Chapter 2: Evolution of Robots and Laws of Robotics
In 1961, a car factory in the United States gave one of its worst jobs to a machine. The job was to pick up hot metal parts as they came out of a die-casting machine, where molten metal is poured into moulds. The parts were heavy, the air near them was hot, and the fumes were harmful to breathe. A person had to stand there all day and lift part after part.
The machine that took over was called Unimate. It was a single heavy arm bolted to the floor. It could not see, it could not hear, and it did not choose anything. It simply repeated a set of stored movements, one after the other, all day.
Sixty years later, there are machines that walk up stairs, sort parcels, and work at the same bench as a person. None of them appeared suddenly. Each one was built on top of the one before it.
Diagram
Fig. 2.0-1 – a simple horizontal timeline running left to right with five marked points: 1920 (a theatre mask, labelled “the word robot”), 1961 (a single fixed factory arm), 1970s (an arm with a camera on top), 1990s (a walking humanoid figure), today (a four-legged robot beside a person). No dates other than these five, and no other text on the drawing.
This chapter will answer each of these five questions.
2.1 Where the Word “Robot” Came From
Think of the jobs at home that nobody wants. Filling water, sweeping the same floor twice a day, carrying heavy bags up the stairs. The work is not difficult to understand. It is just long, repeated and tiring.
People have wanted machines to take over that kind of work for a very long time. Long before electricity, before motors, before computers. That wish is where the story of robots begins.
Machines that moved on their own
In ancient Greece, engineers and inventors such as Hero of Alexandria built devices called automata — machines that could move by themselves for a short while using water, air, weights or steam. One might open a door. Another might make a sound or move a small figure.
An automaton was not a robot. It could not be given a new job. Whatever it was built to do, it did that one thing, in that one order, every single time. Still, it showed something important: movement could come from a machine instead of a hand.
The year the word appeared
For hundreds of years there was no single word for a machine-worker. The word arrived in 1920, and it came from a play, not from a factory.
A Czech writer named Karel Capek wrote a play called R.U.R., which stands for Rossum’s Universal Robots. The play is about artificial workers who are built to do human labour. Capek took the Czech word robota, which means forced or hard labour, and shortened it to “robot”.
Diagram
Fig. 2.1-1 – a simple two-panel drawing. Left panel: an ancient Greek workshop with a small mechanical bird on a stand, labelled “automaton — moves and makes sound, does only what it was shaped to do”. Right panel: a theatre stage in 1920 with a poster reading “R.U.R.”, labelled “the word robot is used for the first time”.
The play was fiction. There was no real robot anywhere in the world in 1920. But the word was easy to say and it fitted an idea people already had, so it spread into almost every language and stayed. Today it is used in English, Hindi and dozens of other languages exactly as Capek wrote it.
2.2 The First Industrial Robots
Look at what happens at a bakery counter during a rush. The person at the counter picks up a box, folds the flap, drops in the same number of biscuits, closes the lid, and slides it across. Then does it again. And again.
A factory is that, multiplied. The same weld in the same spot on the same part, four hundred times a shift. The work does not need new thinking each time. It needs the same movement repeated with the same accuracy, without getting tired and without getting bored.
That is the kind of work the first factory robots were built for. A machine of this sort is called an industrial robot.
Why “programmable” is the important word
An automaton could only do what its gears and levers were shaped to do. An industrial robot is different in one way that matters more than all the rest: its instructions are stored separately from its body. Change the stored instructions, and the same arm does a new job.
This is why factories were willing to pay for them. A machine built for one task becomes scrap when the product changes. A robot arm gets a new set of instructions and keeps working.
But there is an honest limit to this. The first industrial robots had no way of checking what was actually in front of them. If a part was missing, the arm still moved, gripped nothing, and carried nothing to the next station. It could not notice the mistake, because it had nothing to notice with.
Diagram
Fig. 2.2-1 – a single robot arm bolted to the floor beside a conveyor belt. A shaded arc on the floor shows how far the arm can reach, with a safety fence drawn along the edge of the arc and a gate in it. Label the arc “working area”.
2.2.1 Unimate: A Robot for a Dangerous Job
The first industrial robot did not go to work in a laboratory. It went to a car factory, and it was given a job that was harming people.
