Glossary
Humanoid robots in plain words
Every technical term about humanoid robots, explained in a sentence or two — no jargon. Because you shouldn't need an engineering degree to understand what you're buying.
101 terms explained
Use
3C industryIndustrial sector encompassing Computing, Communication and Consumer Electronics. A common term in the Chinese manufacturing context, used to describe production lines for tablets, smartphones and digital accessories. It was in this sector that the first deployments of humanoids in real production took place in early 2026. EN: "3C industry".Automotive Tier-1 supplierA tier-1 supplier to the automotive industry, delivering complete sub-systems (motors, actuators, gearboxes) directly to the manufacturers (OEMs). In the context of humanoid robots, the term refers to players such as Hyundai Mobis, Bosch or Schaeffler, who are moving down the robotics chain via the component they have already mastered: the actuator.Battery lifeHow long the robot runs on a charge.Central enterprise (央企, yangqi)State-owned enterprise supervised directly by SASAC (State-owned Assets Supervision and Administration Commission). In China, central enterprises are the designated first deployers in state plans, such as the 2026 MIIT-SASAC directive on humanoids. Not to be confused with provincial state-owned enterprises or domestic private companies.CobotA robot designed to work safely alongside humans.Companion robotA humanoid robot designed not to carry out productive tasks but to keep a user company: conversation, physical presence, reading emotions. A commercial category distinct from the industrial or service humanoid robot.Contracting authorityEntity (client, industrial buyer) that defines the specifications its suppliers must comply with and places orders accordingly. In the robotics industry, a major contracting authority (e.g. BMW, Amazon) can impose its own evaluation and qualification criteria on robot manufacturers, either in place of or in addition to public standards.Cycle timeThe time a robot (or an operator) needs to complete a single repetitive operation on a production line. A core metric of industrial productivity: a shorter cycle time at constant quality means a higher throughput. It complements the success rate and continuous walking time when assessing a robot in real-world conditions. FR: "temps de cycle".Deployment quotaA volume of units or use cases set by administrative directive, which the designated players (provinces, state-owned enterprises) must reach within an imposed timetable. Different from a market forecast: the quota is a binding commitment accompanied by delivery dates and mandatory categories. A typical instrument of Chinese industrial policy (e.g. directive 工信厅联科函〔2026〕256号: 10,000 humanoids before the end of 2026).Design for Excellence (DfX)A methodological framework for industrial design that simultaneously optimises a product's manufacturability, reliability, maintainability and cost from the design phase onwards; applied by Bosch to Humanoid.ai's HMND 01 robot to guarantee volume production.Fallback systemContractually agreed residual human intervention when the robot's autonomy falls short.First-pass yield (first-pass yield rate)In industrial production, the proportion of units that pass every end-of-line inspection test without rework or repair. A key indicator of manufacturing process maturity; an FPY above 80% is regarded as a significant industrial threshold for humanoid robots.Humanoid robotA human-shaped robot: two arms, two legs, a head.Manufacturing readinessThe maturity of a robotic system for integration into industrial production; does not guarantee operational deployment.Mass productionIndustrial phase in which a product is manufactured on a large scale using standardised processes. In the humanoid robot sector, 2025 was designated by the MIIT as the first year of mass production for the industry. English equivalent: mass production.MES (Manufacturing Execution System)Industrial software for real-time control of manufacturing operations: batch tracking, quality control, component traceability, flow optimisation. Distinct from ERP (planning) and SCADA (equipment supervision). An in-house developed MES is a marker of vertical integration.PayloadThe maximum weight the robot can carry.Phone-home (beaconing)The behaviour of a device (robot, sensor, software) that automatically sends data to its manufacturer or to a remote server without the operator's knowledge. In English: phone-home or beaconing. Cited as a national security risk in debates about Chinese humanoid robots deployed in countries allied to the United States.Rare earthsA group of 17 metallic elements (including neodymium, dysprosium and praseodymium) that are essential to the permanent magnets used in electric actuators and to batteries. China accounts for the bulk of global extraction and processing, making this a point of strategic vulnerability for humanoid robotics supply chains. EN: rare earth elements / rare earths.Scaling up (scale-up)The process by which a robotics technology moves from demonstration to volume industrial deployment.Shipments (expéditions)The number of units physically delivered to a customer or deployed on site over a given period. This is the benchmark metric used by analyst firms (TrendForce, IDC) to gauge market volume for robots; it should be distinguished from orders and from unverified deployment announcements. FR: expéditions.TeleoperationControlling the robot remotely, like a puppet.
