What was Unimate, the first industrial robot, and how did it work?

The word "robot" came from a 1920 play about artificial workers, and for decades robots lived mostly in fiction. In 1961 one finally clocked in at a real factory. It had no face, no voice and no legs. It was a heavy mechanical arm bolted beside a die-casting machine at a General Motors plant in New Jersey, and its job was to pull out red-hot metal parts so that people did not have to.

That machine was Unimate, and it launched the business of building robots for factories.

Who invented it

The idea came from George C. Devol Jr., an American inventor. On December 10, 1954, he filed a patent application for what he called "Programmed Article Transfer," a machine for moving objects automatically under the control of a stored program. The U.S. Patent Office granted it as patent 2,988,237 on June 13, 1961. Its opening lines describe an invention concerned with the automatic operation of machinery, and especially with handling equipment and its controls.

The patent alone did not make a robot. In 1956 Devol met Joseph Engelberger, who became his business partner, and the two talked about science fiction writing. They decided to build a real working robot. Together they founded Unimation, the first company devoted to making robots. The name Unimate came from that company.

When and where it went to work

In 1961 the first Unimate started work for General Motors, at its Inland Fisher Guide plant, located in Ewing Township, New Jersey. It tended a die-casting machine. Die casting means forcing molten metal into a steel mold, called a die, to make a metal part. When the part comes out, it is very hot. Workers who handled those parts risked burns, toxic fumes and injuries from the machinery.

Unimate took the castings out of the machines and also welded parts onto car bodies. The Robot Hall of Fame at Carnegie Mellon University describes these as tasks that are unpleasant for people, which is exactly why they were handed to a machine first.

How it worked

Unimate's arm weighed about 4,000 pounds and was connected to a large control cabinet. The key part was its memory. Devol's patent describes a program controller on which the positions the arm must reach are recorded as codes. As the machine runs, the controller compares where the arm is with the next recorded position, stops it when they match, and then moves on to the next step.

Devol preferred a magnetic recording for this, a rotating drum coated with magnetic material, similar in principle to a tape recorder. He chose it because it could be erased and re-recorded easily and could hold many instructions. That is what made Unimate programmable: change the recording, and the same arm does a different job.

Here is what one work cycle at a die-casting machine might look like when broken into steps, the way a program controller stores them:

Step	Stored instruction	What the arm does
1	Move to position A	Reach into the open die
2	Close gripper	Grab the hot casting
3	Move to position B	Lift it clear of the machine
4	Move to position C	Swing it over a cooling tank or conveyor
5	Open gripper	Release the part
6	Return to start	Wait for the next casting

Each instruction is simple. The value comes from doing them the same way, every time, for hours, without tiring or getting hurt.

Try it: Pick a small repetitive task, like making a cup of tea. Write it as numbered instructions that use only "move to," "grip," "release" and "wait." Then ask what happens if the cup is two inches to the left. That gap is the difference between a programmed robot and an intelligent one.

Why it mattered

Unimate showed that a general-purpose, reprogrammable machine could do dangerous factory work reliably. Because its program could be recorded again, the same arm could be given a new sequence of moves instead of being rebuilt for each job.

It also caught the public imagination. Unimate appeared on The Tonight Show with Johnny Carson, where it putted a golf ball into a cup, poured beer and waved the bandleader's baton. Later Unimate models had up to six programmable axes of motion and could handle parts weighing up to 500 pounds. Unimate robots spread through factories around the world, and in 2003 the Robot Hall of Fame at Carnegie Mellon chose Unimate as an inductee, in the same year as the Mars rover Sojourner and the fictional HAL 9000 and R2-D2.

Is an industrial robot like Unimate a form of artificial intelligence?

Not really, and the distinction is useful. Unimate did not perceive, reason or learn. It replayed a recorded sequence of positions. If a part was out of place, it would close its gripper on empty air. Artificial intelligence, as researchers at places like MIT and Stanford were defining it in the same years, was about programs that solve problems, use language and improve from experience.

Robotics and AI developed side by side, often in the same laboratories, and over time they have converged. In 2019 many factory robots still follow fixed programs much as Unimate did, but a growing number use cameras and machine-learning software to find parts in a bin, check quality or work safely near people. Researchers are also training robot arms to learn grasping skills by trial and error rather than by step-by-step programming.

So Unimate belongs on the AI timeline less as an example of intelligence than as the body that intelligence would eventually be given. It answered the question of whether a machine could do physical work in a real factory. The question of whether that machine could also think about its work is one AI is still answering.