Chapter 1: Introduction to Robotics

In 1961, a machine at a General Motors plant in New Jersey began doing a job nobody wanted. Its task was to lift hot metal castings out of a die-casting machine and stack them — heavy, scalding, repetitive work that burned hands and ruined shifts. The machine was called Unimate. It was an arm roughly the size of an adult, built by George Devol and Joseph Engelberger from a patent Devol had filed seven years earlier. Its instructions were stored on a magnetic drum, and it repeated the same sequence of movements, hour after hour, without pause and without complaint.

The Unimate could not see. If a casting was not where the sequence expected it to be, the arm still closed on empty air and set nothing down. It knew nothing about the room around it, and nothing about the people in it. What it had was a body that could move accurately and a memory of what to do next.

Six decades later, an arm of similar size can watch a moving belt with a camera, tell one part from another, soften its grip for something fragile, and stop the moment a person steps too close. Between those two machines lies almost everything this chapter is about: what turns a mechanism into a robot, what sensing adds to it, and what changes when a machine begins to decide for itself.

Photograph

Fig. 1.0-1 – an early Unimate industrial arm at work in a factory: the heavy drum-shaped base, the single jointed arm and the gripper must all be visible, with the die-casting or press machine it serves in the same frame, so the reader sees a robot working as one part of a production line. An archival factory photograph from the 1960s is preferred. A studio or museum shot of the arm standing alone is usable only if no working-line image can be licensed, because it loses the industrial setting the story depends on; artists’ renderings, film stills and modern arms restyled to look vintage are not usable.

Curiosity Corner

Q1. A washing machine also follows a stored programme and works on its own. Why is it not usually counted as a robot? Q2. The Unimate could not sense anything around it. What exactly does a machine need before it can react to the world instead of ignoring it? Q3. When a robot is described as working “on its own”, how much is really left to the machine and how much to the person supervising it? Q4. What allows a modern arm to tell one part from another on a moving belt, and to change its grip to suit what it has picked up? Q5. Long before a robot is cut and assembled, engineers build it inside a computer. What can they learn there that the workshop cannot tell them?

Keep these five questions in mind as you read. The chapter answers each of them in turn, at the point where you have enough of the subject to see why the answer is what it is.

1.1 What a Robot Is

The word robot is used loosely in ordinary speech. A toy that walks across a table is called a robot, and so is a machine arm welding a car body, and so is a piece of software that answers messages on a website. Three very different things cannot all be the same kind of object.

Engineering cannot work with a word that loose. Before you can specify a machine, choose its parts or judge whether it is doing its job, you need to know what the machine is supposed to be. This section fixes the meaning, and it begins where the word itself began.

1.1.1 From Imagination to Machine

People imagined artificial workers long before anyone could build one. Water clocks that struck bells, mechanical birds, writing dolls driven by cams and springs — these automata fascinated their audiences for centuries. They moved, but they could not be told to do anything different. Each one repeated the single performance its maker had built into its metal.

The name we now use came from a play. In 1920, the Czech writer Karel Capek wrote R.U.R. (Rossum’s Universal Robots), in which manufactured beings are made to do humanity’s labour. He built the name from the Czech word robota, meaning forced labour. The play has no machinery in it that an engineer could copy, but it gave the idea a word, and the word spread faster than any invention.

Side Note

Karel Capek later said that the word was not his own. He credited it to his brother Josef Capek, a painter, who suggested it while Karel was struggling to name the artificial workers in his play.

Two decades later, the writer Isaac Asimov set out his Three Laws of Robotics in a 1942 short story: a robot must not harm a human, must obey human orders, and must protect itself, each law yielding to the one before it. It is worth being clear about what these are. They are a device in fiction, not a standard any real machine is built to. No controller anywhere carries them as instructions. Their value is that they made ordinary readers ask who is answerable for what a machine does — a question the subject still has to answer, and one you will return to in Class XII.

The engineering story runs alongside the literary one, and it is slower. Table 1.1 sets out the milestones that matter most.

Table

Table 1.1 – Milestones in the development of robots; the third column states what each milestone added that did not exist before it | Year | Milestone | What it added | |—|—|—| | 1920 | The word “robot” appears in Karel Capek’s play R.U.R. | A name for the idea of a manufactured worker, long before any such machine existed | | 1942 | Isaac Asimov states his Three Laws of Robotics in a short story | Public discussion of a machine’s obligations, though the laws are fiction and not engineering | | 1954 | George Devol files a patent for a programmed article transfer device | The idea of a machine whose sequence of movements is stored, and can be changed without rebuilding it | | 1956 | Devol and Joseph Engelberger found Unimation | The first company formed to build and sell robots as products | | 1961 | A Unimate arm is put to work at a General Motors plant | Industrial use: a stored sequence taking over hazardous manual handling | | 1969 | Victor Scheinman builds the Stanford Arm | An arm designed from the start to be driven by a computer rather than by mechanical stops and cams | | 1973 | KUKA builds FAMULUS | Among the first arms with six electromechanically driven axes, so a point could be reached from many directions | | 1978 | The SCARA arm is developed in Japan | An arm shaped for one task — fast, accurate assembly of parts fitted from above | | 1997 | The Sojourner rover drives on the surface of Mars | A robot working where an instruction from Earth arrives minutes after it is sent | | 2002 | The Roomba floor cleaner is sold for household use | Sensing and simple decision-making at a price ordinary households could meet | | 2004 | The first DARPA Grand Challenge is held | A long desert course for driverless vehicles; no entrant finished, which showed how hard open-world driving is | | 2013 | Boston Dynamics shows the Atlas humanoid | Walking and balance over uneven ground, a problem wheels avoid rather than solve | | 2023 | The Pragyan rover operates on the Moon after Chandrayaan-3 lands | A rover deployed and driven on the lunar surface by an Indian mission |

Read down the third column and a pattern appears. The first fifty years were spent on the body: joints, drives, and a way of storing a sequence of movements. The middle years put a computer in charge of that body. Only in the last three decades has the hard part been added — machines that take in something about the world around them and change what they do because of it. The arm in Fig. 1.1-1 is the descendant of everything in that table.

Photograph

Fig. 1.1-1 – a present-day six-axis industrial robot arm working on a production line, with the jointed structure from base to wrist clearly visible and the tool at the end of the arm in contact with the workpiece. The surrounding line equipment and safety fencing should be in frame, so the arm is seen as one machine among many rather than as an object on its own. Renderings, concept art and images of arms posed in showrooms are not usable; this must be a photograph of a working installation.

1.1.2 Defining a Robot

Look at what the machines in that table have in common, and three things stand out.

The first is that a robot acts physically on the world. It moves something, or moves itself. A program that sorts a list of names does useful work, but nothing in the room changes when it runs, and it is not a robot.

The second is that a robot is programmable. Its behaviour is fixed by instructions that can be changed without rebuilding the machine. This is exactly what the automata of earlier centuries lacked, and it is what Devol’s 1954 patent introduced. Change the stored sequence and the same arm stacks a different part.

The third is that a robot works from instructions, from what it senses, or from both. A sensor is a device that converts some physical quantity in the world — distance, light, temperature, force — into a signal the machine can use. A controller is the processor that reads those signals, applies the stored instructions and decides what to do. An actuator is the device that turns that decision back into physical motion, such as a motor. Sensors, controller and actuators are examined one at a time in Section 1.2.