The job was at a die-casting machine. Die casting means pushing molten metal into a mould to make a part. When the mould opens, the part that comes out is heavy and extremely hot, and the air around it carries fumes that are unsafe to breathe. Somebody had to stand at that opening and lift out part after part, all day.
Who built it
An American inventor, George Devol, designed a machine for exactly this kind of task and applied for a patent for it in the 1950s. He could not sell it on his own. An engineer named Joseph Engelberger saw what it could be worth, worked with Devol to turn the design into a real product, and set up the company that manufactured it. Their machine was called Unimate.
In 1961, a Unimate arm started work at a General Motors plant in New Jersey, in the United States. It lifted hot die-cast parts out of the machine and stacked them. Later Unimate arms were used for spot welding on car bodies.
Diagram
Fig. 2.2-2 – photo of an original Unimate industrial arm on museum display: a large boxy base with a single thick jointed arm reaching out and a gripper at the end.
How it was told what to do
Unimate did not read a program typed on a keyboard. There were no keyboards on factory machines then. Instead, an operator moved the arm by hand to each position in turn and pressed a button to record it. The arm stored those positions in a magnetic drum memory inside its base, and afterwards it repeated the same sequence of positions on its own.
This way of programming is still used today, and it is a sensible one. The person who knows the job best is the one who sets the positions, and no one has to describe a movement in words.
2.2.2 Research Robots of the 1960s and 1970s
A factory arm that repeats stored positions is useful, but it cannot leave its spot and it cannot cope with anything unexpected. Move the box of parts fifty centimetres to the left and the arm goes on gripping empty air.
So while factories were buying arms, universities were asking a different question. What if a computer controlled the robot, and the robot could look at the world before it moved? Researchers such as Marvin Minsky and Seymour Papert were working on artificial intelligence at this time, and robots became the place to test those ideas on real hardware.
An arm built for a computer
In 1969, Victor Scheinman built the Stanford Arm at Stanford University in the United States. It had six joints, so it could reach a point from many different angles, much as your own arm can touch the same spot on a table with your elbow high or low.
Two things made it important. It ran on electric motors rather than hydraulic pressure, which made it cleaner and easier to control precisely. And it was designed from the start to be driven by a computer instead of by recorded hand movements. It was used for assembly, welding and moving material.
A robot that could move and look
At about the same time, a very different machine was being built at a research institute in California. It was a mobile robot — a robot that moves from place to place instead of staying bolted down. It was called Shakey, and work on it ran from the mid-1960s to 1972.
Shakey had a television camera, a range finder to judge distance, and bumpers to detect contact. A large computer in another room did the working out and sent instructions by radio. Given a goal such as pushing a box off a platform, Shakey could break that goal into a series of steps and carry them out.
Diagram
Fig. 2.2-3 – a labelled sketch of Shakey: a tall wheeled box with a television camera on a mast at the top, a radio antenna above it, a range finder below the camera, and a bump bar around the base. An arrow to one side points to a separate computer in another room, joined by a wavy radio link.
Shakey was slow. It could take many minutes to work out and complete one move, and it only worked in rooms prepared for it, with plain walls and large blocks it could recognise. But it did something no factory arm could do: it decided the order of its own steps.
From the laboratory back to the factory
Ideas from research do reach real workplaces, though it takes time. In 1978, Scheinman’s work led to the PUMA robot, short for Programmable Universal Machine for Assembly. PUMA arms were compact, electrically driven and computer controlled, and they went on to be used widely on car assembly lines and in electronics assembly.
2.4 Generations of Robots
The story so far has run along a timeline: 1920, 1961, 1972, 2000, 2012. A timeline is useful, but it does not tell you what a robot could actually do.
Here is a better way to sort them. Think of the phones sold in any market today. Some are plain keypad phones that make calls and send messages. Some are smartphones with a camera, maps and a hundred apps. Both are on sale this year, so the year tells you nothing. What separates them is capability.
Robots are grouped the same way. The generation of robots means the stage of technology a robot belongs to, judged by its capabilities, its design and the technology built into it.
