and networking hardware to a computer. Newer generations move more data, but both connected devices must support the same capabilities to benefit.

Foundations

Perception

Perception is the process of turning sensor data into an understanding of what is happening. For a robot, this might mean finding a box in a camera image, measuring its distance, and tracking whether it moves.

Industry and Safety

PFL: Power and Force Limiting

PFL limits robot power, force, pressure, and motion so contact remains within defined safety limits. The complete tool, payload, speed, and body region must be evaluated.

Sensors and Perception

Photodiode

A photodiode converts light into electrical current. It is a basic sensing element inside cameras, optical encoders, safety scanners, and many distance sensors.

Simulation and Data

Photorealistic Rendering

Photorealistic rendering creates images intended to resemble real camera output. Realistic appearance can help vision training, but visual realism alone does not guarantee accurate physics.

Foundations

Physical AI

Physical AI describes intelligent systems that sense the real world, make a decision, and cause or coordinate a physical action. A warehouse robot that sees a blocked aisle, chooses another route, and checks its progress is using physical AI.

Foundations

Physical Intelligence

Physical intelligence is the ability to understand and act within the constraints of the real world. It includes ideas such as balance, contact, motion, object behavior, and the consequences of an action.

Simulation and Data

Physics Engine

A physics engine calculates how virtual objects move, collide, fall, bend, or respond to forces. Robotics simulators use it to test motion and contact before running the same behavior on physical hardware.

Industry and Safety

Pick and Place

Pick and place means taking an item from one location and placing it in another. Success depends on perception, grasping, motion, cycle time, and handling mistakes.

Navigation and Control

PID Controller: Proportional Integral Derivative Controller

A PID controller corrects error using its current size, its history, and how quickly it is changing. It is widely used because a well tuned PID can control speed, temperature, position, and many other physical processes.

Industry and Safety

PL: Performance Level

Performance Level expresses the ability of a safety related control function to reduce risk. Categories run from lower to higher reliability and depend on architecture, diagnostics, and component reliability.

Foundations

Planning

Planning means choosing a sequence of steps that can move a system from its current state toward a goal. A mobile robot may plan which aisle to use, while a robot arm plans how to approach and grasp an object.

Industry and Safety

PLC: Programmable Logic Controller

A PLC is a rugged industrial computer that controls machines and processes. It reads inputs, runs reliable control logic, and switches outputs. Physical AI often works alongside PLCs rather than replacing them.

Robotics and Actuation

Pneumatic Actuator

A pneumatic actuator uses compressed air to create movement. It can be simple, fast, and inexpensive for factory tasks, although air compression and precise control can be less efficient than electric alternatives.

Sensors and Perception

Point Cloud

A point cloud is a collection of points that represent the shape of a three dimensional scene or object. LiDAR and depth cameras commonly produce point clouds that robots use for mapping, measurement, and obstacle detection.

Models and Learning

Policy

A policy is the decision rule a robot uses to choose an action from its current situation. It may be hand programmed, learned from examples, or improved through reinforcement learning. The policy is the part that answers what should I do next.

Navigation and Control

Pose

Pose describes an object's position and orientation. Knowing that a part is on a table is not enough for grasping. A robot also needs to know which way the part is facing.

Navigation and Control

Position Control

Position control commands a mechanism to reach and hold a target location or angle. Feedback measures the actual position so the controller can correct error.

Models and Learning

Post Training

Post training is the work done after a model's main training phase to improve its behavior, efficiency, or usefulness. It can include instruction tuning, safety tuning, quantization, or learning from human and robot feedback.

Compute and Memory

Power Efficiency

Power efficiency describes how much useful work a system performs for the energy it consumes. It matters in battery powered robots and drones, but also in factories and data centers where heat, cooling, and electricity affect operating cost.

Industry and Safety

Predictive Maintenance

Predictive maintenance uses condition data to estimate when equipment may need service. The goal is to act before failure without replacing healthy parts too early.