Putting these together gives a definition you can work with.

Definition

A robot is a programmable machine that acts physically on its surroundings, working from stored instructions, from information supplied by its sensors, or from both, and carrying out its task with little or no direct human intervention while it runs.

Notice what the definition does not require. It does not require the machine to look human. It does not require intelligence. It does not even require sensors — the Unimate of 1961 had almost none, and it is still counted as the first industrial robot, because it was programmable and it moved real objects. Sensing is what makes a robot capable of reacting; it is not what makes it a robot.

Common Error

Assuming a robot must look human or must be clever What students do: describe robots as machines that “think like humans” or “have a human shape”, and then refuse to call an industrial arm or a floor cleaner a robot. What happens: half the subject disappears. Almost every robot in real use is an arm, a wheeled base or a flying platform, and most of them carry no intelligence beyond a few rules. The fix: judge a machine by what the definition asks for — physical action, programmability, and working from instructions or sensing. Shape and cleverness are design choices, not requirements.

This also settles a question that troubles most students at the start.

Curiosity Corner

Q1. A washing machine also follows a stored programme and works on its own. Why is it not usually counted as a robot?

A. Because its motion is fixed by its construction. A washing machine can be told how long to run and how fast to spin, but it can only ever turn one drum in one place. A robot’s instructions specify where to move and what to do there, so the same machine can be programmed for a different task without being rebuilt. That flexibility, not the presence of a timer or a start button, is what separates a robot from an ordinary automatic appliance. Table 1.2 sets the two side by side.

Compare Table

1.1.3 Why Robots Are Used

A robot is expensive to buy, to program and to maintain. It is worth that cost only for particular reasons, and knowing those reasons is how engineers decide whether a robot belongs in a task at all.

Work that is dull, and must be identical

Human attention drifts on the four-hundredth repetition of a task; a machine’s does not. A robot performs the same cycle at the same speed for an entire shift, which raises output and, just as importantly, makes every unit alike. The cost is that this advantage only appears above a certain volume. For a workshop making twenty different products a week, the time spent reprogramming and re-tooling can exceed anything the robot saves.

Work that is dangerous or unhealthy

Molten metal, paint spray, grinding dust, chemical tanks, live electrical equipment, structures at height — these are places where a person is at risk simply by being present. Putting a machine there removes the risk to the person entirely. The honest limitation is that the danger has not vanished; it has moved. A robot that fails inside a hazardous area still has to be recovered, and the recovery is the risky job that remains.

Work that demands precision beyond the hand

A robot can place a component within a fraction of a millimetre and return to the same point again and again, which is why electronics assembly and precision welding are so heavily automated. But precise is not the same as correct: a robot that has been taught a position slightly off target will hit that wrong position with beautiful consistency, thousands of times, until someone notices. You will study exactly how this is measured and specified in a later chapter.

Places a person cannot go

Some environments simply cannot hold a human being — the deep seabed, the inside of a pressurised pipe, a reactor building, the surface of another world. Here a robot is not an economic choice but the only way the work happens at all. Fig. 1.1-2 shows one such machine at work. The trade is severe: everything the operators know about the site arrives through the robot’s sensors and its communication link, and when either fails, the machine is beyond reach.

Photograph

Fig. 1.1-2 – a remotely operated underwater vehicle working below the surface, with its manipulator arm, lights, camera housings and tether cable all visible, and the structure it is inspecting or handling in the same frame. The tether must be clearly in shot, since it carries the power and the communication link the vehicle depends on. Surface shots of the vehicle sitting on a deck do not show the working situation and are not usable.

Support, care and access

Robots also do work whose value is not measured in output at all: assisting surgery, helping a patient regain movement after injury, carrying supplies through a hospital, giving a person with limited mobility a way to reach and grip. These uses raise questions that a factory arm does not — about consent, safety around people, and what is lost when care becomes mechanical — and they are treated seriously later in the course.

Set against all of this, robots remain poor at anything unfamiliar. A person handed an unexpected situation improvises; a robot handed one outside its instructions and its sensing usually does the wrong thing at full speed. Every honest reason for using a robot is a reason that fits a task which is well defined, repeated, hazardous, or out of human reach — and every deployment changes the work that the people around it do.

Try Yourself

List five machines in your school or home that run on their own once switched on — a fan regulator, a printer, a lift, a water pump with a float switch, and one of your own choosing. For each, decide using Table 1.2 whether it is a fixed automatic machine or a robot, and write one line giving the reason that decided it.

1.2 How a Robot Works

Open up any robot — a floor cleaner, a welding arm, a rover on another planet — and you find the same five things. It takes in information about the world or about itself. It works out what to do. It moves. It draws energy from somewhere. And its behaviour is fixed by a program that a person wrote.

These five are not a list to be memorised. They are a loop, and the loop is what makes a robot different from a machine that simply runs. Sensing produces a reading; the reading changes a decision; the decision changes the motion; the motion changes the world; and the next reading is taken from the world the robot has just altered. Fig. 1.2-1 shows the whole cycle.

Diagram

Fig. 1.2-1 – flow: the surroundings → sensors → controller (drawn holding a stored program) → actuators → movement of the robot or of an object → change in the surroundings, with the last arrow returning to the sensors block to close the loop. A separate power source block sits below, with arrows going up to the sensors, the controller and the actuators. Label the arrows: signal from sensors to controller, command from controller to actuators, power from the source to each block.

Theory Snapshot

The sense–decide–act loop Statement: A robot operates by repeating a cycle in which its sensors report the current state of itself and its surroundings, its controller applies stored instructions to that report and selects an action, and its actuators carry that action out, altering the state that the next cycle will read. In words: look, decide, move — then look again at the situation you have just changed. Why it holds: an action taken on old information is an action taken blind. Because the world changes while the robot works on it, and because the robot’s own motion is part of that change, the reading must be repeated after every action. Used for: organising the design of any robot, and for locating a fault — a robot behaving wrongly is failing at sensing, at deciding, or at acting, and separating the three is the first step of debugging. Watch out: the loop takes time. If a cycle is slow, the robot acts on a picture of the world that has already gone out of date, which is why a fast-moving machine needs a fast loop rather than a clever one.

1.2.1 Sensing the Surroundings

A sensor converts a physical quantity into an electrical signal. That is all it does, and remembering it will save you from a great deal of confusion later. An ultrasonic sensor does not detect a wall; it produces a number related to the time an echo took to return. Reading that number as “a wall 30 cm away” is the controller’s work, not the sensor’s.

Robots sense in two directions. Some sensors look outward, at the world: distance to an object, the brightness of a surface, the presence of a person. Others look inward, at the robot’s own state: how far a joint has turned, how fast a wheel is spinning, whether the body is tilting. A robot with only outward sensing knows where things are but not where its own arm is, and it cannot work reliably.

Table 1.3 lists sensors you will meet through this course, with the quantity each one actually measures. You will study how each of them works, and where each one fails, in a later chapter.

Table

The abbreviations are worth expanding once. Infrared (IR) light lies just beyond the red end of the visible range; light detection and ranging (LIDAR) works like an ultrasonic sensor but sends out light instead of sound, and sweeps it across a scene.

Curiosity Corner

Q2. The Unimate could not sense anything around it. What exactly does a machine need before it can react to the world instead of ignoring it?

