The robotics industry is developing rapidly, and this comes with changes in terminology that outpace standards. Here we propose a simple classification for contemporary robotic systems, along with corresponding definitions.
At the highest level, this classification separates robots along two independent axes: “Form Factor” and “Level of Autonomy.” The form factor can be thought of as the body of the robot, while the level of autonomy can be thought of as the brain. Most form factors can be paired with most levels of autonomy.
We exclude autonomous vehicles, unmanned aerial vehicles, and any system capable of propelling itself in the water from our definition of “robot.” We expect these systems to be regulated separately.
Form Factors
Robots can be classified into two non-mutually exclusive categories: manipulation and mobility. A robot that fits both categories is a “mobile manipulator.”
Manipulation
A manipulation robot is a robot designed to position one or more end effectors. An end effector is the tool that directly contacts or acts on the workpiece (think grippers, welding torches, screwdrivers, paint sprayer, etc.). The same manipulator can perform a variety of different jobs depending on which end effector is attached. Types of manipulators include:
- Articulated: a single chain of joints, with each joint carrying the load of all subsequent joints in the chain. The most common and versatile manipulation architecture.
- Parallel: multiple arms working together to move a single end effector. Each arm is typically driven by a single motor mounted to a fixed frame, with the moving parts consisting of lightweight passive linkages.
- Cartesian: linear motion along perpendicular rails. Work area can be made arbitrarily large by extending the rails. Cartesian robots encompass most CNC machines, most 3D printers, and gantry systems.
Mobility
Robots designed to move through an environment. Types of mobile bases include:
- Bipedal: Two-legged locomotion. Inherently unstable, requiring continuous active balancing. Enables navigation of environments built for humans.
- Quadrupedal: We use this category to refer to all legged architectures with more than two legs, though four legs is standard. More stable than bipedal. Can traverse rough, unstructured terrain.
- Wheeled: Locomotion via wheels or tracks on a mobile platform. The simplest and most energy-efficient locomotion architecture. Limited to relatively flat surfaces. Dominant in warehouse and logistics applications.
Mobile Manipulation
Combines one of the mobile bases above with one or more manipulators (nearly always articulated). Examples include:
- Humanoids: bipedal base + two articulated arms
- Quadruped with arm: quadrupedal base + one articulated arm
- Wheeled mobile manipulator: wheeled base + 1–2 articulated arms
Levels of Autonomy
A robot’s behavior is governed by its control system: the software and firmware that determines how it moves and acts. High autonomy control systems are often referred to as “models” or “policies.” The same physical robot running a different control system can operate at a different level of autonomy.
Robot control systems can be broadly classified into three categories based on level of autonomy. These levels require a basic understanding of the software/firmware stack:
- A motion is a single physical movement of the robot. Example: repositioning joints to shift the angle of an end effector.
- An action is a sequence of motions that accomplishes something. Example: the motions of reaching, closing a gripper, and lifting combine to form the action “pick up.”
- A task is a sequence of actions that achieves a goal. Example: the actions of opening a box, picking items, carrying them, and placing them on a shelf together form the task “stock the shelf.”
The three layers of the motion control stack correspond to these three concepts:
- Motion control: The execution of individual motions, regulating how the robot physically carries them out.
- Motion planning: Determining how to decompose actions into motions.
- Action planning: Determining how to decompose tasks into actions.
Autonomy is a property of the control system, not the hardware. The same robot running different software operates at different levels of autonomy. This gives drafters three distinct ways to scope a bill:
- The deployed combination: “a robot operating under a high-autonomy control system.”
- The hardware capability: “a robot capable of running a high-autonomy control system.”
- The software alone: “a high-autonomy robotic control system.”
A control system’s autonomy can be classified into three major levels:
Non-Autonomous: Motion control is performed by the control system, but motion planning and action planning are determined entirely by a human. This can either be through pre-programmed instructions or real-time remote commands.
Low-Autonomy: The control system autonomously performs motion planning, adapting how it executes actions in real time using sensor input. Action planning is still determined by a human. All possible action sequences are pre-specified; the system selects among and adapts within them. Examples: pick-and-place with machine vision, wheeled warehouse robots (AMRs) that maneuver around obstacles.
High-Autonomy: The control system autonomously performs action planning, generating action sequences to accomplish its goal. Example: a robot told “load the dishwasher” that figures out which items to pick up and where to place them in the rack.
Other Common Robot Classifications
These terms often come up but don’t warrant their own definitions (because they’re either too broad or too specific). With the exception of CNC machines, we caution against regulation that targets robots according to these classifications.