Sensors and Perception

Pressure Sensor

A pressure sensor measures force spread over an area. It is used in pneumatic and hydraulic systems, process equipment, tactile surfaces, and fluid handling.

Models and Learning

Pretraining

Pretraining gives a model broad abilities before it is adapted to a specific job. A robot model might first learn from large image, video, language, and motion datasets, then receive additional training for warehouse picking.

Software and Connectivity

Private 5G

Private 5G is a cellular network dedicated to one organization or site. It can provide managed coverage and device control for factories, ports, campuses, and mobile machines.

Software and Connectivity

PROFINET

PROFINET is an industrial Ethernet system used to connect controllers, drives, sensors, and other automation equipment with different timing requirements.

Sensors and Perception

Proprioception

Proprioception is a machine's sense of its own body. Joint encoders, motor currents, and force sensors can tell a robot where its limbs are, how they are moving, and what loads they are experiencing.

Software and Connectivity

Protocol

A protocol is a shared set of rules for formatting and exchanging information. Devices can only communicate successfully when they agree on the relevant protocol.

Sensors and Perception

Proximity Sensor

A proximity sensor detects whether an object is nearby without requiring physical contact. Different designs use light, magnetic fields, sound, or electrical properties to detect particular materials and distances.

Models and Learning

Pruning

Pruning removes model connections or values that contribute little to the result. A well pruned model can use less memory and computing while keeping most of its accuracy.

Software and Connectivity

Pub Sub: Publish Subscribe

Publish subscribe communication lets senders publish messages to a named channel while interested receivers subscribe. The participants do not need direct knowledge of one another.

Models and Learning

Q Learning

Q learning estimates how valuable an action is in a particular state. The system learns which choices are most likely to produce a strong long term reward.

Robotics and Actuation

QDD: Quasi Direct Drive

A quasi direct drive actuator combines a high torque motor with modest gearing. It aims for strong output while keeping friction, backlash, and resistance to outside motion relatively low.

Software and Connectivity

QoS: Quality of Service

Quality of service settings control how communication handles priority, reliability, delay, history, and delivery. They help a system treat safety data differently from optional telemetry.

Robotics and Actuation

Quadruped Robot

A quadruped robot walks on four legs. The form provides a useful balance between stability and mobility for inspection, research, public safety, and rough terrain.

Industry and Safety

Quality Inspection

Quality inspection checks whether a product or process meets defined requirements. Sensors and AI can find patterns, but acceptance rules and measurement evidence still need clear engineering ownership.

Models and Learning

Quantization

Quantization represents a model's numbers with fewer bits. This can reduce memory use, power use, and inference time, which helps models fit on edge devices. The tradeoff is that too much compression can reduce accuracy.

Sensors and Perception

Radar: Radio Detection and Ranging

Radar uses radio waves to detect distance, direction, and often speed. It can work in darkness and difficult weather, making it useful for vehicles, drones, safety systems, and measuring motion around industrial equipment.

Simulation and Data

Ray Tracing

Ray tracing simulates how light travels and interacts with surfaces. It can produce realistic images and model sensors whose behavior depends on reflection and visibility.

Robotics and Actuation

Reach

Reach describes how far a robot can extend its working point from its base or body. Maximum reach does not mean every pose, payload, or orientation is possible throughout that entire space.

Navigation and Control

Real Time Computing

Real time computing means producing a result within a required time limit, not simply computing as fast as possible. A safety signal that arrives after the deadline is a failure even if the calculation was correct.

Foundations

Reasoning

Reasoning is the process of connecting information to reach a conclusion or choose a course of action. A robot may reason that a glass is fragile, so it should reduce grip force and move more slowly.

Industry and Safety

Redundancy

Redundancy adds another component, signal, or method so one failure does not immediately remove an important function. Safety systems may compare independent sensors or controllers, but extra parts only help when failures are truly separated and detected.