A. Three things, and all three are needed together. First, a sensor that converts the relevant quantity into a signal — for an obstacle, something that measures distance. Second, a controller that reads that signal and compares it against a stored rule, such as “if the distance is under 20 cm, stop”. Third, an actuator that can carry out the different action the rule calls for. The Unimate had the third but not the first two, so nothing in the world could ever alter what it did next.

Exam Corner

Q. Why do robots carry several different sensors instead of one very good one?

A. Because each sensor converts a different quantity, and each fails in its own way. An ultrasonic sensor is confused by a soft curtain that absorbs its pulse; a camera is defeated by darkness; an infrared sensor is unreliable against a black surface that reflects almost nothing. A robot carrying three sensors that fail under different conditions still has something to work with when one of them goes quiet.

A sensor’s honest limitation is that it reports, and nothing more. It cannot tell you that a reading is wrong. Dust on a lens, a loose connector, a hot day, or an unusual surface will all produce a perfectly confident number that happens to be false, and the controller will act on it.

1.2.2 The Control System

The controller is where the robot’s behaviour actually lives. It reads the sensor signals, applies the stored program, and issues commands to the actuators — thousands of times a second in a fast machine.

In a school robot the controller is usually a microcontroller: a single chip carrying a processor, memory and input-output pins, small enough to run from a battery and cheap enough to build into anything. The board you will use in the laboratory is built around one. At the other end of the range, a driverless car carries several powerful computers. The scale is enormously different; the job is the same.

What the controller does with a reading is decide, and deciding at this level means comparing. A value arrives, it is compared against a threshold or against a target, and one branch of the program is taken rather than another. Nothing more mysterious than that is happening inside most working robots, including many that look impressively lifelike.

Two limits follow directly, and they explain most robot failures you will ever see. The controller knows only what the sensors report — a robot cannot avoid an obstacle it has no way of detecting. And the controller can only choose among actions the program provides — a situation the programmer did not anticipate leaves the machine doing whatever its last instruction said, which is often exactly the wrong thing.

1.2.3 Actuation, Mobility and Manipulation

An actuator converts the controller’s electrical command back into physical motion, closing the loop between decision and effect. Most robots use electric motors, because they are compact, controllable and run directly from a battery. Hydraulic actuators, driven by pressurised fluid, are used where very large forces are needed, and pneumatic actuators, driven by compressed air, where a simple fast push is enough. The motors used in robotics are studied in detail later in the course.

What the robot does with that motion divides robots into two broad kinds. A robot with mobility moves itself — on wheels, on tracks, on legs, or through air or water. A robot with manipulation stays where it is and moves other things, using an arm made of rigid links joined at movable joints. Many machines do both: a rover drives to a rock and then reaches out to it.

At the far end of an arm is the part that actually touches the work.

Definition

An end-effector is the device fitted at the free end of a robot arm that performs the task on the workpiece — a gripper, a welding torch, a suction cup, a drill or a surgical instrument — as distinct from the links and joints that position it.

The distinction matters because the arm and the end-effector are chosen separately and for different reasons. The arm decides where the tool can be placed; the end-effector decides what can be done there. The same arm fitted with a gripper stacks cartons and fitted with a spray head paints panels.

Fig. 1.2-2 shows how these parts sit together in a small wheeled robot of the kind you will build.

Diagram

Fig. 1.2-2 – layout of a small two-wheeled robot on an acrylic chassis, drawn from above with the plate shown as transparent. Label: two motors with wheels at the rear; a castor at the front; the motor driver module; the controller board; an ultrasonic sensor mounted facing forward at the front edge; the battery pack at the centre. Draw thin lines from the sensor to the controller and from the controller to the motor driver, and thicker lines from the battery to the motor driver and to the controller, with a legend marking the thin lines “signal” and the thick lines “power”.

Actuators bring their own limits, and they are physical rather than logical. An arm can reach only so far and lift only so much. Nothing starts or stops instantly, because links and loads have mass, so a fast arm overshoots unless it is slowed in advance. And an actuator strong enough to be useful is strong enough to injure, which is why the machines in Fig. 1.1-1 work behind fencing.

1.2.4 The Energy Source

Every block in Fig. 1.2-1 needs power, and where that power comes from shapes the whole design. A robot can be tethered — supplied through a cable from the mains supply, like a factory arm bolted to the floor — or it can carry its energy on board in a battery, like a floor cleaner or a drone. Spacecraft add a third option, generating power from sunlight through solar panels, because no other supply is available.

The choice is a real engineering trade, not a detail.

Compare Table

The last row of that table hides a circular problem worth noticing now. A mobile robot that needs to run longer carries a bigger battery; the bigger battery is heavier; the heavier robot needs more energy to move; and some of the extra capacity is spent carrying the extra capacity. Battery types, their characteristics and how to calculate a robot’s power requirement are taken up in a later chapter.

Safety Note

The lithium polymer packs used in the school laboratory store a large amount of energy in a soft foil case. Never puncture, bend, crush or short the terminals of a pack, never charge one unattended or on a flammable surface, and never continue to use a pack that has swollen or been dropped hard. Charge only on the charger provided, under supervision.

1.2.5 Programming and Intelligence

The last of the five is the one that decides everything the other four are used for. A robot’s hardware sets what is possible; its program sets what actually happens.

Programs used in robots sit at three broad levels, and it is useful to see them as a progression.

The first is a fixed sequence. The robot performs stored movements in order, taking no account of its surroundings. The Unimate worked this way, and so does a great deal of industrial equipment today, because when the parts always arrive in the same place a fixed sequence is fast, cheap and completely predictable.

The second is rule-based reaction. The program reads sensors and chooses among stored responses: if the distance ahead falls below a threshold, stop and turn. Almost every robot you will build in this course works at this level, and so does most equipment in use worldwide. It is not primitive — it is what makes behaviour reliable enough to trust.

The third is learned behaviour, where the response is not written out by the programmer but derived by the machine from large quantities of data. This is where artificial intelligence enters robotics, and it is taken up in Sections 1.3 and 1.5.

Common Error

Describing a sensing robot as one that “thinks” or “understands” What students do: write that a robot “sees the wall and understands that it must turn”, or that it “knows” where it is. What happens: the explanation cannot then account for the robot’s failures, which are the interesting part. A machine that understood a wall would not drive into a glass door or a black cloth. The fix: describe what the machine actually does — a value is measured, compared against a stored threshold, and a branch of the program is taken. The behaviour can be sophisticated without any understanding behind it, and saying so precisely is what lets you predict where it will break.

Take away any one of the five and the loop stops. Without sensing, the robot cannot react; without a controller, nothing decides; without actuators, nothing happens; without power, nothing runs; without a program, the hardware sits still. Every chapter that follows is a closer look at one of these five, and every robot you meet later is an arrangement of them, chosen for a particular job.

1.3 Autonomous and Semi-Autonomous Operation

Two robots can be doing the same job and yet be completely different machines. One is being steered, moment by moment, by a person watching a screen. The other was told where to go an hour ago and has been working out the route for itself ever since. Both arrive at the same place. What separates them is not what they do, but how much of the deciding happens inside them.

That question — how much decision-making sits in the machine — runs through the whole subject. It decides which sensors the robot needs, how much computing it must carry, what happens when the link to the operator drops, and who is answerable when something goes wrong.

1.3.1 Levels of Human Involvement

It helps to separate a task into three kinds of decision, because different machines divide them differently.

