Industrial Robotics: Programmed Paths and Factory Arms

The word arrived in a play. The arm arrived later

Josef Čapek suggested the word. Karel Čapek’s play R.U.R., written in 1920 and staged in 1921, sent “robot” out from the Czech robota, forced labor. The figures in that play were artificial workers. They were not a Unimate, and Karel later complained that people had turned his robots into mechanisms of sheet metal. He had not meant a jointed arm in a factory.

Keep the play where it belongs: a word, and a warning about labor. Do not draw a line from a Prague stage in 1921 to a die-cast machine in 1961. Engineers did not build the arm because a playwright named it. They built it because a plant wanted a motion repeated in a place a person kept getting burned.

Before the arm, the motion was cut into the iron

A transfer machine, a cam, a line of relays: those already did physical work. The assembly page and the PLC page said what they cost you when the product changed. You moved a station. You pulled wires. You cut a new cam. A robot does not abolish that. It moves the path into a memory you can overwrite, and it leaves the fixture, the tool, and the reach of the arm where they were.

Numerical control, on the computers page, had already aimed a cutter from numbers. A robot is a cousin of that idea, aimed at handling and at a tool the cutter is not. Cousin is the right word. A mill that follows a tape is not an ancestor that “became” an arm.

George C. Devol Jr. and the programmed transfer patent

Devol’s patent is titled “Programmed Article Transfer.” It describes a machine that can store a cycle of motions and repeat them, for moving articles, not a general mind and not a humanoid. In the patent he wrote that the invention made available, for the first time, a more or less general-purpose machine for cyclic digital control. That sentence is the inventor describing his own claim. I am not adopting “for the first time” as a finding. Other people were already positioning axes with numbers. What the patent stakes out is a programmable transfer arm.

A patent is not a shipment. The paper is 1954 and 1961. The machine in a plant is a separate date.

Joseph Engelberger and Unimation

Engelberger met Devol and put a company, Unimation, behind the idea. He spent years talking plant managers into letting a machine take a hot part. That is a real job, and it is why the arm left the patent and stood on a floor. His own book, Robotics in Practice, is his account of that work. It is not a neutral history, and it is not a reason to hand him a title.

“Father of robotics” is the same kind of title this series already refused for the PLC. Devol had the patent. Engelberger had the company and the sales. A crew built the hardware. General Motors had the job that was burning people. Later a Japanese licensee, Kawasaki, built the machines too. One man does not get the century.

1961: a hydraulic arm, a hot casting, a taught sequence

The installation the trade tells is a Unimate shipped in 1961 to a General Motors plant in New Jersey, pulling pieces from die-casting. I was not in the aisle. I am reporting the account the industry repeats, because that is the evidence we have: a first commercial arm in a plant, on a job that was repetitive and hostile, not a laboratory trick with a block of wood.

The early Unimate was hydraulic. It repeated positions it had been led through. It was not a modern electric robot computing a smooth path through six joints. It was strong enough for the work and dumb in the useful way: the same reach, again, while a person stayed out of the heat. Reliability was a fight. Plants that bought the second and third machines were not buying a solved technology. They were buying a motion they were willing to maintain.

How the arm actually moves

Count the pieces, or you will talk as if the software lifts the metal. A base. Links. Joints. An actuator on each joint you mean to move — a hydraulic cylinder on the early machines, an electric motor and a gearset on most later ones. A device that reports where the joint is, when the designer paid for one. A controller that stores the points. A tool on the end. Power. A fence, unless someone has done the work to say the fence can come down.

Several joints moving together can put the tool at a point in a volume the arm can reach. They cannot put it outside that volume because the program asked. Reach, payload, and speed are hardware. The program spends them. It does not enlarge them.

On a later controller, the box does compute: given a point in space, what angle does each joint need. That calculation is real, and it is later and tighter than a 1961 drum of taught points. Do not describe the first Unimate as if it already had it.

A servo is feedback. It is not Watt’s governor grown up

Turning a motor on and leaving it is not how you hit a point. A servo compares the angle you asked for with the angle the joint reports, and it corrects. Encoders and resolvers are the usual reporters. Speed and torque can be in the loop too, on the machines that bother.

The steam page had a governor that held an engine’s speed. The family resemblance is feedback: measure, compare, correct. It is not a bloodline. A flyball on a steam engine did not become a wrist encoder. Say “closed loop” if you mean the joint is checking itself. Do not say the robot inherited the governor.

