Types of Robots
The Families of Robots
Robots come in families defined by the job they do. Learn the main ways of classifying them: by mobility, by application, by autonomy. Plus why shape follows task.
beginner · 12 min read
Lesson content
There is no universal robot. That is not a failure of imagination. It is physics.
A machine built to carry 300 kg cannot also fit through a doorway. A machine built to fly cannot also push against something heavy.
So robots come in families. Each family is a bundle of design choices that only make sense together and picking one choice quietly commits you to the rest.
The most useful first cut is does the base move?, because that one choice determines almost everything else about the machine.
| Fixed base (manipulators) | Mobile base | |
|---|---|---|
| What moves | The tool, within a work envelope | The whole robot, through an environment |
| Knows where it is | Exactly, from joint encoders and geometry | Approximately, since it must localise and can be wrong |
| Power | Mains cable; effectively unlimited | Batteries; the dominant design constraint |
| Main problem | Reaching a pose accurately | Knowing where it is and not hitting anything |
| Payload | Up to hundreds of kg | Limited by traction and by battery mass |
| Examples | 6-axis arm, SCARA, delta, cobot | AMR, drone, quadruped, rover |
What moves
- Fixed base (manipulators)
- The tool, within a work envelope
- Mobile base
- The whole robot, through an environment
Knows where it is
- Fixed base (manipulators)
- Exactly, from joint encoders and geometry
- Mobile base
- Approximately, since it must localise and can be wrong
Power
- Fixed base (manipulators)
- Mains cable; effectively unlimited
- Mobile base
- Batteries; the dominant design constraint
Main problem
- Fixed base (manipulators)
- Reaching a pose accurately
- Mobile base
- Knowing where it is and not hitting anything
Payload
- Fixed base (manipulators)
- Up to hundreds of kg
- Mobile base
- Limited by traction and by battery mass
Examples
- Fixed base (manipulators)
- 6-axis arm, SCARA, delta, cobot
- Mobile base
- AMR, drone, quadruped, rover
That split explains something that surprises almost everyone at first.
An industrial arm can place a part to within a tenth of a millimetre. A mobile robot the size of a fridge may not know where it is to within ten centimetres.
Why such a gap? The arm is bolted to the floor and measures every joint directly, so its geometry is a known fact. The mobile robot has to infer where it is, from noisy sensors and wheels that slip.
- Manipulators
Arms: a chain of powered joints ending in a tool, on a fixed base.
The workhorse of industry. Reach, payload and repeatability define one completely.
- Ground mobile robots
Wheeled, tracked or legged machines that move over a surface.
Efficient on flat ground; the wheel/leg choice is decided almost entirely by the terrain.
- Aerial robots
Multirotors and fixed-wing UAVs.
Unmatched reach and viewpoint; punished on endurance and payload by the cost of staying up.
- Marine and underwater robots
ROVs on a tether, AUVs running free.
Pressure, corrosion and the fact that radio does not travel through water, so no GPS and no Wi-Fi.
- Humanoids and legged robots
Machines shaped for human environments, or for terrain wheels cannot cross.
Enormously capable and enormously complex; justified only when the shape genuinely pays off.
- Swarms
Many simple units acting on local rules, with no leader.
Resilient and scalable; individually near-useless, which is the point.
- Soft robots
Machines that move by deforming rather than by rotating rigid joints.
Intrinsically safe and adaptable; hard to model and hard to control precisely.
- Mobile manipulators
An arm on a mobile base: both families at once.
The combination inherits both sets of problems, which is why it is still a research frontier in unstructured settings.
Two other ways of sorting robots come up constantly, so keep them handy.
By application: industrial, service, medical, field, defence, domestic, research.
By autonomy: teleoperated, assisted, semi-autonomous, fully autonomous.
Any one robot sits somewhere on all three axes at once. A surgical robot, for instance, is a fixed-base manipulator, a medical application and almost entirely teleoperated.
Example
Reading a task and predicting the robot
"Inspect the underside of a bridge." Fixed base is out, because the robot must travel. Ground is out, because there is no surface under a bridge. That leaves aerial or a suspended crawler and the payload is a camera rather than a tool, so a multirotor with a gimballed camera it is. Endurance will be the binding constraint, so plan for battery swaps.
"Load parts into a CNC machine, 400 times a shift." Fixed position, repetitive, known part locations, moderate payload. That is a 6-axis arm on a pedestal, or a cobot if the operator shares the space. No perception needed if the parts are fixtured.
In both cases the task, not a preference for a kind of robot, produced the answer.
| Family | Buys you | Costs you |
|---|---|---|
| Industrial arm | Precision, payload, speed, reliability | It cannot leave its pedestal |
| Wheeled mobile | Efficiency and simple odometry | Needs continuous, flattish ground |
| Legged | Terrain wheels cannot cross | Complexity, energy, control difficulty |
| Aerial | Reach and viewpoint | Endurance, payload, weather, regulation |
| Humanoid | Fits human spaces and tools | Cost and control complexity, for a shape rarely optimal |
| Swarm | Resilience and scale | No single unit can do the job |
| Soft | Safety and adaptable grasping | Imprecision and modelling difficulty |
Why don't we just build one robot that does everything?
Because the requirements physically fight each other.
Payload wants mass and stiffness. Mobility wants lightness.
Precision wants rigidity. Safety around people wants softness.
Endurance wants efficiency. Speed wants raw power.
A robot that tried to satisfy all of them would be mediocre at each and expensive at all.
Humanoids are the closest anyone has come and their argument is subtler than it looks. Nobody claims the human shape is optimal for any task. The claim is that the world is already built for that shape, around doors and stairs and handles and tools. So one mediocre generalist might still beat buying five specialists.
Picking the robot family first and fitting the task to it.
Describe the work volume, the payload, the variation and the environment first. The family usually falls out of those four.
Assuming a humanoid is the advanced version of an arm.
They are different families for different jobs. A 6-axis arm outperforms any humanoid at manipulation on a bench and always will.
Treating mobile manipulators as a solved combination.
Base position error lands directly on the arm's accuracy. In unstructured settings this is still a research problem, not a product choice.
Key takeaway
- The first and most useful split is fixed base versus mobile base, because it determines nearly everything else.
- Fixed-base robots know exactly where they are; mobile robots must estimate it and can be wrong.
- Also classify by application and by level of autonomy; every robot sits on all three axes.
- Each family buys one thing and pays for it with another. There is no free combination.
- Shape follows task. Describe the job and the family usually names itself.
Finished this lesson?
Guest progress is saved on this device.