There is the goal: clean this floor, inspect that weld, drive to that rock. There is the plan: which route, which order, which approach. And there is the moment-to-moment motion: turn the left wheel a little faster now, stop now, close the gripper now.

A machine under direct human control leaves all three with the person. A machine at the other extreme is given the goal and works out the plan and the motion by itself. Between the two lies everything interesting.

Definition

An autonomous robot is one that carries out its assigned task without human intervention during operation, using its own sensors, stored program and decision-making to plan its actions and control its motion.

Definition

A semi-autonomous robot is one that performs part of its task on its own but depends on a person for the remaining decisions — typically for the goal, for supervision, or for approving critical actions.

Read those two definitions carefully, because the qualifying phrases carry the meaning. “Without human intervention during operation” does not mean without a human anywhere. Somebody chose the goal, wrote the program, positioned the machine and can stop it. A rover on Mars is autonomous in exactly this sense while it drives, and entirely directed by people in every other sense.

Common Error

Reading “autonomous” as “nobody is involved” What students do: write that an autonomous robot “works completely by itself and needs no human at all”. What happens: the statement cannot be defended, because every autonomous machine in existence was given its task, its rules and its limits by people, and can be commanded to stop. The fix: say what is actually autonomous — the planning and the moment-to-moment control during the task. Autonomy is about which decisions are delegated, not about the absence of people.

Compare Table

The clearest case for full autonomy is a machine that cannot be steered in time. A rover on Mars, of the kind shown in Fig. 1.3-1, is separated from its operators by a radio path that takes several minutes each way, so a command and its reply are typically eight to forty-eight minutes apart depending on where the two planets are in their orbits. A person watching for a wheel to slip could do nothing about it; by the time the picture arrived on Earth, the rover would already have dug itself in. The delay, not ambition, is what forces the deciding into the machine.

Photograph

Fig. 1.3-1 – a planetary rover on a surface, either a mission photograph from Mars or the Moon or an engineering model under test on a rough terrain yard. The wheels, the mast carrying the cameras and the solar panels or instrument deck must all be visible, and the surrounding ground should be visibly uneven, since the terrain is the reason the machine must decide for itself. Artists’ impressions and film imagery are not usable; a mission or agency photograph is required.

Curiosity Corner

Q3. When a robot is described as working “on its own”, how much is really left to the machine and how much to the person supervising it?

A. Usually less than the phrase suggests. The person sets the goal, writes or selects the program, defines the limits the machine may not cross, and retains the power to stop it. What the machine holds is the planning and the moment-to-moment control while the task runs. In a semi-autonomous system the machine holds even less: a surgical system holds the steadiness of the instrument while the surgeon holds every decision about where it cuts.

1.3.2 Sensors as the Robot’s Means of Perception

Section 1.2 established that a sensor produces a number. Autonomy needs something more than numbers. Before a robot can plan a route, it must hold some picture of what is around it — where the walls are, where the gap is, which way it is facing.

Building that picture from raw readings is called perception. It is a step of processing that sits between the sensors and the decision, and it is where most of the difficulty of autonomy actually lies.

The picture is built by combining sensors rather than trusting any one of them, a technique called sensor fusion. Consider a small robot working along a corridor, as in Fig. 1.3-2. Counts of wheel rotation tell it roughly how far it has travelled, but wheels slip on a dusty floor, so the estimate drifts. A gyroscope tells it how much it has turned, which corrects the heading. An ultrasonic sensor reports the distance to the wall on the right, which corrects the position across the corridor. No one of these three is sufficient; together they hold a usable estimate.

Diagram

Fig. 1.3-2 – a small wheeled robot part-way along a straight corridor, drawn from above. Show three sensing inputs on the left half: a cone spreading forward from the ultrasonic sensor to the wall ahead, marked “distance ahead”; a short arc at the wheels marked “wheel rotation counted”; a curved arrow at the body marked “turn rate from gyroscope”. On the right half, in a bordered panel headed “the robot’s internal picture”, draw the same corridor as straight lines with the robot’s estimated position and heading marked by an arrow, and a dashed line showing the planned path through the gap ahead. An arrow labelled “perception” runs from the left half to the panel.

Exam Corner

Q. If two sensors disagree, which one should the robot believe?

A. Neither, on its own. A well-designed system knows what each sensor is good for and weights the readings accordingly — trusting the gyroscope for heading and the wheel counts for distance, for instance, rather than asking either to do the other’s job. Where two sensors report the same quantity and disagree badly, the sensible response is not to pick a winner but to treat the disagreement as a fault signal and slow down or stop.

The limitation is blunt and worth stating plainly: perception can only be as good as the sensing beneath it. A robot whose sensors cannot detect glass has no picture of a glass door, and no amount of processing will invent one. Every autonomous failure you read about begins here far more often than it begins in the decision-making.

1.3.3 How Artificial Intelligence Enables Autonomy

Artificial intelligence (AI) is the field concerned with making machines perform tasks that would ordinarily be said to require human intelligence — recognising objects, understanding speech, planning, and choosing between options under uncertainty. Machine learning (ML) is the part of AI in which a system derives its own rules from data instead of being given them by a programmer.

In an autonomous robot, AI does its work at five points along the loop of Section 1.2.

Perception and interpretation. Algorithms turn raw sensor data into recognised things: this region of the camera image is a person, that return from the LIDAR is a wall and not rain.
Decision-making. Given the interpreted scene, the system selects an action — slow down, go around to the left, wait.
Planning and path-finding. Route-finding algorithms, of which Dijkstra’s algorithm and A* are the best known, compute a path from the present position to the goal that avoids the obstacles the perception step has placed on the map.
Control and execution. The chosen path is converted into actual commands to the motors, moment by moment.
Learning and adaptation. Some systems adjust their own behaviour over time from the outcomes of past attempts, improving without being reprogrammed.

Only the last of these five requires machine learning. The first four can be, and very often are, done with ordinary rules and arithmetic written by an engineer. This matters more than it may seem. A great deal of successful autonomy in factories and warehouses contains no learning at all, because the surroundings are controlled enough that fixed rules cover every case. AI earns its place where the surroundings cannot be controlled — an open road, an unmapped field, a room full of people.

There is a cost to buying that ability. A rule-based robot can be read: you can look at the program and say exactly what it will do at 20 cm. A system that has learned its behaviour from data cannot be read the same way, so predicting its response to a situation it has never met is genuinely hard. It also needs large quantities of data to learn from, and it will reproduce whatever was wrong or missing in that data.

1.3.4 What Autonomy Gains and What It Costs

Autonomy is worth its complexity for four reasons.

It removes fatigue, so the thousandth cycle is performed exactly like the first. It takes people out of places that are dangerous to occupy. It scales — a fleet of self-directed vehicles in a warehouse needs no fleet of operators, so adding a machine costs only the machine. And it works where supervision is impossible, whether because the delay is too long, as on Mars, or because events happen faster than a person can respond.

Against that stand four costs, and a serious engineer holds both lists at once.

Sensors fail, and they fail quietly. A dirty lens or a wet sensor produces confident wrong readings that the machine has no way to doubt, and an autonomous system acts on them immediately and at full speed. A supervised system has a person who might notice.

Environments are unpredictable. A rule set or a learned model built for one situation meets an unfamiliar one — an unusual obstacle, an unexpected surface, a person doing something no one anticipated — and there is no guarantee about what it will do.