The tool is the job. The arm is the reach

A bare wrist does not unload a press. The end effector does: a gripper, a weld gun, a paint gun, a vacuum cup, a dispenser. Change it and the same arm can be aimed at another task, after someone reteaches the path, rebuilds the fixture, and checks the new hazard. A gun that was safe as a handling job can be a different machine once it carries a sharp edge or a live torch.

Modularity is real. It is not magic. The catalog of tools is a catalog of jobs a person still has to prove out, one cell at a time.

The auto plants bought the repetition. They did not buy every task

Spot welding, handling, and paint were the early fits: the same motion, a hard environment, a volume high enough to pay for the cell. Assembly of a loose, floppy, or slightly different part waited. A robot that must feel a clip seat is a harder problem than a robot that must put a weld on a fixture that does not move.

Use spread as it happened. Unimation’s arm was licensed and copied outward. Other builders, in the United States, Europe, and Japan, sold electric arms with better path control. The automobile industry was the big early customer. It was not a law that every factory then bought a robot. A shop with fifty different parts and a short run often kept the person. It often still does.

The arm is not the cell

The PLC page already split this job, and this page will not steal it back. The robot controller owns the path of the arm. A PLC, in a great many cells, owns the room: conveyor, clamp, gate, the word that a part is present, the word that the press is clear. They handshake. Part there. Area safe. Arm clear. Task done. Get the handshake wrong and both boxes are healthy while the cell is wrecking a fixture.

There is no law that the PLC is the boss. Write the split down before the crash writes it for you.

Most arms still assume the part is where the fixture put it

The first generations repeated a path in a world the fixture held still. That is still how a large share of industrial arms earn their keep. A sensor can add a check: force when a gripper stalls, a proximity switch when a nest is empty, an encoder on the joint you already needed. Those are permits and corrections. They are not awareness of a room.

Vision can find a part that is not fixtured to the millimeter, or read an orientation, or fail a surface. It can also be fooled by light, oil, and a part the camera was never shown. A camera on a robot is a sensor with a lot of bits. It is not a pair of eyes. Use it where the cell was proved. Do not say the arm became adaptive because a lens was bolted on.

A cobot is a risk assessment, not a friendly species

A conventional industrial arm is fenced because speed, mass, and a tool can kill. In the mid-1990s J. Edward Colgate and Michael Peshkin, at Northwestern, built devices they called cobots for a General Motors question: how a person and a mechanism might share a task. The first of those devices did not power the motion themselves. The person supplied the force. The mechanism steered. That is not the powered arm later sold under the same nickname.

The nickname drifted. It now sits on electric arms with force limits, speed limits, and monitored stops, meant for applications a risk assessment has passed. The name does not pass the assessment. A “collaborative” arm with a knife, a heavy payload, or a fast program is a regular hazard with a softer sales word. People and arms can share a station when the application was built for that. They do not share it because the brochure used the word cobot.

Wheels are the next page, not a soul

An arm bolted to the floor is not the only machine the trade calls a robot. A cart that follows a wire or a tape in the floor is a guided vehicle. A later cart that picks a path around a pallet is a different control problem. Both still move because someone mapped the building and someone answers when a sensor is dirty. Calling that “autonomous” without saying what it cannot see is how a warehouse brochure gets ahead of the truck.

That story is the warehouse page. I will not finish it here. An arm that can reach and a cart that can roll are not one invention, and neither one decided to go to work.

Where this sits in the history of automation

A cell you can buy now may put an electric arm, an encoder, a PLC, a camera, and a network in one fence. The arm can be told to find a part that is a little out of place, to take a path that avoids a post, and to stop when a force says the grip is wrong. Those are options on a purchase order. They are not the 1961 machine, and they are not installed in every plant that owns an arm.

What gets called artificial intelligence may classify a picture or flag a drift. It does not become responsible for the stop, and it does not inherit the job of the interlock. A person still owns the decision to let the arm run when the guess is wrong. I am describing equipment. I am not describing a creature.

What I want you to keep

Josef suggested the word. Karel’s 1920 play sent it out. The play’s robots were not factory arms. Devol’s patent was filed on December 10, 1954, and issued on June 13, 1961, as a programmed transfer machine. His “for the first time” is his claim, not ours. Engelberger and Unimation got an arm onto a floor. He is not the father of robotics. The installation the trade cites is a Unimate at GM in New Jersey in 1961, hydraulic, repeating taught points on hot die-cast work.

A servo checks a joint. It is not Watt’s governor. A tool change is still a new job to prove. A PLC still owns a lot of the cell. Vision is a sensor. A cobot from the mid-1990s did not power itself, and a powered arm with that nickname is only as safe as the application. The cart belongs to the warehouse page. And the arm moves the path it was given. It does not move on its own.