Glossary
Glossary
The terms DEPLOY uses with strict definitions. Where industry usage is loose or inflated, these reflect the registry's bar for what counts.
55 terms · canonical reference: registry.deploy.report/glossary ↗
Autonomous vehicles & robotaxi
AV stacks, robotaxi services, sensor modalities, regulatory framework.
Disengagement
An event in which a human takes control of an autonomous vehicle operating driverlessly. The California DMV publishes annual counts; one of the few comparable cross-company autonomy metrics.
Geofenced deployment
A deployment that operates only inside a defined geographic boundary, pre-mapped and validated. Most current robotaxi services are geofenced.
HD map
Centimeter-scale, lane-level map used as prior knowledge in AV perception and planning. HD-map vs map-less is a defining architectural choice.
Lidar
Light Detection and Ranging — laser-based 3D ranging sensor. Standard in most autonomous-vehicle stacks, notably absent from Tesla's vision-only approach.
Operational design domain
SAE J3016 term for the operating conditions an autonomous system is designed to handle. A vehicle outside its ODD is off-spec, not 'extending capability.'
Permit
Robotaxi context: regulatory authorization to operate a specific service type in a specific jurisdiction. Testing, driverless testing, and commercial passenger are escalating tiers.
Phantom braking
Unintended emergency-braking event with no real threat present. Indicates perception false positives. NHTSA-flagged in Tesla Autopilot reviews.
Remote assistance
Human-in-the-loop layer for autonomous vehicles in ambiguous situations. Distinct from full teleoperation and from a safety driver. Every commercial robotaxi service uses some form.
Robotaxi
A fully driverless commercial passenger service: a passenger pays for a ride and the vehicle operates with no human safety driver on board.
SAE levels of driving automation
Six levels (0–5) defined by SAE J3016 specifying which driving tasks the automation handles and under what conditions the human remains responsible. Cross-company SERP for 'level 4 autonomy.'
Vision-only
Autonomous-vehicle perception using cameras only — no lidar, no radar. Tesla's distinctive architectural choice since the 'Tesla Vision' transition.
Humanoid robots & hardware
Actuation, mechanics, end-effectors, and the mechanical architecture choices that shape what humanoids can do.
Backdrivability
Whether external force can move a joint without the motor commanding it — the dividing line between cobot-safe and traditional industrial arms.
Cobot
Collaborative robot. Designed to operate safely in shared space with humans without safety enclosures. Defined by ISO 10218 and ISO/TS 15066 power-and-force-limiting requirements.
Degrees of freedom
Count of independent axes a robot can articulate. The per-subsystem breakdown (legs · arms · hands) is more informative than the platform total.
Dexterous manipulation
Multi-fingered manipulation requiring fine motor control — in-hand reorientation, multi-finger force regulation, tool use. Distinct from parallel-gripper pick-and-place.
Direct-drive
Motor coupled directly to the joint output, no gear reduction. High backdrivability and bandwidth at the cost of torque density. MIT Cheetah is canonical.
End-effector
The terminal device on a robot arm that interacts with the world. In humanoids, typically a multi-fingered hand; in industry, a gripper, suction cup, or specialized tool.
Form factor
A robot's physical configuration: bipedal humanoid, wheeled general-purpose, quadruped, industrial arm, autonomous vehicle, or fixed-pedestal.
Gait cycle
The complete sequence of leg motions from one foot's contact with the ground to its next contact. Cycle frequency, duty factor, and step length together specify a gait.
Harmonic drive
A compact zero-backlash gear transmission with 50:1 to 320:1 ratios. Standard in humanoid joints. Two suppliers (Harmonic Drive SE and Nabtesco) own essentially the entire global market.
Humanoid robot
A general-purpose robot with a bipedal, human-scale form factor, designed to operate in environments built for humans without modification.
Hydraulic actuator
Robot joint driven by pressurized fluid. High peak force per kg but heavy support infrastructure; the dominant architecture in the original Atlas, retired in 2024.
Proprioception
A robot's internal awareness of its own body state via encoders, IMUs, and torque sensors. Distinct from exteroception, which reads the external world.
Series elastic actuator
A robot joint architecture with a spring placed in series between motor and output. Trades control bandwidth for backdrivability, impact tolerance, and direct torque sensing.
Tendon-driven
Actuation architecture where joints are driven by cables routed from remote motors. Standard for high-DOF hands; less common at major body joints.
AI, models & control
Foundation models, training paradigms, and the control stacks that turn learned policies into robot behavior.