Responsibility becomes harder to place. When a machine that chose its own action causes harm, the question of who is answerable — the operator, the owner, the programmer, the manufacturer — has no obvious answer, and different people argue it differently. This is treated properly in Class XII, where you will be asked to weigh the arguments rather than accept one.

And a system that decides for itself must be protected. If its sensor inputs, its instructions or its communication link can be interfered with, its behaviour can be interfered with too, which is why safety and security are designed into an autonomous system from the start rather than added afterwards.

Try Yourself

Take one task from your school — carrying trays in the canteen, or watering plants along a corridor. Write down the three kinds of decision from Section 1.3.1 for that task, then decide which of them you would hand to a machine and which you would keep with a person. Give one reason for each choice, and name the sensor that the machine would need for any decision you have handed over.

1.4 Types of Robots

There is no single correct way to sort robots, and looking for one is a waste of effort. Engineers use several classifications side by side, because each answers a different question.

Ask what a robot is for, and you classify by application. Ask how much of the deciding it does, and you classify by control mechanism. Ask whether it moves and how, and you classify by mobility. Ask what shape its body takes and how its joints are arranged, and you classify by mechanical structure. The same machine appears in all four schemes at once, and that is not a flaw in the schemes.

1.4.1 Classification Based on Application

This is the classification used most often, because it is the one a customer thinks in. It groups robots by the work they are bought to do. Table 1.6 sets out the categories, with the demand each setting places on the machine — which is the part worth learning, since the demand is what drives the design.

Table

Two rows deserve a note. Military robots are listed here because they are a real and substantial part of the field; their capabilities and the arguments they raise are examined in Class XII, where the ethical questions are set out as competing positions rather than settled ones. And medical robots sit deliberately in the semi-autonomous band of Section 1.3 — a surgical system holds the instrument steady, but the surgeon decides every cut.

1.4.2 Classification Based on Control Mechanism

Section 1.3 established the underlying idea: what varies is how much of the deciding lives in the machine. The standard categories follow from it directly.

A manual or remote-controlled robot is directed by a person throughout. Every movement corresponds to something the operator does with a joystick, a switch or a handset, and the machine stores no plan of its own.

A semi-autonomous robot performs parts of its task by itself but relies on a person for the rest, usually for the goal or for approving critical actions.

A fully autonomous robot is given a goal and carries it out without intervention while it runs, planning and controlling its own motion from what it senses.

A tele-operated robot is controlled from a distance, with the operator working from visual or audio feedback returned by the machine.

The last of these is worth separating carefully from the first, because the two are often confused. A remote-controlled machine may be operated in plain sight, with the person watching the machine itself. In tele-operation the person cannot see the machine at all and works entirely through what its cameras and microphones send back — inside a pipe, down a mine shaft, under water, or in another room of a hospital. That changes the engineering completely: the communication link becomes a critical component, its delay becomes part of the control problem, and losing it means losing the machine.

1.4.3 Classification Based on Mobility

Robots may also be grouped by whether they move and by the medium they move through.

Stationary robots are fixed in place, usually arms bolted to a floor or a frame beside the work they serve. Mobile robots move through their surroundings on wheels, legs or tracks. Aerial robots — drones — fly. Underwater robots operate submerged, either on a tether from a surface vessel or under their own direction. Humanoid robots are built to resemble the human body and to move as it does.

The last category shows something useful about classification in general.

Common Error

Treating the classification schemes as mutually exclusive What students do: ask which single category a machine “really” belongs to, and treat a humanoid as though it could not also be a service robot. What happens: the description of any real machine falls apart. A planetary rover is an exploration robot by application, fully autonomous by control, mobile by mobility, and articulated in the structure of its instrument arm — all at once. The fix: state which classification you are using before you place a machine in it. Each scheme sorts by a different property, so the same robot correctly occupies one place in each.

1.4.4 Classification Based on Mechanical Structure

This classification is the most technical of the four and the most useful to a designer, because a robot’s structure decides the space it can actually reach.

Definition

The work envelope of a robot is the complete set of points its end-effector can reach, given the arrangement of its links and joints and the limits of travel at each joint.

The work envelope is fixed by the structure and cannot be programmed around. An arm that moves only along three straight axes sweeps out a box; an arm whose first joint rotates about a vertical axis sweeps out a shape with a round outline. Choosing a structure is therefore choosing the shape of the space you can work in, before any program is written. Fig. 1.4-1 compares the main arrangements.

Diagram

Fig. 1.4-1 – four arm structures drawn side by side, each with its work envelope shaded behind it. (a) Cartesian: three straight sliding joints at right angles, arrows along each, envelope drawn as a rectangular box. (b) Cylindrical: a vertical rotating base joint with a curved arrow, a vertical sliding joint and a horizontal sliding joint, envelope drawn as a hollow cylinder. (c) Spherical or polar: a rotating base, a pivoting shoulder joint and one sliding telescopic link, envelope drawn as part of a hollow sphere. (d) Articulated: a rotating base with three pivoting joints in series, envelope drawn as an irregular rounded volume. In every drawing, hatch the fixed base, mark each joint with a circle for a pivot and a short double line for a slide, and label the links in order from base to end-effector.

Table

The abbreviation is worth expanding: Selective Compliance Assembly Robot Arm (SCARA) describes exactly what the design does. The arm is stiff in the vertical direction, so it can press a part down firmly, but it gives slightly in the horizontal plane, which lets a peg find a hole instead of jamming against its edge. That single property is why so much electronic assembly uses this shape.

Delta robots take the opposite approach to weight. Instead of one heavy arm carrying its own motors along with the load, the motors stay on the fixed frame overhead and drive three light arms that meet at the tool. Very little mass has to be accelerated, so the tool can be moved and stopped extremely quickly — which is why they appear over conveyor belts, as in Fig. 1.4-2.

Photograph

Fig. 1.4-2 – a delta robot working above a moving conveyor in a packaging or food line, mounted from an overhead frame, with the three arms converging on the tool and products visible on the belt below. The overhead mounting and all three arms must be in frame, since the point of the photograph is that the motors stay on the frame and only the light arms move. Renderings and exhibition-stand photographs without a working line are not usable.

1.4.5 Hybrid and Specialised Types

Some robots do not sit neatly in any of the four schemes, either because they combine categories or because they represent a newer line of work.

Swarm robotics uses many small, simple, identical robots that coordinate with one another instead of one large capable machine. No individual holds the plan; the useful behaviour appears from the local rules each one follows and from what they sense of their neighbours. The attraction is robustness — losing one unit out of fifty barely matters — and the difficulty is that behaviour arising from many interactions is hard to predict and hard to guarantee.

Bio-inspired robots borrow a solution from a living organism: a snake-like body that moves through rubble by pushing against its surroundings, or a legged design that walks where wheels would sink.

Soft robots are built from flexible materials rather than rigid links. A soft gripper conforms to whatever it holds, so it can lift a ripe tomato or a hen’s egg without the force control a rigid gripper would need, and it is inherently safer in contact with a person. Fig. 1.4-3 shows one at work. The cost is precision: a body that yields cannot hold a position as exactly as a rigid one, and its behaviour is much harder to model.