Industrial robot: At its most specific, an articulated arm on a fixed base, with little to no sensing or AI integration. Often used more broadly: sometimes synonymous with “articulated,” other times referring to any robot on a fixed base with little to no AI integration. There is no agreed-upon industry definition.
Cobot (collaborative robot): An articulated arm with safety features that allow it to operate near humans. These tend to be lower-payload and have a greater degree of autonomy than standard industrial robots.
SCARA robot: A type of articulated robot in which all rotary joint axes are vertical and parallel to each other, with least one linear joint for vertical motion. Cheaper, smaller, typically used for electronics assembly.
Delta robot: A type of parallel robot that consists of three arms. Primarily used for high-speed, low-payload tasks in a small area.
AMR (autonomous mobile robot): A low-autonomy wheeled robot without manipulation capabilities, typically used for material transport in warehouses and logistics.
Humanoid: A mobile manipulation robot with a bipedal locomotion base and two articulated arms. Occasionally also used to refer to two-armed manipulators with a wheeled-base (e.g. Sunday and Dexmate). Usually developed by full-stack companies alongside high-autonomy control systems.
CNC machine: A cartesian robot used for subtractive manufacturing (milling, drilling, cutting) or additive manufacturing (3D printing). Legislation that does not intend to cover CNC machines and similar equipment should scope to more specific definitions (e.g., “articulated robot” rather than “robot”).
Service robot: An ISO classification (ISO 8373:2021) defined as a robot “in personal use or professional use that performs useful tasks for humans or equipment.” This is an overly broad definition that is poorly suited for legislation that needs to target systems by their capabilities.
Key Components
Actuators
Discussions of robotic components frequently use the terms motors, actuators, gearboxes, reducers, drives, joints, and screws interchangeably, which is a common source of confusion. In supply-chain discussions, “Actuator” typically refers to rotary actuators: the complete joint modules present in all robotic systems with articulated manipulation or legged mobility.
A rotary actuator consists of:
- A servo motor (receives power and spins at high speeds, includes an encoder that senses the joint’s position).
- A reducer (turns speed from the motor into torque at a fixed ratio).
- An output bearing (supports the output flange, allowing it to rotate under the loads imposed by the arm segment).
Cobot joints often include an additional encoder and occasionally a torque sensor for faster collision detection. Many rotary actuators also include a brake, which closes when power is cut to hold the joint in place.
The reducer itself goes by many names, some which are architecture-specific and some which are architecture-agnostic. These include gearbox, harmonic drive, cycloidal drive, and others. As a heuristic, anything with “gear” or “reducer” in the name is a reducer, though not everything with “drive” in the name is.
Each servo motor is powered by a corresponding servo drive. The servo drive is the lowest-level motion control computer. It delivers precise electrical currents to the motor according to motion control commands issued by the higher-level controller.
Sensors
Sensors are a broad category encompassing many components with different purposes. Types of sensors include:
- Position encoders: These are found in actuators and typically lie directly on the servo motor. They sense the current position of a joint.
- Vision sensors (cameras): These are typically found in robots with some degree of autonomy. Used to identify objects (to interact with) and obstacles (to avoid) in the environment.
- Proximity/distance sensors: LiDAR is the most common example, but alternatives include ultrasonic and infrared sensors. Depth cameras sense both vision and distance.
- Torque/force sensors: These are also found in actuators. They measure the load on joints.
- IMUs: These combine accelerometers and gyroscopes to assist in balance. Found in most legged robots.
- Touch sensors: Experimental sensors for measuring texture and slip. Rarely used outside of high-end end effectors.
- Other sensors: Sensors which are not typically used as continual input for the robot’s control system. Includes microphones, thermometers, and voltage sensors for measuring battery life.
Controller
The controller is the computer that runs the control system’s software.
Non-autonomous robots (e.g. traditional industrial robots, CNC machines, and certain cobots) typically use proprietary controllers programmed in vendor-specific languages. Their closed architectures make third-party software integration difficult.
Low-autonomy robots typically add a small GPU which runs a computer vision model to process camera input. In these architectures, the proprietary controller still handles motion control.
High-autonomy systems require a fundamentally different controller: GPU-based compute hardware capable of running AI inference onboard. Action planning and motion control often run as separate models on the same hardware. While onboard compute is the dominant standard for today’s high-autonomy robots, only the motion control layer strictly requires it. Action planning is more latency-tolerant, making it theoretically viable to run on cloud infrastructure instead.
Definitions
These are definitions that we recommend incorporating into legislation and that are designed to be combined modularly. Legislation can scope its coverage by selecting one or more definitions from the Hardware section and combining with one or more definitions from the Control System section.
