Glossary
The working vocabulary of physical AI, defined in a sentence or two. Each definition is written to make sense on its own.
- Action chunkingRobot Models
- Having a policy predict a short sequence of future actions at once instead of one action per step. It smooths motion and lets a slow model drive a fast control loop.
Read the article → - Autonomous mobile robotAMR · Hardware
- A mobile robot that plans its own path through a facility using onboard sensing and a map, rather than following fixed guides such as tape or wires the way an automated guided vehicle does.
- Collaborative applicationCobot application · Safety & Risk
- A robot application in which people and the robot share a workspace during operation. Standards treat collaboration as a property of the whole application — robot, tool, workpiece and task — not of the robot arm alone.
Read the article → - Degrees of freedomDoF · Hardware
- The number of independent ways a mechanism can move. A typical industrial arm has six; a humanoid with dexterous hands can have forty or more.
Read the article → - Domain randomizationSimulation & Data
- Training a policy in simulation while randomly varying physics and visual parameters, so that the real world looks like just one more variation it has already seen.
Read the article → - Embodied AIFoundations
- AI that learns and acts through a body situated in an environment. The older, research-oriented name for what industry now mostly calls physical AI.
Read the article → - End effectorHardware
- The device at the end of a robot arm that interacts with the world: a gripper, suction cup, hand, welding torch or other tool. It is part of the robot application and must be included in its risk assessment.
- Functional safetySafety & Risk
- The part of a machine’s safety that depends on a control system performing correctly — for example, stopping the robot when a light curtain is broken. Its required reliability is expressed as a performance level or safety integrity level.
Read the article → - Imitation learningRobot Models
- Training a robot policy from demonstrations of the task, usually collected by a human teleoperating the robot, rather than from a reward signal.
- Performance levelPL · Safety & Risk
- A rating from PL a to PL e, defined in ISO 13849-1, of how reliably a safety-related part of a control system performs its safety function. PL e is the most demanding.
Read the article → - Physical AIFoundations
- AI systems that perceive the physical world, decide what to do, and act on it through a machine body such as a robot arm, a mobile robot, a humanoid or a vehicle.
Read the article → - PolicyRobot Models
- The function that maps what a robot observes to what it does next. In learned systems the policy is a neural network; in classical systems it is hand-written control logic.
- Power and force limitingPFL · Safety & Risk
- A collaborative safeguarding method in which contact between robot and person is allowed, but force and pressure are kept below limits set for each region of the body.
Read the article → - Reinforcement learningRL · Robot Models
- Training a policy by trial and error against a reward signal. In robotics it is mostly done in simulation, where millions of attempts are cheap and failures break nothing.
- Risk assessmentSafety & Risk
- The structured process of setting a machine’s limits, identifying its hazards, and estimating and evaluating the risk of each one, so that risk can be reduced to an acceptable level. ISO 12100 defines the method.
Read the article → - Robot foundation modelRobot Models
- A large model pretrained on broad robot and non-robot data that can be adapted to many tasks and robot bodies, instead of being trained from scratch for one.
Read the article → - Sim-to-real gapSimulation & Data
- The drop in performance when a policy trained in simulation runs on a physical robot, caused by differences between simulated and real physics, sensing and timing.
Read the article → - Speed and separation monitoringSSM · Safety & Risk
- A collaborative safeguarding method in which sensors track the distance between a person and the robot, and the robot slows and then stops before that distance falls below a calculated protective separation distance.
Read the article → - TeleoperationSimulation & Data
- A human controlling a robot remotely in real time. It is used both to collect demonstration data for training and to recover robots in the field when autonomy fails.
- Vision-language-action modelVLA · Robot Models
- A model that takes camera images and a natural-language instruction as input and outputs robot actions. Most are built by adding action output to a pretrained vision-language model.
Read the article → - World modelRobot Models
- A model that predicts how the environment will change in response to actions. Robots can use one to plan, to evaluate a policy without running it, or to generate synthetic training data.