Technical
ActuatorThe motorised "muscle" that moves a joint.Affective AIOn-board artificial intelligence module that analyses facial expressions, tone of voice and speech patterns to assess the user's emotional state and adapt the robot's behaviour accordingly. Distinct from object manipulation or navigation AI.Autonomous navigationA robot's ability to plan and execute its own trajectories without real-time human intervention, as opposed to teleoperation. In humanoid competition (e.g. the Beijing 2026 marathon), the autonomous/remote-controlled distinction determines the official ranking via a penalty coefficient. EN: «high>ªautonomous navigation».BipedWalks on two legs, like a human.Cluster controlSimultaneous control of several robots from a single platform.Cross-embodimentThe ability of an AI system to control robots with different morphologies (wheels, legs, arms) without being designed for a single chassis. The data collected by one robot enriches the model used by another. EN: cross-embodiment.Data Factory (Usine à données)Infrastructure dedicated to the large-scale production of training data for robots, whether physical (a human motion capture centre) or software-based (a synthetic generation pipeline using simulation). The term gained currency with NVIDIA's Physical AI Data Factory Blueprint (March 2026) and the TUM RoboGym. It signals a paradigm shift: data is no longer a by-product of deployment, it is manufactured deliberately. EN: data factory, training data factory.Dexterous manipulationA robot's ability to perform fine, precise movements with its hands (gripping, assembly, handling fragile objects), as opposed to simple locomotion. Measured in particular by the number of degrees of freedom concentrated in the hand. FR: manipulation dextre.Digital twinSynchronised virtual replica of a physical system (robot, environment, factory), used for simulation, testing and AI training without any risk to the real hardware. In English: digital twin. A common term in sim-to-real work and production optimisation.DOF (degrees of freedom)The number of independent joints a robot has.Dynamic balanceThe ability of a mechanical or robotic system to maintain its stability while in motion, through continuous adjustments (as opposed to static balance). A fundamental prerequisite for autonomous bipedal walking; inherited from techniques developed for self-balancing exoskeletons. EN: dynamic balance / dynamic stability.Embodied AIAn artificial intelligence tied to a physical body (a robot) that perceives and acts in the real world.End-to-end neural networkMachine learning architecture in which a single neural network directly processes raw inputs (sensors, cameras) and produces action outputs, without a pipeline of hand-written sub-modules. It does away with bespoke, hand-crafted logic in favour of a general-purpose model trained on data. In English: *end-to-end neural network*.ExoskeletonMotorised frame worn by a human to amplify strength or assist mobility. In humanoid robotics, manufacturers with a background in exoskeletons (dynamic balance, posture control) are transferring that expertise to autonomous bipedal robots. EN: exoskeleton.Firmware / OTA updateThe robot's internal software, updated remotely.Foundation modelA versatile pre-trained AI model that can be adapted to a wide range of tasks or robot morphologies.Gripper / handThe hand or claw at the end of the arm that grasps objects.HEIS (Humanoid Embodied Intelligence Standardization)Chinese standardisation framework for humanoid robots and embodied intelligence, led by the MIIT. The first HEIS reference framework (2026) covers 52 standards across the entire humanoid value chain: components, intelligence, applications, safety.Human in the loopThe presence of a human supervisor in the decision-making chain of a system presented as autonomous.Imitation learning (apprentissage par imitation)The robot learns by imitating human movements captured through teleoperation.Large language model (LLM)An AI model trained on vast amounts of text, able to understand and generate language — and, increasingly, to drive robots.LiDARA 'laser radar' that lets the robot see in 3D.Locomotion platformHumanoid robot designed primarily for walking, balance and load carrying, without advanced dexterous manipulation capability. Dominant metric: number of units delivered, endurance, payload. Stands in contrast to dexterous manipulation as a unit of account. EN: locomotion platform.Mecha (mecha robot)Cockpit robot piloted by a human on board.Motion capture (mocap)A technique for recording the movements of a human body (or an object) using inertial, optical or mechanical sensors, in order to produce trajectory and pose data. Used to generate real-world physical training data: a human trainer performs gestures that are captured and then used to train the robot by imitation. Example systems: Xsens (Hexagon), MANUS gloves. EN: motion capture, mocap.Onboard AIThe artificial intelligence running inside the robot itself.Perception stackThe set of hardware and software components that allow a robot to perceive its environment: cameras, depth sensors (LiDAR, etc.) and the associated processing. In 2025-2026, this item alone accounts for around 20,000 USD on a humanoid robot — the equivalent of the target cost of the complete machine — making it the main obstacle to economic viability.Performance benchmarkStandardised test framework for assessing a robot's capabilities in a neutral and comparable way.Physical AIA strategic paradigm aimed at unifying autonomous driving and humanoid robotics on a single software platform, exploiting the perception and decision-making synergies between the two fields. Term popularised by Mobileye/Amnon Shashua in 2026. EN: Physical AI.Physics engine (moteur