Robotics glossary
The key technical terms in robotics and AI – briefly explained, with references to real models.
- Actuator
- The “muscle” of a robot: a drive unit made up of a motor, gearbox and sensors that moves a joint. Modern humanoids rely almost exclusively on electric actuators; Boston Dynamics switched Atlas from hydraulic to electric drives.
- Auto-empty station
- A base station that automatically empties a robot vacuum's dustbin.
- Axial flux motor
- A flat electric motor with high torque density – attractive for compact, powerful robot joints.
- Backdrivability
- A joint can be moved from the outside without the gearbox locking up. This makes robots safer and more sensitive when in contact with people.
- BLDC motor
- Brushless DC motor – the standard drive in robot joints, drones and robot vacuums.
- Boundary wire
- A cable laid in the ground that defines the mowing area of a classic robotic lawn mower.
- CE marking
- The manufacturer's declaration that a product meets EU requirements – a prerequisite for sale in the EU.
- Cobot
- A collaborative robot that can work alongside people without safety fencing, provided the risk assessment permits it.
- Collaborative operation
- Humans and robots work in the same workspace. ISO/TS 15066 specifies the force limits.
- Cycloidal drive
- An impact-resistant, compact gearbox with a high reduction ratio; partially backdrivable.
- Degrees of freedom (DoF)
- The number of independently movable axes. Manufacturers count inconsistently – with or without hands, per hand or for both hands. That's why you should always compare total DoF and hand DoF separately.
- Dexterous hand
- A skillful multi-fingered hand, usually with five fingers and 6 to more than 20 degrees of freedom, often with tactile sensors.
- Edge AI (on-device)
- The AI runs directly on the robot instead of in the cloud. This reduces latency and improves data privacy.
- Embodied AI
- Artificial intelligence with a body: AI that perceives through sensors and acts in the physical world through actuators.
- End effector
- The tool at the end of a robot arm – a hand, gripper, suction cup or screwdriver.
- F/T sensor
- A force/torque sensor that measures forces and torques along six axes, e.g., at the wrist.
- Fleet learning
- What one robot learns is passed on to the entire fleet via an update.
- Foundation model
- A large AI model pretrained on broad data that can be adapted to many tasks – in robotics, for example, NVIDIA GR00T or Physical Intelligence π0.
- Harmonic drive (strain wave gearing)
- A zero-backlash gearbox with a very high reduction ratio in a small space – widely used in arm and wrist joints.
- Hot-swappable battery
- A battery that can be replaced during operation or by the robot itself. It enables near-continuous operation.
- Humanoid
- A robot with a humanlike build: head, torso, two arms and usually two legs. Wheeled variants are called “wheeled humanoids.”
- Imitation learning
- The robot learns by imitating human demonstrations – via teleoperation, motion capture or video.
- IMU
- Inertial measurement unit: measures acceleration and angular velocity and helps the robot keep its balance.
- IP rating
- Protection against dust and water in accordance with IEC 60529. IP54 means protected against dust and splashing water; IP67 means dust-tight and protected against temporary immersion.
- LiDAR
- A laser scanner that measures the surroundings in three dimensions – standard in robot vacuums, delivery robots and legged robots.
- LLM
- Large language model: a language model for dialogue and task understanding, often acting in robots as a “translator” between humans and the control system.
- Mobile manipulator
- A robot arm on a mobile platform that fetches materials, sorts items or tends machines.
- NVIDIA Jetson Thor
- An edge AI computer based on the Blackwell architecture that is becoming the standard in many high-end humanoids in 2025/26.
- Open source
- Software or hardware designs are publicly available and may be used and modified – e.g., with pib, Poppy or InMoov.
- Payload
- The maximum weight a robot can lift or carry. Note: specifications per arm vs. total, and lifting vs. carrying, are often mixed up.
- Planetary gearbox
- A robust, efficient gearbox with a moderate reduction ratio; common in legged robots.
- Policy
- The learned control model that generates movements from sensor data.
- QDD (quasi-direct drive)
- A powerful motor with a low gear ratio: backdrivable, impact-resistant and sensitive – typical of modern legged and humanoid robots.
- RaaS (Robotics-as-a-Service)
- Robots are rented or subscribed to instead of purchased – with maintenance and updates included in the price.
- Reinforcement learning
- Learning through trial and reward, usually millions of times over in simulation.
- RGB-D camera
- A color camera with additional depth information, e.g., Intel RealSense.
- ROS 2
- Robot Operating System: open-source middleware and the standard in research, and increasingly in industry.
- RTK-GNSS
- Satellite navigation with correction data for centimeter-level accuracy – used by wire-free robotic lawn mowers.
- SDK
- Software development kit: tools and interfaces for writing your own programs for the robot.
- Sim-to-real
- Transferring skills trained in simulation to real robots.
- SLAM
- Simultaneous localization and mapping: the robot builds a map while simultaneously determining its position within it.
- Tactile sensing
- Touch and pressure sensors, usually in the fingertips, for sensitive grasping.
- Teleoperation
- Remote control by a human, for example using a VR headset. Also used to collect training data – and it is not autonomy.
- Tendon drive
- The motors sit in the body, and force is transmitted to the fingers or joints via cables. This makes limbs light and compliant.
- TOPS
- Trillion operations per second: a measure of a chip's AI computing power, e.g., NVIDIA Jetson.
- Torque density
- Torque per kilogram of drive mass – a key metric for dynamic robots.
- VLA model
- Vision-language-action model: translates camera images and voice commands directly into robot movements, e.g., Figure Helix or NVIDIA GR00T.
- VLM
- Vision-language model: understands images and language together.
- Whole-body control
- Coordinated control of all joints, including balance – a prerequisite for grasping while walking.
- World model
- An AI model that predicts how the environment will change as a result of an action – the basis for planning.