Diagram
Fig. 2.4-1 – a staircase of five rising steps, left to right. Step 1: “mechanical only, no sensing”. Step 2: “basic sensors”. Step 3: “vision and touch”. Step 4: “learns from data, decides its next action”. Step 5: “works safely alongside people”. Each step drawn slightly higher than the one before, with no dates written on the drawing.
Robots are usually grouped into five such generations. The first four describe machines that exist and are working now. The fifth is still emerging, which means it describes what is being built and tested rather than what is finished.
2.4.1 First and Second Generation Robots
First generation: strong, steady and blind
First generation robots were mechanical machines with no sensing and almost no computing. They held a fixed sequence of movements and repeated it. Industries used them for work that was dangerous or numbingly repetitive, such as assembly line work. Unimate is the standard example.
Everything about them followed from one fact: they could not check anything. Whether the tray in front of them was full or empty, the arm ran its sequence. So the factory had to be built around the robot — parts delivered to an exact spot, at an exact moment, every time.
Within those limits they were excellent. They were fast, they did not tire, and they held the same position hour after hour. Machines of this type are still bought today for work where nothing varies.
Second generation: the first sense of the outside world
The obvious weakness was the blindness. A missing part, a part lying crooked, a tray that had not arrived yet — none of it made any difference to the machine’s behaviour, and the mistake travelled down the line.
Second generation robots fixed part of this by adding basic sensors, most commonly a proximity sensor and simple control sensors. A proximity sensor reports whether an object is close by, without touching it. You meet one every time you put your hands under a tap in an airport washroom and the water starts.
Now the robot could hold its movement until the sensor reported that a part was there. See Fig. 2.4-2 for the difference this makes to the same simple task.
Diagram
Fig. 2.4-2 – two rows showing the same pick-and-place task. Top row, labelled “first generation”: three frames of an arm swinging to an empty tray, closing its gripper on nothing, and carrying nothing to the next station. Bottom row, labelled “second generation”: three frames of an arm with a small proximity sensor on the gripper, waiting with the tray empty, then closing the gripper once the sensor reports a part present.
These robots also needed something to read the sensor and act on it. That job went to a microcontroller — a small, complete computer on a single chip, built to run one program and control connected hardware directly — and to the early computer systems of the time. Being able to react to a signal made second generation robots far more versatile than the ones before them, and they could handle a wider range of tasks. The PUMA arm, short for Programmable Universal Machine for Assembly, is the usual example.
2.4.2 Third and Fourth Generation Robots
Third generation: seeing and feeling
Knowing that something is there takes a robot only so far. Suppose parts arrive on a belt lying at any angle, or suppose the object to be picked up is an egg. Now the robot needs to know where exactly the part is, which way round it is lying, and how hard it is squeezing.
Third generation robots added the sensors that answer those questions: vision systems and touch sensors.
A vision system is a camera together with a program that measures things in the picture — the position of an edge, the outline of a shape, the distance between two marks. The camera by itself only supplies an image. The measuring is done by the program.
Touch and force sensors report how hard the gripper is pressing. That is what allows a robot to hold something fragile: it increases its grip until the force reading reaches a set value, and then stops squeezing.
With these, a robot can handle changing conditions to some extent. If the part has shifted two centimetres, the vision system measures the new position and the arm goes to it. Third generation robots could therefore take on more complex tasks and interact with their surroundings far more efficiently. The ABB IRB 6700, a large industrial arm used for heavy work in car factories, is an example of a robot of this level.
Fourth generation: rules worked out from data
Even a robot with a camera still follows rules that a person wrote by hand. Somebody decided what counts as an edge, what counts as a correct part, what to do in each case. That works while the situations can be listed. It stops working when they cannot.
Nobody can list every way a warehouse floor might be blocked, or every angle at which a parcel might be lying. This is where artificial intelligence (AI) comes in — the branch of computer science that builds systems able to carry out tasks that normally need human judgement, such as recognising an object or choosing a route.
The part of AI that matters most here is machine learning, in which a program works out its own rules from many examples instead of being given the rules directly. Show a system thousands of labelled pictures of parcels and it settles on the pattern that separates a parcel from a shadow. Nobody wrote that rule down.
Fourth generation robots are built around this. They have more autonomy, choosing their next action from what their sensors report, and their performance improves as more data is collected. They are used in healthcare, logistics, inspection work and self-driving vehicles. Boston Dynamics Spot, a four-legged robot used mainly for inspection and collecting data at industrial sites, is a well-known example.
