George C. Devol

The patent for the first programmable industrial robot — Unimate, a machine that could be taught a sequence and then repeat it, not the invention of robotics itself

George C. Devol lived from 1912 to 2011. In 1954 he filed a patent for a “programmed article transfer” device. The machine that later carried the name Unimate was a hydraulic arm that could be taught a sequence of motions and then repeat that sequence without a person standing at the controls for every cycle. That is a real change in factory machinery. It is not the invention of the robot, and it is not the moment automation began.

Earlier pages on this site refused to treat ancient automata, Jacquard’s cards, or Vaucanson’s mechanisms as computers or self-running plants. The same caution belongs here. Unimate followed a taught path. It did not sense a changing environment and rewrite its own instructions. It did not decide. It repeated what it had been shown, within the limits of its hydraulics and its controller. That was enough to be useful on a factory floor. It was not enough to make it a thinking machine.

The patent and the first installations sit in the 1950s and early 1960s. Joseph Engelberger later helped turn the idea into a company and into machines that General Motors and others would buy. Devol is the name on the original patent. The commercial history is a partnership and a market, not a single inventor working alone in a garage.

Interpretive illustration of George C. Devol with an early Unimate industrial robot arm
A teaching picture. Not his face. The arm and the control cabinet stand in for the Unimate idea. The machine followed a taught sequence. It did not invent the next one.

The 1954 patent

Devol’s U.S. Patent 2,988,237, “Programmed Article Transfer,” was filed in 1954 and issued in 1961. The claim that matters is that a machine could be taught a series of positions and motions and then repeat them. The teaching was done by leading the arm through the path or by setting controls. Once the sequence was stored, the machine could run the cycle again. That is the practical difference from a fixed cam or a hard-wired mechanism. You could change the job without rebuilding the hardware.

The patent does not describe a machine that sees, that plans, or that corrects itself when the part is in the wrong place. It describes a device that can store and replay a path. Later industrial robots added sensors, better controllers, and more flexible programming. Those additions belong to other decades and other engineers. Do not move them back to 1954 to make the first patent do more than it did.

Unimate on the factory floor

The first Unimate installations were in the early 1960s, notably at a General Motors die-casting plant. The arm unloaded hot parts from a machine. That was a job that was hot, repetitive, and unpleasant for a person. The robot did not need to understand die casting. It needed to reach the same places in the same order, cycle after cycle. When the sequence was wrong, someone retaught it. The machine did not notice that the part had stuck or that the next cycle should be different.

Engelberger’s company, Unimation, built and sold the machines. Devol supplied the original concept and the patent. The commercial success of the early industrial robot is therefore a company story as well as an inventor’s story. A patent is not a factory full of robots, and a single installation is not an industry. The later spread of industrial robots belongs to many firms, many engineers, and decades of other work.

Hold this line: Unimate could be taught a path and could repeat it. It could not sense that the path was wrong and invent a better one. Teaching is not learning in the later sense of a system that updates its own model. A hydraulic arm that follows a stored sequence is a useful factory machine. It is not a robot that thinks.

What the early industrial robot did not do

  • It did not invent the idea of a mechanical arm or of automatic machinery. Fixed automation and transfer lines were already in factories.
  • It did not see, plan, or correct itself. It repeated a taught sequence within the accuracy of its hydraulics and its controller.
  • The first installations were not a general-purpose robot that could be dropped into any factory job. They were machines for specific, repetitive tasks.
  • Engelberger and Unimation were necessary for the commercial machines. The patent alone did not put robots on production lines.
  • Later robots with sensors, vision, and more flexible programming are not “Unimate finished.” They are later machines with later capabilities.

What I want you to keep

1912–2011. U.S. Patent 2,988,237, filed 1954, issued 1961, for a programmed article-transfer device. The machine later called Unimate could be taught a sequence and then repeat it. Early installations, including at General Motors, used it for hot, repetitive handling. Joseph Engelberger and Unimation turned the idea into commercial machines. The robot followed a stored path. It did not sense a changing world and rewrite its own instructions. That is the page. The button after this one goes home. It does not go on to a machine that Unimate was “preparing.”

References and Further Reading

  1. George C. Devol, “Programmed Article Transfer,” U.S. Patent No. 2,988,237, filed 1954, issued 1961. The patent. Start here. A device that can store and replay a path. Not a claim to have invented the robot or artificial intelligence.
  2. Joseph F. Engelberger, Robotics in Practice (1980) and later writings. The commercial and industrial side of the early Unimate story. Engelberger’s account is useful for the company and the installations. It is not a substitute for the patent.
  3. The first Unimate installations at General Motors in the early 1960s are documented in industrial and robotics histories. A single plant installation is not the birth of an industry.
  4. Encyclopaedia Britannica, “George Devol.” A short life. Where it treats the Unimate as the origin of modern robotics in a broad sense, follow the patent and the limits of a taught sequence.