Photograph

Fig. 1.4-3 – a soft robotic gripper picking up a delicate item such as a fruit, an egg or a piece of laboratory glassware, close enough that the flexible fingers are seen bending around the object’s shape. The contact between the soft fingers and the object must be clearly visible, since that conformity is the whole point. Images of rigid metal grippers and product photographs of the gripper alone, with nothing held, are not usable.

Modular robots are assembled from interchangeable units that can be rearranged into a different machine for a different job — a wheeled base one week, a climbing chain of segments the next.

What these four have in common is that each gives up something a conventional robot has. Swarms give up predictability, bio-inspired designs give up mechanical simplicity, soft robots give up precision, and modular robots give up the rigidity of a purpose-built frame. Each is chosen when the thing it gains matters more than the thing it loses, which is the shape of nearly every engineering decision in this subject.

1.5 New Age Robotics Systems

The robots of the twentieth century were built to repeat. They were placed in surroundings arranged to suit them — parts arriving in the same orientation every time, fences keeping people out — and within those arrangements they were superb.

The machines described in this section work the other way round. Instead of the surroundings being controlled to suit the robot, the robot is built to cope with surroundings that are not controlled: a belt where the parts arrive at any angle, a factory aisle with people walking through it, a road. These are called new age robotics systems, and what they have in common is that the flexibility has moved out of the fixture and into the software.

1.5.1 Defining Features of New Age Robotics

A new age system typically shows several of the following features. Very few show all of them.

Artificial intelligence is built into the machine, so it can recognise objects and speech and select an action rather than following a fixed list. It operates on its own for long stretches, navigating surroundings that were never surveyed for it. It can work safely alongside people in a shared space instead of behind fencing — a collaborative robot, or cobot. It improves with use, adjusting its behaviour from the outcomes of earlier attempts. It can be monitored and commanded from anywhere, because it is connected to a network rather than to a single control panel. And in some designs, many small units coordinate rather than one large one, the swarm approach already described in Section 1.4.5.

Notice that every feature in that paragraph is about handling variation. That is the single thread running through the whole idea.

1.5.2 The Technologies That Make Them Possible

None of these features come from robotics alone. Each rests on a technology developed largely outside the field and brought into it. Table 1.8 names them, together with what each costs — because a table of benefits with no costs teaches nothing about choosing.

Table

Two of those rows pull against each other and are worth reading together. Cloud computing offers a robot far more processing than it could ever carry, but every request travels to a server and back. Edge computing keeps the processing on board, where it is limited but immediate. Real systems split the work: anything that must happen within milliseconds — stopping for an obstacle — stays on the robot; anything that can wait — analysing a week of performance data — goes to the cloud.

1.5.3 Advanced Artificial Intelligence and Machine Learning

Section 1.3.3 introduced artificial intelligence and machine learning. What is meant by advanced AI is not a different technology but a shift in what the system can be asked to do: understand data it was not shown before, hold some awareness of context, interact through speech or images rather than through a keypad, and adapt without being reprogrammed.

The capability rests on a handful of methods you should be able to name. Deep learning uses artificial neural networks with many layers to handle large and complicated data. Computer vision recognises and interprets what a camera sees. Natural language processing (NLP) interprets and generates human language. Reinforcement learning arrives at a strategy by trying actions and being rewarded or penalised for the outcome.

Underneath all of them sits the same four-step process, shown in Fig. 1.5-1.

Diagram

Fig. 1.5-1 – flow: collected data → training, where an algorithm builds a model from the data → evaluation, where the model is tested on data it has not seen → deployment, where the trained model makes predictions on new inputs, with an arrow returning from deployment back to collected data to show new experience being fed in. Label the arrow from training to evaluation “model”, and the arrow into deployment “trained model”. Mark the evaluation block with a side note “if performance is poor, return to training”.

Theory Snapshot

Machine learning Statement: A machine learning system derives its own decision rule from a body of example data, by adjusting an internal model until its outputs on that data are close to the correct ones, and then applies that rule to inputs it has not seen before. In words: instead of being told the rule, the machine is shown many examples and works the rule out. Why it holds: many tasks are easy to demonstrate and very hard to specify. Nobody can write down the rule that distinguishes a cracked casting from a sound one, but ten thousand labelled photographs of each can be supplied. Used for: recognising objects, classifying defects, predicting failures, interpreting speech, and choosing actions where the correct choice was never written down. Watch out: the model can only be as good as its examples. Trained on one factory’s lighting, it will fail under another’s, and it gives no warning when it does — it simply returns a confident wrong answer.

That last point separates the two ways of building behaviour, and the distinction matters when you choose between them.

Compare Table

Curiosity Corner

Q4. What allows a modern arm to tell one part from another on a moving belt, and to change its grip to suit what it has picked up?

A. Two separate capabilities working together. A camera above the belt supplies images, and a computer vision model — trained beforehand on many labelled examples — identifies which part each object is and where it lies, giving the arm a target it was not programmed with in advance. The grip is a second matter entirely: force sensing in the gripper reports how hard it is squeezing, and the controller closes only until that force reaches the value stored for that part, so a carton and an egg are held with quite different pressures. Recognition comes from learning; the gentle grip comes from sensing and control.

1.5.4 Applications in Smart Manufacturing

Manufacturing has automated for longer than any other sector, so it shows most clearly what advanced AI adds to automation that already existed. Four uses cover most of it.

Predictive maintenance replaces fixed servicing schedules. Machines are fitted with sensors for vibration, temperature and sound, and a model trained on past failures learns what the signature of an approaching failure looks like. The machine is then serviced shortly before it would have failed, rather than on a calendar or after a breakdown has stopped the line.

Quality inspection uses computer vision. A camera photographs every unit and a trained model flags defects — a missing solder joint on a circuit board, a mismatched colour, a part fitted the wrong way round. Inspecting every unit rather than a sample becomes affordable, which is the real change.

Process optimisation uses records of past production to recommend settings — temperature, pressure, feed rate — that raise yield or cut waste, adjusting them as conditions drift instead of holding one setting all day.

The fourth is the one that changed the shape of the factory floor itself.

System Profile

Collaborative robot (cobot) Working principle: a cobot is a robot arm designed to operate in the same space as people without fencing. It runs at limited speed and force, senses contact or proximity, and stops or retreats when it detects a person or an unexpected obstruction. Its motors are sized and its surfaces shaped so that a collision at its permitted speed does not injure. Components: a lightweight jointed arm; sensors for force, torque or proximity at the joints; a controller that continuously checks speed and force against safety limits; an end-effector chosen for the task; a power supply. Advantages: it can be placed beside an existing workstation without rebuilding the cell, so a small workshop can automate one step at a time; it is quick to reprogram, often by guiding the arm by hand through the required motion; and it lets a person and a machine share a task, each doing the part they are better at. Limitations: the limits that make it safe also make it slow and light. A cobot cannot match a fenced industrial arm for speed or payload, it costs more per unit of work done, and it is only safe within the operating limits its installation was assessed for — raising its speed for productivity removes the very property that allowed the fence to be left out. Where it is used: small-batch assembly, machine tending, packing, laboratory sample handling, and testing lines where the product changes often.

Fig. 1.5-2 shows the arrangement that results, and it is worth looking at carefully, because the absence of the fence is the whole point.

Photograph

Fig. 1.5-2 – a collaborative robot arm working at a bench directly alongside a person, with no fencing or light curtain between them, both handling parts of the same task. The person’s hands and the arm must be within reach of each other in the frame, and the arm should be the lightweight jointed type. Photographs showing a caged industrial arm, or an operator standing well away from the machine, do not show collaborative operation and are not usable.