Robotics and Actuation

Repeatability

Repeatability measures how consistently a robot returns to the same position under the same conditions. A robot can be highly repeatable even if that position is slightly offset from the commanded one.

Models and Learning

Reward Model

A reward model estimates how desirable an AI output or action is. It can help guide training when quality is easier to judge than to describe with a simple formula.

Sensors and Perception

RGB Camera: Red Green Blue Camera

An RGB camera records ordinary color images using red, green, and blue channels. It provides rich visual detail at relatively low cost, but a single RGB image does not directly measure distance.

Sensors and Perception

RGBD Camera: Red Green Blue Depth Camera

An RGBD camera provides color images and depth measurements together. Robots use the combined data to recognize an object and estimate where its surfaces are in space.

Simulation and Data

Rigid Body Simulation

Rigid body simulation models objects that move and collide without changing shape. It works well for many robot links, tools, boxes, and machine parts.

Industry and Safety

Risk Assessment

A risk assessment identifies hazards, estimates likelihood and severity, and chooses measures to reduce risk. It must consider the complete application, not only the robot.

Models and Learning

RL: Reinforcement Learning

Reinforcement learning teaches a system through rewards and penalties. A robot tries actions, observes the result, and gradually favors behavior that earns better outcomes. Simulation is often used because real robot trial and error can be slow or unsafe.

Models and Learning

RLHF: Reinforcement Learning from Human Feedback

RLHF improves a model using human preferences or evaluations. People compare outputs, and that feedback helps train the model toward behavior judged more useful or appropriate.

Foundations

Robot

A robot is a programmed machine that senses, moves, or manipulates objects with some degree of independence. A robot can use advanced AI, simple automation, or both. Having a mechanical body does not automatically make a machine intelligent.

Robotics and Actuation

Robot Arm

A robot arm is a linked mechanism that positions a tool or object. Its joints, reach, payload, accuracy, and end effector determine which tasks it can perform.

Simulation and Data

Robot Demonstration

A robot demonstration is a recorded example of how a task should be performed. It may come from direct operation, remote control, motion capture, or a human physically guiding the robot.

Robotics and Actuation

Robot Joint

A robot joint connects links and allows a particular kind of movement, such as rotation or sliding. Its actuator, transmission, bearings, sensors, wiring, and controls often form one integrated module.

Foundations

Robotics

Robotics is the field concerned with designing, building, programming, and operating robots. It brings together mechanical engineering, electronics, software, sensing, control, and increasingly machine learning.

Simulation and Data

Robotics Simulator

A robotics simulator provides a virtual place to develop and test robot software, sensors, motion, and environments before using physical equipment.

Compute and Memory

ROCm: Radeon Open Compute

ROCm is AMD's open software platform for GPU computing. It provides tools and libraries for AI, scientific computing, and other workloads on supported AMD hardware.

Industry and Safety

ROI: Return on Investment

ROI compares the financial benefit of an investment with its cost. A physical AI calculation should include integration, downtime, service, labor changes, risk, and actual utilization.

Sensors and Perception

Rolling Shutter

A rolling shutter records different image rows at slightly different times. It can reduce camera cost, but rapid motion may make straight objects appear bent or stretched.

Software and Connectivity

ROS 2: Robot Operating System 2

ROS 2 is the newer generation of ROS designed for modern robotics, distributed systems, security options, and more dependable communication across different computers.

Software and Connectivity

ROS Action

A ROS action manages a longer task that provides progress and can be canceled. Driving to a waypoint is a typical action because it takes time and may need interruption.

Software and Connectivity

ROS Bag

A ROS bag records messages from a ROS system for later playback and analysis. It helps reproduce field conditions without needing the robot present.

Software and Connectivity

ROS Node

A ROS node is a program that performs one logical job inside a ROS system. One node may read a camera, another may detect objects, and another may send movement commands.

Software and Connectivity

ROS Service

A ROS service handles a short request and response. One node asks for a calculation or state, and another returns the result.