Diagram
Fig. 2.4-3 – photo of a four-legged inspection robot walking over uneven ground at an industrial site, with a camera unit mounted on its back.
There is an honest limit here that matters. A robot that learns from data is only as good as the data it was given. If a delivery robot’s data came entirely from wide, smooth footpaths, it will handle a narrow, broken one badly — and it will not be able to tell anyone why it failed, because no person wrote the rule it was following.
2.4.3 Fifth Generation Robots
Everything in this chapter so far has already been built. The fifth generation has not, or not fully. It is described as emerging, which means the machines are being designed and tested now, and the description is of what they are expected to do rather than of what is finished and working.
Fifth generation robots are expected to have greater autonomy still, more advanced AI, and the ability to work alongside humans safely and collaboratively. The examples usually given are advanced humanoid robots, and robots meant for complex tasks in unstructured environments.
The hard word in that sentence
The whole difficulty sits in that last term.
An unstructured environment is a place that was not arranged for a machine — where objects are not in fixed positions, the lighting changes, people and animals move about, and no two situations are quite alike. A kitchen at home is one. A field after heavy rain is one. A collapsed building is one.
Remember how the first factory robots worked: the factory was arranged around the machine. Every generation since has moved a little away from that, and the fifth generation aims to finish the journey — a robot that fits into a place as it already is, and needs nothing rearranged for it.
Diagram
Fig. 2.4-4 – two scenes side by side. Left, labelled “structured”: a factory cell with a fixed jig, taped floor markings, parts in identical trays, constant lighting. Right, labelled “unstructured”: a home kitchen with a chair pushed out at an angle, utensils lying in different spots, a dog on the floor, sunlight falling across half the room.
Where this actually stands
It is worth being honest about the gap. Impressive demonstrations of walking, lifting and sorting exist, but most of them are run in conditions the team chose. Several problems remain genuinely unsolved: gripping unfamiliar objects reliably, working for long hours on a battery, and proving to a safety inspector that a machine is safe around children and elderly people.
The collaborative robots of today are a first step towards this, and they are deliberately kept slow and light so that a bump does no harm. A machine strong enough to lift a person, moving freely in a home, is a much harder safety problem and is not close to being settled.
2.5 What Robotics Means
A car is a machine. Designing cars, building them, testing them, repairing them and improving them is not a machine — it is a whole field of work, with people trained in it.
The same distinction applies here. A robot is a machine. Robotics is the field of work and study built around those machines.
Three kinds of work in one field
Building even a simple robot needs three different sorts of skill together, and that is what makes robotics its own field rather than a corner of some other one.
There is the mechanical side: the body, the joints, the arms or wheels, the motors that move them, and making sure the whole thing takes the weight it must without bending.
There is the electronics side: the sensors, the circuits that carry their signals, the power supply, and the board that everything plugs into.
And there is the computing side: the program that reads the sensor values, works out what to do, and sends instructions to the motors. Since the fourth generation, this side increasingly includes AI.
Diagram
Fig. 2.5-1 – three overlapping circles in a triangle. Circle one labelled “mechanical — body, joints, motors”. Circle two labelled “electronics — sensors, circuits, power”. Circle three labelled “computing — the program that reads sensors and drives motors”. The overlap in the centre labelled “robotics”.
Leave out any one of the three and there is no robot. A body with no sensors cannot tell what is around it. Sensors with no program produce readings nobody uses. A program with no body can compute all day and move nothing.
2.6 The Three Laws of Robotics
When robots first appeared in stories, they nearly always turned on the people who built them. That was the whole plot: clever machine, careless inventor, disaster.
An American science fiction writer named Isaac Asimov found this dull and untrue to how engineering actually works. Nobody sells a pressure cooker without a safety valve. If robots were ever manufactured and sold, he reasoned, they would come with safety rules built in, and the interesting stories would be about what happens when those rules meet a situation nobody planned for.
So for his stories he wrote a set of rules that every robot in them was built to follow. They are called the Laws of Robotics, and there are three.
The three laws
First law: A robot must not injure a human being, and must not stand by doing nothing while a human being comes to harm.