The larger framework these four uses belong to — connected machines, data-driven factories, and the place of people within them — is studied in Class XII under smart manufacturing and Industry 4.0.

1.5.5 Applications in Healthcare

Healthcare is the setting where the semi-autonomous idea of Section 1.3 matters most, because the consequences of a wrong decision fall on a patient.

In imaging and diagnosis, models trained on large collections of labelled scans highlight regions that may show disease, drawing a radiologist’s attention to them. The system does not diagnose. It sorts and flags, and a clinician decides — which is both an ethical position and a practical one, since the model is confident even when it is wrong.

In surgery, robotic systems such as the da Vinci system translate a surgeon’s hand movements at a console into very small, steady movements of instruments inside the patient. The machine removes tremor and scales the motion down; every decision about where the instrument goes remains with the surgeon.

In rehabilitation and monitoring, powered wearable devices support a patient’s limb during therapy, and networked sensors track heart rate, movement or glucose continuously so that deterioration is noticed early rather than at the next appointment.

The gain in each case is precision, consistency and reach — a specialist’s attention extended further than it could otherwise go. The costs are equally real: high price, the need for trained staff, and the question of responsibility when a recommendation from a model turns out to be wrong. Class XII takes up healthcare robotics as a full unit, and takes up that question with it.

1.5.6 Applications in Autonomous Vehicles

A driverless vehicle is the hardest of these three cases, because the surroundings are shared with people who are not following its rules.

It begins with sensing, and no single sensor is enough. Cameras read colour, lane markings and signs but are weakened by darkness and glare. LIDAR measures distances precisely in three dimensions but is degraded by heavy rain and fog. Radar sees through weather and measures speed directly but resolves shapes poorly. Ultrasonic sensors cover the last metre or two, for parking. Fig. 1.5-3 shows how the coverage of these sensors is arranged so that the weaknesses do not line up.

Diagram

Fig. 1.5-3 – a car drawn from above at the centre, with the coverage of each sensor type shaded around it: a long narrow forward cone labelled “radar, long range”; a wide forward cone labelled “camera”; a full circular sweep around the vehicle labelled “LIDAR”; and short arcs close to the front and rear bumpers labelled “ultrasonic, parking range”. Mark the overlapping region ahead of the vehicle and label it “overlap: more than one sensor covers this area”. Add a legend giving the quantity each sensor measures.

The readings are then fused into one picture of the scene, and the vehicle must locate itself within it. Satellite positioning through the global positioning system (GPS) gives a rough location but is lost in tunnels and degraded between tall buildings, so it is combined with sensor readings using simultaneous localisation and mapping (SLAM), in which the vehicle builds a map of its surroundings and works out its position within that map at the same time.

Only then can the vehicle plan. A route is computed to the destination, obstacles are avoided, and — the genuinely hard part — the likely behaviour of other road users is predicted, so that the vehicle acts on where a pedestrian will be rather than where they are. The chosen path is finally converted into steering, braking and acceleration commands, smoothly enough that a passenger is not thrown about.

Two honest limits close this section. First, performance falls where the training and the design assumed conditions that are not present — unfamiliar road markings, an unusual vehicle, an obstruction nobody anticipated. Second, driving automation exists in degrees rather than as a single switch, and many vehicles described loosely as self-driving are semi-autonomous systems that assume a person is watching and ready to take over. Reading such a system as fully autonomous is not a harmless mistake; it is precisely the mistake that causes accidents.

1.6 Components of Robots: From Visualisation to Creation

Every component in Section 1.2 had to be made by somebody. The sensor, the controller board, the motor and its gears, the frame they are all bolted to — none of these are natural objects. Each began as an idea, was drawn and tested inside a computer long before any material was cut, and was then manufactured to dimensions measured in fractions of a millimetre.

That path from idea to part is the subject of this section. It matters for a practical reason: the accuracy a robot can achieve, the loads it can carry and the reliability it shows in service are all decided at the design and manufacturing stage. No amount of clever programming recovers a joint that was made with slack in it.

1.6.1 System Visualisation

A robot is difficult to hold in the head. It has a mechanical structure, a set of electrical connections, a flow of signals and a program, and each of these behaves differently. Describing all of that in words alone is nearly impossible, which is why engineers draw.

Definition

System visualisation is the graphical representation of a system’s components, its structure, its behaviour and the interactions between its parts, used to understand, analyse, communicate about and improve that system.

Visualisation serves five distinct purposes, and it is worth separating them because a drawing made for one purpose is often useless for another.

It supports design understanding, letting an engineer see the layout and how the parts interconnect. It supports debugging and testing, since a simulated system can be run and observed before anything physical exists. It supports communication, so that a manager, a customer or a technician can follow what the system does without reading the code. It supports documentation, leaving a record of how the system is built for whoever maintains it in three years’ time. And it supports optimisation, because bottlenecks and wasted motion are far easier to see in a picture than to find in a table of numbers.

The forms it takes are equally varied: block diagrams for architecture, flowcharts for sequences, simulation models for behaviour under different conditions, three-dimensional models and schematics for the physical system, and data dashboards for watching a running machine. You have already read several of them in this chapter — Fig. 1.2-1 is a block-and-flow diagram of a robot’s operation, and Fig. 1.5-3 is a coverage layout.

In robotics, visualisation carries an extra load. A robot is built by mechanical, electrical and software engineers who use different vocabularies, and a shared drawing is often the only thing all three can read. Fig. 1.6-1 shows the situation this makes possible: the machine tested on a screen before it is trusted with a real task.

Photograph

Fig. 1.6-1 – an engineer at a workstation with a three-dimensional model or simulation of a robot arm on the screen, and the corresponding physical arm on the bench in the same frame. Both the screen content and the real machine must be recognisable as the same design, since the point is the correspondence between the model and the hardware. Stock images of generic screens covered in graphics, with no matching physical machine, are not usable.

1.6.2 Design and Creation

Visualisation is not decoration added after the thinking. It is the medium the thinking happens in, and it runs through two stages that overlap continuously.

Design is the planning and structuring of the system: what parts it will have, what each will do, and how they will interact. In practice it does four jobs. It structures the concept, mapping out boundaries, modules and the flow of data before anything is built. It exposes problems early, when a conflict between two parts is a line to be moved rather than a component to be scrapped. It gives the different disciplines a common language. And it keeps the eventual user in view, so that the machine can actually be operated, reached into and repaired by a person.

Creation is the making: turning those designs into working hardware. Digital models are built and simulated, prototypes are produced from the design files, the running system is monitored through dashboards that display speed, temperature and data flow, and every change is documented so the next version starts from an accurate record rather than from memory.

Neither stage happens once. Fig. 1.6-2 shows the cycle they actually form.

Diagram

Fig. 1.6-2 – flow: requirement → design drawn in CAD → simulation and checking → prototype manufactured → tested against the requirement → revision of the design, with the arrow from revision returning to the design block to close the loop. Add a second arrow from testing straight out to a block marked “released to use”, drawn as a dashed line labelled “only when the test is passed”. Mark each of the four arrows leaving simulation, prototype and testing with the cost of finding a fault at that point: “redraw”, “one part”, “whole machine”.