Software and Connectivity

ROS Topic

A ROS topic is a named channel for continuous streams of messages. A camera node can publish images to a topic, and any authorized node that needs those images can subscribe.

Software and Connectivity

ROS: Robot Operating System

ROS is an open source collection of software tools and communication patterns for building robots. Despite the name, it is usually middleware running on an operating system rather than the operating system itself.

Robotics and Actuation

Rotary Actuator

A rotary actuator creates turning motion. Motors are the most familiar example, but hydraulic and pneumatic rotary actuators are also used where a machine needs to rotate a joint, shaft, valve, or tool.

Navigation and Control

Route Planning

Route planning chooses a broader path through a road network, facility, or delivery area. It may consider distance, traffic, restrictions, energy, and task priority.

Navigation and Control

RTK: Real Time Kinematic Positioning

RTK improves satellite positioning by using correction data from a known reference. Under good conditions, it can provide centimeter level accuracy for surveying, agricultural machines, drones, and autonomous vehicles.

Navigation and Control

RTOS: Real Time Operating System

An RTOS is an operating system designed to run important tasks with predictable timing. It is common in motor controllers, drones, vehicles, medical devices, and industrial equipment where missed deadlines can affect performance or safety.

Compute and Memory

Runtime

A runtime is the software environment that manages a program while it is executing. It can handle memory, device communication, scheduling, and access to processor specific functions.

Industry and Safety

Safety Laser Scanner

A safety laser scanner monitors a defined area and triggers a protective response when a person or object enters. Its placement and configured zones are part of the safety design.

Industry and Safety

Safety PLC: Safety Programmable Logic Controller

A safety PLC is designed and certified to perform safety functions with monitored, fault tolerant behavior. It may manage emergency stops, guards, scanners, and safe motion.

Industry and Safety

SAT: Site Acceptance Test

A SAT tests equipment after installation at the operating site. It confirms connections, utilities, safety, interfaces, and performance in the real environment.

Industry and Safety

SCADA: Supervisory Control and Data Acquisition

SCADA software monitors and supervises equipment spread across a plant, utility, or large site. It gathers data, displays conditions, records history, and lets authorized operators send higher level commands.

Robotics and Actuation

SCARA: Selective Compliance Assembly Robot Arm

A SCARA robot moves quickly in a horizontal plane while remaining stiff vertically. It is widely used for assembly, insertion, packaging, and moving small parts.

Simulation and Data

Scenario

A scenario is a defined situation used to train or test a system. It specifies relevant actors, conditions, events, and expected outcomes.

Simulation and Data

Scenario Generation

Scenario generation creates varied situations for simulation or testing. Good generators include normal work, rare events, difficult combinations, and realistic failure conditions.

Software and Connectivity

SDK: Software Development Kit

An SDK is a collection of code, tools, examples, and documentation for building applications on a platform, device, or service.

Robotics and Actuation

SEA: Series Elastic Actuator

A series elastic actuator places a spring between the motor and the load. The spring helps measure force, absorb impacts, and create safer contact behavior.

Models and Learning

Self Supervised Learning

Self supervised learning creates learning tasks from the data itself. A model might hide part of an image or video and learn to predict what is missing, allowing it to learn useful structure without extensive human labeling.

Sensors and Perception

Semantic Segmentation

Semantic segmentation assigns a category to each pixel in an image. This gives a machine a detailed map of areas such as floor, wall, person, vehicle, vegetation, or defect.

Compute and Memory

Semiconductor Process Node

A process node names a generation of chip manufacturing technology, often with a number such as 5 nanometer or 3 nanometer. A smaller number does not guarantee a better product, but newer nodes can improve density, speed, or efficiency.

Sensors and Perception

Sensor

A sensor measures something about a machine or its environment and turns it into data. Cameras measure light, encoders measure motion, and temperature sensors measure heat. Sensors give physical AI its connection to current reality.