Second law: A robot must obey the orders given to it by human beings, except where those orders would conflict with the First Law.
Third law: A robot must protect its own existence, as long as doing so does not conflict with the First or Second Law.
The order is the whole idea
Notice that each law is allowed to act only when it does not clash with the one above it. The order is not decoration.
Human safety sits at the top and cannot be overruled. Obedience comes next, so an order that would hurt someone is refused. The robot’s own survival comes last, so a robot will let itself be damaged rather than disobey an instruction or allow a person to be hurt.
Diagram
Fig. 2.6-1 – three horizontal bands stacked vertically. Top band, widest, labelled “First Law — do not harm a human”. Middle band labelled “Second Law — obey human orders”. Bottom band, narrowest, labelled “Third Law — protect itself”. A downward arrow at the side labelled “a lower law applies only when it does not clash with a higher one”.
What the three laws did achieve was a way of thinking. They put the idea into people’s heads that a machine’s responsibilities should be decided before it is built, not after something goes wrong. That idea is now taken seriously by the people who design robots and the governments that regulate them, even though the wording they use is nothing like Asimov’s.
2.7 Summary
- Ancient engineers built automata — machines that moved on their own using water, air, weights or steam. They could not be given a new task, so they were not robots.
- The word “robot” was first used in 1920 by the Czech writer Karel Capek in his play R.U.R., from the Czech word robota, meaning forced or hard labour.
- An industrial robot is a programmable machine used in a factory for a physical task. Because its instructions are stored separately from its body, the same arm can be given a new job.
- Unimate, invented by George Devol and produced with Joseph Engelberger, began work at a General Motors plant in 1961. It lifted hot die-cast parts and later did spot welding. It had no sensors, so the factory had to be arranged around it.
- In the 1960s and 1970s, research robots added computer control and sensing. The Stanford Arm (1969) was electric, had six joints and was built to be driven by a computer. Shakey (1972) was a mobile robot with a camera, a range finder and bumpers, and could work out the order of its own steps. The PUMA arm (1978) brought this work back into factories.
- Smaller, cheaper computers and sensors let robots leave the fence. Autonomy means carrying out a task without a person guiding each movement.
- A humanoid robot is built in a human shape so it can use stairs, doorways and spaces made for people. Its main difficulty is balance, which it manages with tilt sensors, force sensors in the feet and constant motor corrections. ASIMO, introduced by Honda in 2000, is the best known example.
- A collaborative robot works beside a person with no cage. It is kept safe by being light, slow, and able to sense force and stop on contact. Baxter (2012) is an example. Its trade-off is low speed and light loads.
- Robots are grouped into five generations by capability, not by the year they were made.
- First generation: mechanical only, no sensing, used for dangerous or repetitive work. Example: Unimate.
- Second generation: basic sensors such as proximity sensors, controlled by microcontrollers and early computers, so more versatile. Example: PUMA.
- Third generation: vision systems and touch sensors, so the robot can measure where things are and how hard it is pressing, and can cope to some extent with change. Example: ABB IRB 6700.
- Fourth generation: greater autonomy, with artificial intelligence and machine learning working out rules from data instead of a programmer writing every rule. Used in healthcare, logistics, inspection and self-driving vehicles. Example: Boston Dynamics Spot.
- Fifth generation: still emerging. Aimed at high autonomy, advanced AI, and safe collaborative work with people in unstructured environments — places not arranged for a machine.
- Robotics is the branch of engineering and science dealing with designing, building, programming and using robots. It joins mechanical work, electronics and computing.
- The three Laws of Robotics were written by Isaac Asimov for his fiction: a robot must not harm a human or allow harm through inaction; it must obey human orders unless that conflicts with the first law; it must protect itself unless that conflicts with the first or second law.
- The laws are ordered, so a lower law applies only where it does not clash with a higher one. They are fiction. Real robots are made safe by force limits, guards, emergency stops, tested software and legal safety standards.
2.8 Key Terms
- Artificial intelligence (AI) — the branch of computer science that builds systems able to carry out tasks that normally need human judgement, such as recognising an object or choosing a route.
- Automata — early machines that could move by themselves for a short while using water, air, weights or steam, but could not be given a new task.