physique)Software that simulates the laws of physics (contacts, gravity, friction, rigid-body dynamics) to enable robots to be trained in a virtual environment. Critical to the validity of sim-to-real transfer: a faithful physics engine narrows the gap between simulated behaviour and real-world behaviour. Examples: Newton 1.0 (NVIDIA/Linux Foundation), Kamino (Disney Research). EN: physics engine, physics simulator.Pre-loaded movementA gesture sequence recorded in advance that the robot replays identically, without deciding it itself.QuadrupedA four-legged robot, dog-style (e.g. Boston Dynamics Spot).Real-world physical training dataData derived from concrete interactions between a robot and the physical world (movements, forces, trajectories), as opposed to synthetic data generated by simulation. Considered the main bottleneck in embodied robotics: expensive to capture, it determines the quality and generalisation of action models. EN: real-world physical training data.Real2Sim2RealRobot training pipeline: capture data from the real world, transfer it into a simulator for augmentation and training, then reimport the trained model onto the physical robot. Reduces the need for teleoperation and direct human demonstrations. EN: Real2Sim2Real.Reinforcement learning (RL)A training method in which an agent learns by interacting with an environment (real or simulated) and maximising a reward signal. In robotics, it is used extensively in simulation (e.g. NVIDIA Isaac Lab) to produce control policies, which are then transferred to the physical robot (see: sim-to-real). Distinct from imitation learning: the agent does not need human demonstrations, but does require a well-defined reward function. EN: reinforcement learning (RL).Robotics foundation modelA general-purpose pre-trained model capable of driving different robotic morphologies without retraining.RolloutComplete sequence of actions carried out by a robotic model; its duration measures decision latency.SDKThe toolkit for programming the robot yourself.Sensor harnessA device worn by a human operator fitted with sensors (vision, IMU, force/tactile sensors) to record trajectories and interactions in real-world conditions, without a physical robot. The data produced in this way is used to train embodied AI behaviour models. EN: sensor harness / motion-capture suit for embodied AI., short_definition: Sensor suit worn by a human to generate robot training data.Sim-to-real (simulation to reality)Training a robot in simulation, then transferring the skills to the real world.Tactile sensorA device that measures force, pressure or slippage on contact with a surface. In humanoid robotics, it allows the robot's hand to modulate its grip according to the fragility or mass of the object. Recent implementations (e.g. e-Flesh from Assured Robot Intelligence) combine 3D-printable microstructures, magnets and magnetometers.TFLOPSA unit of computing power: one trillion floating-point operations per second.Training dataThe examples (movements, images, measurements) a robot learns a task from.Training stackThe complete software suite used to train a robot model: simulation platform (e.g. Nvidia Omniverse/Isaac), data pipeline, foundation model (VLA or otherwise), sim-to-real loop. The most upstream layer in the production chain of a robot behaviour; a dependency on a third-party stack (e.g. Nvidia) is structural, not incidental.VLA (Vision-Language-Action)AI model that translates images and language into motor commands for a robot.World modelAI model that predicts how the physical world will evolve in order to guide robotic action.World model (modèle du monde)A predictive model of world physics, used to learn behaviours without intensive teleoperation.
Buying & cost
A-shareA security listed in renminbi (yuan) on the Shanghai or Shenzhen stock exchanges (the Chinese A-share market). Historically reserved for Chinese institutional investors, the A-share market remains distinct from H-shares (Hong Kong) and American ADRs. The STAR Market (科创板) is a segment of it focused on technology and innovation.Average selling price (ASP)Ratio of revenue to the number of units sold over a given period. A key indicator in hardware for measuring the trajectory towards commoditisation or upmarket positioning. A sustained fall in ASP points either to a price war or to a deliberate volume strategy (cost reduction through economies of scale). English acronym: ASP (Average Selling Price).BOM (Bill of Materials)Exhaustive list of the components and materials required to manufacture a product, together with their unit costs. For a humanoid robot, the BOM details each sub-assembly (actuators, sensors, on-board computing, chassis) and makes it possible to estimate the cost of series production. EN: Bill of Materials (BOM).Capitalisation tableA document setting out a company's capital structure: shareholders, percentage of ownership, type of securities (ordinary shares, preference shares, options). Also known as a cap table. It makes it possible to identify who holds financial interests in a company, including institutional investors, funds and strategic partners.CE markingThe maker's declaration that the product meets EU rules.CommoditisationThe process by which a product loses its differentiation and competes on price alone.Contract manufacturerIndustrial firm producing the robot on behalf of the designer (contract manufacturer).Customer acquisition costWhat a company spends to win a customer — sometimes selling at a loss to capture something else, like data.Customs dutiesA tax on entry into the EU, based on country of origin.DecacornPrivately held startup valued at more than USD 10 billion. An extension of the terms "unicorn" (USD 1 billion) and "hectocorn" (USD 100 billion). In English: *decacorn*.Decoupling (découplage)A strategy aimed at reducing economic and technological interdependence between two countries or geographical areas, particularly between the United States and China. In robotics, it takes the form of bringing production back home, diversifying component suppliers and imposing restrictions on the export of critical technologies. Not to be confused with offshoring (moving to a third country) or reshoring (complete repatriation). FR: découplage.DepreciationSpreading a purchase cost over several accounting years.Dual-class sharesCapital structure in which certain shares (typically the founder's) carry a number of voting rights greater than their weight in the capital. This allows the founder to retain operational control after an initial public offering or a fundraising round. English equivalent: dual-class shares or supervoting shares.HS code (customs)The code that classifies a product for customs.Influencer marketingA promotional strategy in which a brand pays a third party (here, a robot) to associate its image with a product or service in front of a digital audience. Refers to the commercial model adopted by certain humanoid robots deployed as brand ambassadors rather than as production tools.IPO (initial public offering)Initial Public Offering: the first listing of a company's shares on a stock market. It allows pre-existing shareholders to realise their capital gains. In robotics, the IPO is often planned as early as the previous funding round, with financial investors buying into the company beforehand in order to benefit from the valuation at issue.LeasingRenting the robot over 2-3 years, with a buyout option at the end.Preferred supplier (fournisseur préféré)Contractual status granted to a supplier guaranteeing it priority of allocation over a defined volume of purchases, often accompanied by a multi-year commitment. In the humanoid actuator supply chain, this is a key industrial chokepoint: Schaeffler has been named preferred supplier to Humanoid.ai for more than 50% of its actuators through to 2031.Production capacity (units/year)Maximum volume of units a manufacturer can assemble over a given period, expressed in units per year. In the robotics press, often referred to by the English shorthand *capacity* or by analogy with the steel industry term "tonnage". Used to assess a humanoid robot manufacturer's move to industrial scale.RaaS (Robot as a Service)Renting a robot by subscription, services included.Series AThe first significant round of institutional funding after the seed or pre-seed stage, generally led by venture capital firms. In humanoid robotics, Series A rounds reach unusually high amounts (several hundred million dollars) because of hardware costs and time-to-market delays.Series BVenture capital funding stage following Series A, generally reserved for startups that have validated their business model and are seeking to fund their industrial or commercial scale-up. In robotics, Series B rounds regularly reach several tens or hundreds of millions of dollars.Series CThe third major round of institutional funding for a startup, typically at a valuation ranging from several hundred million to several billion USD. It follows Series A (late seed) and Series B (growth). In English: *Series C*.Sovereign wealth fundA public investment vehicle managed by a state or a public entity, which deploys capital in strategic sectors. In humanoid robotics, their entry into a manufacturer's capital signals structural state backing, well beyond mere financing. E.g.: Big Fund III (National AI Industry Fund) moved into embodied AI in March 2026 via Galbot.StablecoinA cryptocurrency whose value is pegged to a reference asset (dollar, gold, basket of assets) in order to minimise volatility. Used as both a payment tool and a store of value within the crypto ecosystem. In English: *stablecoin*.StandardisationThe process of developing and publishing enforceable technical standards, defining the minimum requirements (safety, performance, interoperability) for a category of products or systems. In robotics, standardisation sets the regulatory framework that manufacturers must comply with. Not to be confused with certification (verification of compliance with a standard) or regulation (the applicable law).State subsidyDirect or indirect financial support granted by a government to strategic companies or sectors. In humanoid robotics, this refers to Chinese schemes (national AI-robotics funds, provincial governments) and American mechanisms (IRA, CHIPS Act) that are accelerating deployments. Not to be confused with leasing and RaaS, which are private mechanisms. EN: *state subsidy* / *government grant*.Supply chainThe full set of players, flows and processes involved in designing, manufacturing and delivering a product, from raw materials through to the end user. In humanoid robotics, it refers in particular to the dependence on Chinese components (actuators, rare earths, PCBs) and to strategies for geographical diversification (near-shoring, friend-shoring). EN: *supply chain*.TAM (total addressable market)Total Addressable Market: an estimate of the maximum revenue a product or service could generate if the entire potential market were captured. Used by analysts and investors to calibrate long-term projections for the robotics sector. EN: Total Addressable Market (TAM).Value chainSequence of value-creating stages from design through to deployment.VATA 20 % tax on the price, recoverable by a business.Vertical integrationA strategy whereby an economic player controls several links in the same value chain — production, distribution, marketing. In the humanoid robot ecosystem, it refers in particular to a broadcaster, a fund or a manufacturer simultaneously holding stakes in the makers it promotes or works with.