Sensors and Perception

Sensor Calibration

Sensor calibration measures and corrects how a sensor differs from the real value or expected geometry. Poor calibration can make accurate hardware produce misleading results.

Sensors and Perception

Sensor Fusion

Sensor fusion combines information from several sensors to create a more reliable view of the world. A mobile robot may combine cameras, LiDAR, wheel encoders, and an IMU because no single sensor works perfectly everywhere.

Sensors and Perception

Sensor Noise

Sensor noise is unwanted variation in a measurement. It can come from electronics, heat, vibration, lighting, interference, or the limits of the sensing method.

Navigation and Control

Sensor Registration

Sensor registration aligns data from different sensors into the same space and time. Without it, a camera object and a LiDAR point may appear to be in different places.

Robotics and Actuation

Servo Motor

A servo motor works with feedback so a controller can command position, speed, or torque precisely. Industrial robots rely on servo systems to move joints accurately and respond to changing loads.

Navigation and Control

Setpoint

A setpoint is the desired value a controller tries to maintain, such as a speed, position, temperature, pressure, or force.

Navigation and Control

Settling Time

Settling time is how long a system takes to reach and remain near its target after a command or disturbance.

Industry and Safety

SIL: Safety Integrity Level

SIL is a functional safety measure describing the required risk reduction and reliability of a safety function under specific standards. A higher level requires stronger evidence and engineering controls.

Simulation and Data

SIL: Software in the Loop

Software in the loop testing runs production software against simulated hardware and environments. It helps find logic problems early and supports repeatable automated tests.

Compute and Memory

Silicon Interposer

A silicon interposer is a thin layer that provides dense electrical connections between chips in one package. It is commonly used to place HBM close to a GPU or accelerator.

Models and Learning

Sim to Real

Sim to real is the process of taking a model or robot skill developed in simulation and making it work on real equipment. The challenge is closing the gap between the clean simulated world and messy physical reality.

Simulation and Data

Simulation

Simulation creates a virtual environment where machines, sensors, and physical events can be tested. It can reduce cost and risk, but results are only as trustworthy as the models, assumptions, and validation behind them.

Compute and Memory

SiP: System in Package

A SiP places several separate chips inside one package so they work as a compact system. It can combine processors, memory, radios, and other components without manufacturing them as one piece of silicon.

Foundations

Situated Intelligence

Situated intelligence means intelligence shaped by the current environment, body, and task. The same action can be useful in one situation and unsafe or ineffective in another.

Navigation and Control

SLAM: Simultaneous Localization and Mapping

SLAM lets a machine build a map while also estimating its position inside that map. It is useful when a robot enters an unfamiliar area or cannot rely on a strong external positioning signal.

Compute and Memory

SoC: System on Chip

An SoC combines several computer functions on one chip, often including CPU cores, graphics, AI acceleration, memory controllers, and communications. This saves space and power in robots, drones, vehicles, and smart cameras.

Simulation and Data

Soft Body Simulation

Soft body simulation models objects that bend, stretch, or deform. It is useful for cloth, cables, food, soft grippers, and flexible materials.

Robotics and Actuation

Soft Robotics

Soft robotics uses flexible materials and compliant structures rather than only rigid links. Soft systems can conform to delicate or irregular objects but may be harder to model precisely.

Robotics and Actuation

Solenoid

A solenoid uses an electromagnetic coil to create short linear movement. It is common in locks, latches, valves, and simple switching mechanisms.

Models and Learning

Sparsity

Sparsity means many values in a model or dataset are zero or unused. Hardware and software designed for sparse work can skip unnecessary calculations and save energy.

Foundations

Spatial Intelligence

Spatial intelligence is the ability to understand positions, shapes, distances, directions, and relationships in space. Robots need it to navigate rooms, reach objects, and reason about where things can fit.

Compute and Memory

SRAM: Static Random Access Memory

SRAM is fast memory that does not need the constant refreshing used by DRAM. It is more expensive and takes more chip area, so it is commonly used for small, fast caches and buffers close to processing units.

