Automatic is not autonomous
A machine can be automatic without being autonomous. A washing cycle or a factory arm repeating the same weld follows a fixed script. An autonomous robot, in the sense used by researchers, senses its surroundings, decides what to do and acts without a person controlling each step. This history follows the move from the first, to the second, starting with factory robots.
Programmable arms
The modern industrial robot is usually traced to the American inventor George Devol, who filed a patent in 1954 for a programmable transfer machine, granted in 1961, and to the engineer and entrepreneur Joseph Engelberger, who turned the idea into a business. Their company built a hydraulic arm called Unimate. One was installed at a General Motors plant in New Jersey in 1961, where it is generally described as handling hot metal die-cast parts, a dangerous task for people. Robots like this were taught a sequence of positions and then repeated it, with no sense of the world beyond the position of their own joints.
Through the 1970s, electric drives replaced hydraulics in many designs. The Swedish company ASEA and the American company Cincinnati Milacron both introduced electric, computer-controlled arms in the early 1970s. Victor Scheinman’s Stanford arm in 1969 and the PUMA arm of the late 1970s, built with General Motors, influenced the layout of many later robot arm designs, and in Japan a design known as SCARA was developed around 1978 for assembly work. Car makers were the first major customers, using robots for welding and painting, and Japan became a leading adopter in the 1980s.
These machines were precise and tireless but not autonomous. They needed the workpiece to be in exactly the right place and were typically fenced off from people.
Shakey and the research robots
At the same time, laboratories were trying to build robots that decided for themselves. The best-known early example is Shakey, developed at SRI (then the Stanford Research Institute) between about 1966 and 1972. Shakey was a wheeled platform with a camera, a range finder and bump sensors, connected by radio to a large computer that planned its actions. It could be asked to move an object from one room to another and would work out a sequence of steps. It was slow, and the environment was simplified, but the project produced lasting ideas in artificial intelligence, including methods for planning and for finding routes.
Through the 1980s, researchers such as Rodney Brooks at MIT argued that useful autonomy might come from simple reactive behaviors layered together, rather than from an elaborate internal model of the world. This helped shift the field toward robots that respond to sensor readings in real time.
Moving beyond the factory
Practical autonomous robots first appeared in settings that could be controlled. Automated guided vehicles followed wires or markers in factories and warehouses from the 1950s onward and later used lasers or cameras to find their way. Hospital delivery robots appeared in the 1980s and 1990s. Warehouse systems in which mobile robots carry shelves to workers became prominent in the 2000s.
At home, the breakthrough was modest in ambition but large in numbers: the robotic version of household vacuum cleaners. Such machines reached mass markets in the early 2000s. They use bump sensors, infrared and later cameras or laser distance sensors to cover a room and avoid obstacles. They are among the few autonomous robots many people own. The wider story of robots built for care and domestic roles belongs to the history of service robots.
Vehicles and drones
The same ideas of sensing and planning fed other fields. Experimental self-driving cars were being tested in the 1980s, and US government-sponsored contests in 2004 and 2005 pushed researchers to build vehicles that could cover long off-road routes by themselves. Unmanned aircraft and underwater vehicles have similarly moved from remote control toward preset or self-directed missions. The details of those fields lie outside the robot-arm lineage described here.
What limited progress
Several obstacles slowed the arrival of general-purpose autonomous robots. Computers were slow and expensive for decades. Sensors were noisy, and cameras produced data that software struggled to interpret. The real world is full of surprises compared with a lab with painted floors. The physical side is tough too, since safe, strong, cheap arms and hands are hard to build.
Gains in computing power, cheap sensors and machine learning in the 2000s and 2010s changed that, though most successful robots still work in structured environments or on narrow tasks. A robot that can fold laundry in an arbitrary home or work reliably in a messy kitchen remains largely a research goal.
Looking back
The path from Unimate to a robot vacuum is not a straight line. Industrial robots grew by being very good at repetition, while research robots like Shakey explored decision making but were impractical. The two threads have been converging, as factory robots gain vision and force sensing and mobile robots gain better arms. Autonomy is best thought of as a spectrum, with different machines at different points along it.