- Autonomy — the ability of a robot to carry out a task without a person guiding each movement.
- Collaborative robot (cobot) — a robot designed to work safely beside a person without a fence, by being light and slow and by stopping when it senses contact.
- Generation of robots — a group of robots sharing the same level of capability: how much they can sense, how much they can work out for themselves, and how far they can adapt.
- Industrial robot — a programmable machine used in a factory for a physical task such as lifting, welding or placing parts, usually an arm fixed in one position.
- Laws of Robotics — the three ordered safety rules written by Isaac Asimov for his fiction: do not harm a human, obey human orders, protect itself, each applying only where it does not clash with the one above.
- Machine learning — a way of building a program so that it works out its own rules from many examples instead of being given the rules directly.
- Microcontroller — a small, complete computer on a single chip, built to run one program and control connected hardware directly.
- Mobile robot — a robot that moves from place to place instead of staying fixed in one position.
- Proximity sensor — a sensor that reports whether an object is present within a short distance, without touching it.
- Robotics — the branch of engineering and science that deals with designing, building, programming and using robots.
- Unstructured environment — a place whose contents, layout and conditions are not fixed or known in advance, so a robot working there cannot rely on things being where they were last time.
2.9 Objective Questions
The word “robot” was used for the first time in a play written in 1920 by:
- George Devol
- Karel Capek
- Isaac Asimov
- Victor Scheinman
Fill in the blank: The word “robot” comes from the Czech word ____________, which means forced or hard labour.
State whether true or false: A robot arm bought this year must be a fifth generation robot, because generations are decided by the year a robot is made.
A sensor that reports whether an object is present within a short distance, without touching it, is called a:
- vision system
- force sensor
- proximity sensor
- tilt sensor
In one line each, name the robot described: (i) the arm that began lifting hot die-cast parts at a car plant in 1961, and (ii) the mobile research robot that used a camera and worked out the order of its own steps.
Match the capability in Column A with the generation in Column B.
Column A Column B a. Mechanical only, no sensing at all i. Second generation b. Basic sensors read by a microcontroller ii. Third generation c. Vision systems and touch sensors iii. Fourth generation d. Rules worked out from data by machine learning iv. First generation Which of the following is the correct order of the three Laws of Robotics, from highest priority to lowest?
- Obey orders, protect itself, do not harm a human
- Do not harm a human, obey orders, protect itself
- Protect itself, do not harm a human, obey orders
- Do not harm a human, protect itself, obey orders
2.10 Concept Questions
Define robotics. Name the three kinds of work it brings together.
An automaton in ancient Greece could move by itself, yet it is not called a robot. Explain the difference.
The first industrial robots worked only in factories that were arranged around them. Explain why, with reference to what these robots could and could not sense.
Differentiate between third generation and fourth generation robots, giving two clear points of difference.
A collaborative robot is deliberately built to be light and slow, even though faster and stronger arms are available. Give the reason for this design choice, and state one job it makes the cobot unsuitable for.
State the three Laws of Robotics in order. Explain why the order in which they are placed is an essential part of the idea.
A humanoid robot is walking across a room when the tilt sensors in its body stop sending readings. What would happen, and why? Relate your answer to how such a robot keeps its balance.
2.11 Case-Based Questions
- A small biscuit factory in Jodhpur is buying two machines for its packing hall. The details offered by the supplier are given in Table 2.4.
- Which generation of robots does Machine A belong to? Give one reason from the table.
- Machine B works at the same bench as a worker, with no fence. Explain how it is kept safe.
- The hall has two tasks: stacking sealed cartons onto a pallet all day, and placing a printed slip into each packet as a worker checks it. Say which machine suits each task, and why.
A hospital wants a robot that can carry medicines from the pharmacy to the wards on its own. The corridors are busy. Trolleys are parked in different places each day, visitors walk in both directions, and strong sunlight falls across one stretch in the afternoon.
- Is this corridor a structured or an unstructured environment? Give one reason from the description.
- The robot must not bump into a trolley that was not there yesterday. Explain what the robot needs in order to manage this, and why a stored route by itself is not enough.
- A member of staff suggests that the hospital simply program Asimov’s First Law into the robot so that it can never harm anyone. Explain why this cannot be done, and state two real measures that would be used instead.