Industry and Safety

SRMS: Safety Rated Monitored Stop

A safety rated monitored stop brings robot motion to a stop and monitors that condition while a person is in the shared workspace.

Industry and Safety

SSM: Speed and Separation Monitoring

SSM adjusts or stops robot motion based on the measured distance between people and the hazardous system. Safe sensing and stopping performance are essential.

Navigation and Control

Stability

Stability means a controlled system remains bounded and returns toward expected behavior after a disturbance. An unstable controller can make motion grow into dangerous oscillation.

Navigation and Control

State Estimation

State estimation combines measurements and a model to estimate important conditions that cannot be observed perfectly. It might track a robot's position, speed, orientation, battery condition, or the motion of nearby objects.

Robotics and Actuation

Stepper Motor

A stepper motor moves in controlled increments called steps. It is useful for positioning at modest cost, though it can lose position if the load exceeds its ability unless feedback is added.

Sensors and Perception

Stereo Vision

Stereo vision uses two cameras with slightly different viewpoints to estimate depth, much like human eyes. The system matches the same features in both images and uses their difference to calculate distance.

Robotics and Actuation

Stiffness

Stiffness describes how strongly a structure resists deformation under load. A stiff robot holds position well, while deliberate flexibility may improve safe contact.

Industry and Safety

STO: Safe Torque Off

Safe torque off prevents a motor drive from producing torque through a safety rated function. Stored energy, gravity, and other hazards may still require additional measures.

Sensors and Perception

Strain Gauge

A strain gauge measures tiny changes in a material as it stretches or compresses. Load cells and force sensors often use strain gauges to measure applied force.

Sensors and Perception

Structured Light

Structured light projects a known pattern onto a scene and measures how the pattern bends across surfaces. That distortion can be converted into a three dimensional shape.

Models and Learning

Supervised Learning

Supervised learning trains a model with examples that include the correct answer. A defect detector might learn from images labeled acceptable, scratched, cracked, or missing a component.

Simulation and Data

Synthetic Data

Synthetic data is created by software instead of collected directly from the real world. It can produce rare, dangerous, or expensive training examples, but the model still needs testing on real operating conditions.

Simulation and Data

Synthetic Environment

A synthetic environment is a digitally created world used for training, testing, or visualization. It may represent a factory, road, warehouse, home, or outdoor site.

Simulation and Data

System Identification

System identification builds a mathematical model from measured input and output data. Engineers use it when a physical system is too complex or uncertain to model from theory alone.

Industry and Safety

Systems Integrator

A systems integrator combines equipment, software, controls, safety, and workflows into a working solution. Integration quality often determines whether promising technology delivers useful production.

Sensors and Perception

Tactile Sensor

A tactile sensor measures contact, pressure, or texture. Placed on a robot hand or gripper, it can help detect whether an object is slipping, how firmly it is held, or where contact occurs.

Industry and Safety

Takt Time

Takt time is the production pace needed to meet customer demand with the available working time. It provides a target rhythm for balancing work.

Industry and Safety

TCO: Total Cost of Ownership

TCO includes purchase price plus integration, infrastructure, software, energy, maintenance, support, training, downtime, and disposal over the useful life of a system.

Compute and Memory

TDP: Thermal Design Power

TDP is a planning value used to size cooling for a processor or chip. It is not always the same as actual electrical use, but it helps engineers judge whether a device can operate safely inside a compact enclosure.

Industry and Safety

Telemetry

Telemetry is data sent from a machine so its condition and behavior can be monitored elsewhere. It may include location, battery level, temperatures, errors, task progress, sensor health, and performance trends.

Simulation and Data

Telemetry Log

A telemetry log is a recorded history of machine status and events. Teams use logs to reproduce failures, compare performance, and understand what happened before an incident.

Simulation and Data

Teleoperation

Teleoperation means controlling a machine from a distance. It can handle tasks that are too dangerous or uncertain for full autonomy and can also collect demonstrations for training robot policies.