Hardware Definitions
Definitions largely track ISO 8373, with some modifications for simplicity and ease of use.
(1) ROBOT.— The term “robot” means a programmable, actuated mechanical system designed to perform manipulation, locomotion, or both, and that locomotes, if at all, only by means of contact with the ground. The term does not include a motor vehicle as defined in section 30102 of title 49, United States Code.
(2) MANIPULATOR.— The term “manipulator” means a mechanical structure composed of one or more segments connected by joints, that positions an end effector in order to interact with objects in a robot’s environment. The term does not include the end effector itself.
(3) END EFFECTOR.— The term “end effector” means a device attached to a manipulator that directly contacts or acts upon a workpiece.
(4) MANIPULATION ROBOT.— The term “manipulation robot” means a robot designed to position one or more end effectors in order to interact with objects in its environment.
(5) ARTICULATED ROBOT.— The term “articulated robot” means a manipulation robot in which each manipulator consists of a single open chain of rigid segments connected in series by joints, in which at least two of those joints are rotary.
(6) PARALLEL ROBOT.— The term “parallel robot” means a manipulation robot in which two or more arms or linkages connect a common base to a single end effector or moving platform, forming a closed-loop mechanical structure.
(7) CARTESIAN ROBOT.— The term “cartesian robot” means a manipulation robot that positions its end effector by moving along two or more mutually perpendicular linear axes.
(8) MOBILE ROBOT.— The term “mobile robot” means a robot capable of traversing its environment under its own power.
(9) LEG.— The term “leg” means a mechanism of interconnected links and joints that is actuated to support and propel a mobile robot through intermittent contact with the travel surface.
(10) LEGGED ROBOT.— The term “legged robot” means a mobile robot that traverses its environment by means of one or more legs.
(11) BIPEDAL ROBOT.— The term “bipedal robot” means a legged robot that traverses its environment by means of two legs.
(12) QUADRUPEDAL ROBOT.— The term “quadrupedal robot” means a legged robot that traverses its environment by means of more than two legs.
(13) WHEELED ROBOT.— The term “wheeled robot” means a mobile robot that traverses its environment by means of wheels, tracks, or similar continuous-contact mechanisms.
(14) MOBILE MANIPULATION ROBOT.— The term “mobile manipulation robot” means a robot that is both a manipulation robot and a mobile robot.
Control System Definitions
Unlike the hardware definitions, the following are not drawn from ISO 8373, which does not classify robots by level of autonomy. They track the Levels of Autonomy described above and can be freely combined with one or more of the hardware definitions.
(1) CONTROL SYSTEM.—The term “control system” means the software and firmware that govern a robot’s behavior. For purposes of the definitions in this section—
(A) the term “motion” means a single physical movement of a robot;
(B) the term “action” means a sequence of motions that together accomplish a discrete operation, such as picking up an object;
(C) the term “task” means a sequence of actions that together accomplish a goal, such as stocking a shelf;
(D) the term “motion control” means the regulation of how a robot physically executes a motion, such that sensor input may affect the quality of execution but does not alter which motions the robot performs;
(E) the term “motion planning” means the decomposition of an action into the individual motions required to carry it out, including the real-time adaptation of those motions to sensor input; and
(F) the term “action planning” means the decomposition of a task into a sequence of actions, including the determination of which objects the robot interacts with and in what order.
(2) NON-AUTONOMOUS CONTROL SYSTEM.—The term “non-autonomous control system” means a control system that performs motion control, but in which both motion planning and action planning are performed by a human. A non-autonomous control system may include—
(A) a scripted system, in which the robot’s motions are fully pre-determined by a human programmer; or
(B) a teleoperated system, in which the robot’s motions are directed in real time by a human operator issuing commands from a location other than on the robot itself, whether through full teleoperation or through shared control in which the control system performs motion control while the human directs motion planning.
(3) LOW-AUTONOMY CONTROL SYSTEM.—The term “low-autonomy control system” means a control system that autonomously performs motion control and motion planning, adapting the robot’s motions to sensor input in real time, but in which action planning is performed by a human or system that selects among pre-specified action sequences. The robot’s possible action sequences are pre-specified; the control system selects among and adapts within those sequences, but does not generate novel action sequences.
(4) HIGH-AUTONOMY CONTROL SYSTEM.—The term “high-autonomy control system” means a control system that autonomously performs action planning, generating sequences of actions that were not individually pre-specified by a human and autonomously determining which objects in the robot’s environment to interact with.