Cross-embodiment
Training or transferring robot policies across different physical platforms. The central scaling claim of foundation-model-for-robotics work — with the same strict-vs-loose split as zero-shot.
Diffusion policy
Robot control architecture that generates actions by denoising from a noisy initial state. Default building block in modern VLA models.
Foundation model for robotics
A large multimodal AI model designed to generate robot behavior across multiple hardware platforms. Often abbreviated VLA (vision-language-action) or robot-FM.
Imitation learning
Training a robot policy from demonstrations rather than reward signals. Strict version is behavior cloning; looser versions iteratively correct for distribution drift.
Inverse kinematics
Computing joint angles needed to reach a desired end-effector pose. Largely subsumed by whole-body control in current humanoid stacks.
Loco-manipulation
Simultaneous locomotion and manipulation — picking something up while walking, carrying it across uneven terrain. The defining humanoid control problem.
Physical AI
AI systems designed to perceive, reason about, and act in the physical world. The defining property is embodied interaction: sensors, actuators, and a decision layer that connects them.
Reinforcement learning
Training paradigm where an agent learns by maximizing reward through trial-and-error. In robotics, mostly happens in simulation due to real-world data cost.
Sim-to-real transfer
Training a robot control policy in simulation and deploying it on physical hardware. The strict version is zero-shot transfer with no real-world fine-tuning.
Teleoperation
Full remote control of a robot by a human operator. The dominant data-collection method for imitation-learning datasets. Distinct from remote assistance.
URDF
Unified Robot Description Format — the XML schema describing a robot's kinematic and dynamic structure. The de-facto standard across simulators, motion planners, and control software.
Vision-language-action model
A neural architecture that ingests visual observations + natural-language instructions and emits robot actions in a single forward pass. Often abbreviated VLA.
Whole-body control
A control framework that solves for joint torques across all DOF simultaneously against multiple competing objectives — balance, end-effector position, joint limits — as a single optimization at each timestep.
World model
Learned predictive model of an environment that can simulate consequences of hypothetical actions. Central to model-based RL and 2024–2025 research efforts.
Zero-shot generalization
A policy performing tasks it was never trained on. Strict version: no gradient updates on the target task. Looser versions involve few-shot fine-tuning or prompt-conditioning.
Deployment & operations
Deployment classifications, operator relationships, safety bars, and the editorial standards Deploy applies to coverage.
Deployment
A robot doing real economic work at a named operator site, with a published source confirming the operation. Not a shipment, sale, demo, or MOU.
Fleet
Count of active vehicles or robots a single operator currently has performing work. Not installed base, manufacturing run, or shipment volume.
Greenfield deployment
Robot deployment at a new facility designed from inception to accommodate it. Distinct from brownfield (retrofitted into existing sites). Greenfield is a weaker capability claim.
Incident
A verified event in which a deployed robot or autonomous vehicle caused harm, behaved abnormally with public reporting, or was subject to regulatory action.
Maturity stages
Deploy's five-step ladder for classifying robot models: research, prototype, pilot, commercial, production. Each stage is a higher trust signal than the one before it.
Operator
The entity that runs a robot or vehicle in the field. May or may not be the same as the manufacturer.
Verified
A claim about a robot, company, deployment, or incident confirmed against primary sources at the registry's methodology bar. Distinct from credible claims.
Manufacturing, business & strategy
Cost engineering, supply chain, business models, and the strategic framings the registry uses.
Bill of materials
Structured list of components, parts, and assemblies in a finished product with unit costs. The single most-cited humanoid cost-down metric — routinely conflated with delivered unit cost.
Brain company vs. hardware company
How Deploy classifies physical-AI companies: hardware (the robot), brain (the AI foundation models and software), or both.
First-pass yield
Share of units coming off a manufacturing line meeting full specification without rework. The threshold separating prototype-with-rework from real manufacturing.
Robot-as-a-Service
Business model where customers pay a recurring fee for robot capacity rather than purchasing. RaaS is the dominant proposed humanoid pricing model.
Throughput
Robotic-manufacturing rate — units per hour, per day, per shift. Distinct from cycle time, production volume, and shipment volume. Routinely conflated in humanoid coverage.
Tier 1 supplier
Direct-to-OEM automotive component supplier — the layer immediately upstream of vehicle assembly. Bosch, Magna, Continental are canonical. Now positioning into humanoids.
Unit economics
Revenue and cost profile of a single unit of operation at the margin. In robotaxi: one vehicle's daily margin against allocated costs.
Vertical integration
Owning multiple layers of the value chain — AI stack, hardware, manufacturing, deployment operations. The dominant strategic posture across physical AI today.