Sensors and Perception

Temperature Sensor

A temperature sensor measures heat in a component or environment. Robots use these readings to protect motors, batteries, processors, and power electronics.

Compute and Memory

Tensor Core

A tensor core is a specialized calculation unit inside some accelerators. It performs the matrix math used by neural networks very quickly, helping speed up model training and inference.

Simulation and Data

Test Set

A test set is held back until evaluation so it can provide a fairer measure of performance on unseen examples.

Sensors and Perception

Thermal Camera

A thermal camera detects infrared energy related to temperature rather than visible color. It can find overheating equipment, inspect buildings, see through some low light conditions, and identify temperature differences invisible to ordinary cameras.

Compute and Memory

Thermal Throttling

Thermal throttling reduces processor speed when temperature becomes too high. It protects hardware but can cause a robot or edge system to perform more slowly during sustained workloads.

Compute and Memory

Throughput

Throughput measures how much work a system completes over time. A vision server may have high throughput because it processes many camera frames, even if each individual frame takes longer than it would on a low latency edge device.

Simulation and Data

Time Series Data

Time series data is a sequence of measurements ordered by time. Motor current, vibration, temperature, position, and production rate are common industrial examples.

Sensors and Perception

ToF Sensor: Time of Flight Sensor

A time of flight sensor measures distance by timing how long emitted light or sound takes to return. Many depth cameras use infrared light for this measurement.

Models and Learning

Token

A token is a small unit of information processed by a model. In language it may represent part of a word. In robotics, tokens can also represent images, sensor states, or actions.

Compute and Memory

TOPS: Trillions of Operations per Second

TOPS is a common peak performance measure for AI accelerators. It counts trillions of operations per second, often at a specific number format. A higher TOPS number does not automatically mean an application will run faster.

Robotics and Actuation

Torque

Torque is a turning force. More torque can help a joint lift a heavier load or accelerate faster, but the useful result also depends on speed, gearing, arm length, heat, and control.

Navigation and Control

Torque Control

Torque control regulates the turning force produced by a motor. It allows more natural contact behavior and precise force management in robot joints.

Compute and Memory

TPU: Tensor Processing Unit

A TPU is a processor designed by Google for machine learning calculations. It is built to handle the matrix operations used by neural networks efficiently and appears in Google cloud infrastructure and selected edge devices.

Models and Learning

Training

Training is the process through which a model learns from examples or experience. The model makes predictions, measures its mistakes, and adjusts its internal parameters. Most training happens before a model is installed on a machine.

Simulation and Data

Training Set

A training set is the portion of a dataset used to adjust a model's parameters. It should represent the situations the model needs to learn.

Navigation and Control

Trajectory Planning

Trajectory planning adds timing to a planned movement. It defines where a robot should be, how fast it should move, and how its speed should change throughout the motion.

Models and Learning

Transfer Learning

Transfer learning reuses knowledge learned for one task or dataset in another. A model trained to recognize everyday objects may provide a stronger starting point for industrial inspection than a model trained from nothing.

Models and Learning

Transformer

A transformer is a neural network architecture that learns which parts of its input deserve attention. It became important in language models and is now widely used for vision, video, robotics, and models that combine several kinds of data.

Software and Connectivity

TSN: Time Sensitive Networking

TSN is a group of Ethernet standards that make delivery timing more predictable. It helps ordinary network technology support industrial control, synchronized machines, and other applications where important data cannot arrive whenever it happens to arrive.

Robotics and Actuation

UAV: Uncrewed Aerial Vehicle

A UAV is an aircraft operated without a pilot onboard. It may be remotely controlled, partly autonomous, or highly autonomous depending on the mission and regulations.

Also called Drone

Robotics and Actuation

UGV: Uncrewed Ground Vehicle

A UGV is a ground machine operated without a person onboard. Examples include inspection rovers, defense vehicles, farm machines, and robots used in dangerous sites.

Sensors and Perception

Ultrasonic Sensor

An ultrasonic sensor sends out high frequency sound and measures the returning echo. It provides simple distance or presence information at low cost, though soft materials, angles, and environmental noise can affect readings.

Compute and Memory

Unified Memory

Unified memory lets different processors access a shared pool of memory instead of maintaining completely separate copies. This can simplify programming and reduce unnecessary data movement between the CPU and an accelerator.

Models and Learning

Unsupervised Learning

Unsupervised learning looks for patterns in data without being given a correct label for every example. It can help group similar machine behaviors, discover unusual operating conditions, or organize large sensor datasets.

Industry and Safety

Validation

Validation confirms that a system satisfies the real user need and intended use. Verification asks whether it was built to specification. Validation asks whether the right system was built.

Simulation and Data

Validation Set

A validation set is separate data used during development to compare models and choose settings without judging them on the final test set.

Robotics and Actuation

Valve

A valve controls the direction, pressure, or flow of air or liquid. Hydraulic and pneumatic machines use valves to command actuators and manage force.

Navigation and Control

Velocity Control

Velocity control regulates how fast a motor, wheel, or joint moves. It is commonly used inside larger systems that plan position and motion.

Simulation and Data

Virtual Commissioning

Virtual commissioning tests machine controls and production logic against a digital model before installation. It can reduce startup time and reveal integration problems earlier.

Models and Learning

VLA: Vision Language Action Model

A VLA model connects what a robot sees, what a person asks, and the actions needed to complete the task. A request such as place the red cup in the sink can become a sequence of robot movements.

Models and Learning

VLM: Vision Language Model

A VLM works with images and language together. It can describe a scene, answer questions about an image, or connect a written instruction to something visible. By itself, a VLM usually explains what it sees rather than controlling a robot.

Navigation and Control

VO: Visual Odometry

Visual odometry estimates movement by comparing features across a sequence of camera images. It can help a robot or vehicle track how far it moved and how its direction changed between frames.

Compute and Memory

VPU: Vision Processing Unit

A VPU is a low power processor designed for computer vision. It can run tasks such as object detection or depth processing inside a camera or edge device without depending on a large GPU.

Compute and Memory

VRAM: Video Random Access Memory

VRAM is the memory available to a graphics processor. For AI, VRAM determines how large a model or batch of data can stay close to the GPU without repeatedly moving information from slower system memory.

Models and Learning

WAM: World Action Model

A world action model connects possible actions with how the world may change afterward. Rather than only predicting the next image, it can help a robot consider which action is most likely to produce the desired result.

Navigation and Control

Waypoint

A waypoint is a location a robot, drone, or vehicle is instructed to reach. A mission may connect several waypoints while local control handles the exact movement.

Models and Learning

WFM: World Foundation Model

A world foundation model learns patterns about how physical scenes and events change over time. It can help generate simulations, predict possible future states, or provide useful knowledge for robots and autonomous vehicles.

Sensors and Perception

Wheel Odometry

Wheel odometry estimates movement from wheel rotation. It is simple and useful, but errors grow when wheels slip, skid, or travel over uneven surfaces.

Robotics and Actuation

Wheeled Robot

A wheeled robot moves efficiently on suitable surfaces. Wheels simplify mechanics and energy use, but stairs, gaps, loose terrain, and large obstacles can limit access.

Software and Connectivity

WiFi: Wireless Fidelity

WiFi connects devices over local wireless networks. It is convenient for robots and sensors, but coverage, interference, roaming, congestion, and security affect industrial reliability.

Robotics and Actuation

Workspace

A robot's workspace is the full region its tool can reach. Joint limits, collisions, tool orientation, payload, and the surrounding equipment reduce the practically usable area.

Industry and Safety

Zero Trust

Zero trust is a security approach that does not automatically trust a user or device because of its location. Access is verified, limited, monitored, and reviewed continuously.