The costs written on that figure are the reason the loop is arranged as it is. A mistake caught in the model costs an afternoon’s redrawing. The same mistake caught in a machined part costs the part and the time to make another. Caught after the robot is installed and running, it can cost the machine, the work it was doing, and possibly the safety of the person beside it. Every hour spent on the left of that diagram buys back several on the right, and this is the single most useful habit an engineer can form early.

1.6.3 CAD Modelling

The tool that made this way of working ordinary is computer-aided design.

Definition

CAD modelling — computer-aided design modelling — is the use of specialised software to create precise digital representations of physical objects, components or systems in two or three dimensions, holding not only their shape and dimensions but also their materials, tolerances, surface finishes and assembly relationships.

That last clause is what separates a CAD model from a drawing. A drawing shows what a part looks like; a model knows what the part is. Because the software holds the material, it can compute the part’s mass. Because it holds the assembly relationships, it can swing a joint through its full travel and report whether the arm strikes its own base. Because it holds tolerances, it can be sent directly to a machine that will produce the part within them.

Four things a CAD model lets an engineer do, before any material is touched:

Check for interference — whether two parts occupy the same space at any point in the motion.
Simulate motion and load — how the structure deflects when the arm is fully extended with its heaviest payload.
Compute mass properties — the weight of the assembly and where its centre of mass sits, which decides whether a mobile robot tips over when it accelerates.
Produce manufacturing data — the model becomes the input to the machine that makes the part, through computer-aided manufacturing (CAM) software that converts the geometry into machine instructions.

Different work needs different software, and there is no single package that does everything well.

Table

Curiosity Corner

Q5. Long before a robot is cut and assembled, engineers build it inside a computer. What can they learn there that the workshop cannot tell them?

A. Things that would otherwise only be discovered by breaking something. In the model they can swing every joint through its full range and see whether links collide; extend the arm to its limit and see how far it sags under load; compute where the centre of mass sits and whether the robot tips when it turns quickly; and check that a bolt can actually be reached by a spanner once everything is assembled. A workshop can tell you all of this too — but only after the parts have been made, and each answer costs a set of parts. The model answers the same questions for the price of a redraw, and answers them while changing the design is still cheap.

Try Yourself

Take the chassis plate of a robot in your school laboratory and measure it: overall length and width, the diameter and spacing of the motor mounting holes, and the plate thickness, recording every value in millimetres. Then write beside each dimension how much variation you think would still allow the robot to be assembled — that figure is its tolerance, and Section 1.6.4 explains why stating it matters.

1.6.4 Precision Manufacturing

A design is only as good as the parts that come back from the workshop.

Definition

Precision manufacturing is the production of components to tight tolerances using controlled processes and accurate machinery, so that every part made conforms to its specification closely enough to be interchangeable.

The idea at its centre is tolerance: the amount by which a dimension is allowed to differ from its stated value. No process produces an exact size, so a specification that says only “8 mm” is not a specification at all. A shaft seat given as 8.00 mm with a tolerance of ±0.02 mm accepts any part between 7.98 mm and 8.02 mm, and rejects the rest. Tighten the tolerance and the part costs more; loosen it and the assembly develops play.

In a robot, that play is not a cosmetic matter. A gearbox with slack between its teeth lets the output shaft move a little without the motor turning at all, so the controller commands a position and the arm settles somewhere near it. The precision of the machine’s positioning is set in the workshop long before any program runs.

The processes used are chosen by what the part must achieve.

Computer numerical control (CNC) machining cuts metal or plastic to shape with a rotating tool guided by instructions from the CAD model, and holds tolerances of a few micrometres — a micrometre being one thousandth of a millimetre. Laser cutting produces flat profiles quickly from sheet material such as acrylic or MDF, which is how most school robot chassis are made. Three-dimensional printing builds a part up in layers, which suits complicated shapes and one-off prototypes. Injection moulding forces molten plastic into a mould, expensive to set up and extremely cheap per part once running, so it is used only in quantity. Fig. 1.6-3 shows the output of one of these processes on a scale you will meet directly.

Photograph

Fig. 1.6-3 – a set of laser-cut acrylic or MDF robot chassis parts, still held in the sheet by small tabs with the cut outlines and mounting holes clearly visible, and one assembled chassis of the same design beside the sheet. The cut edges and the hole pattern must be legible, and the assembled part must visibly correspond to the cut profiles. Photographs of a laser cutter’s exterior, or of finished robots with no cut sheet in frame, are not usable.

Common Error

Treating a three-dimensional printed part as a precision part What students do: print a gear, a bearing seat or a motor mount and fit it straight into an assembly that needs accurate location. What happens: layer-built parts shrink slightly as they cool and their holes come out undersized, so the gear meshes roughly, the shaft binds or the mount sits at a small angle — and the arm ends up with play the controller cannot correct. The fix: choose the process by the tolerance the part needs. Print brackets, covers and spacers freely; for anything that locates a shaft or transmits torque accurately, machine the part or drill and ream the critical holes after printing.

Safety Note

Cutting, printing and soldering equipment is used only under supervision and only by students trained on that machine. Keep hands and hair clear of any moving tool, use eye protection where the school’s rules require it, solder with a stand and with the extraction or ventilation running, and let hot parts cool before handling them. Power down and unplug a machine before clearing a jam or changing a tool.

1.6.5 Creating Sensors, Actuators and Control Units

The three subsystems of Section 1.2 are each made by combining CAD modelling with precision manufacturing, and each poses a different problem.

Making sensors

A sensor’s working parts are extremely small. Inside it may sit a thin diaphragm that flexes under pressure, a strain gauge whose resistance changes as it stretches, a piezoelectric layer that produces a voltage when squeezed, or an electrode a few micrometres across. These are modelled in CAD so that dimensions, layer thicknesses and material properties can be set exactly, and so the designer can simulate how the structure will respond to a change in temperature, pressure or force before committing to a fabrication run.

They are then made by processes borrowed from the semiconductor industry: microelectromechanical systems (MEMS) fabrication, in which mechanical structures are etched directly into silicon; photolithography, which patterns those structures using light and a light-sensitive coating; thin-film deposition, which lays down layers a fraction of a micrometre thick; and laser micromachining for fine features. The accelerometer in a robot or a phone is a MEMS device — a tiny mass suspended on silicon springs, small enough that several would fit across a full stop. Its outer housing, by contrast, is made conventionally by CNC machining or injection moulding, so that it fits the mechanical system around it. How such a sensor actually converts motion into a signal is studied later in the course.

Making actuators

An actuator is a mechanical problem rather than a microscopic one. CAD is used to model the gears, shafts, bearings, linkages and motor mountings, and to simulate the motion, the distribution of torque, the stresses in the loaded parts and the heat the motor will generate — all of which decide whether the chosen actuator is adequate before one is bought or built.

Manufacture then has to hold the tolerances that the simulation assumed. Gear teeth and shaft seats are produced by CNC milling and turning; complicated internal shapes that a cutter cannot reach are made by electrical discharge machining, which removes metal with controlled sparks in a fluid; and prototype housings come from high-resolution printing. The reason for this care is alignment: a motor mounted a fraction of a degree out of true loads its bearings unevenly, wastes power in friction, and wears out early.

Making control units

A control unit is an electronic problem, and it is designed with electronic design automation (EDA) software — the third row of Table 1.10. The engineer first captures the schematic, which states what is connected to what, and then lays out the printed circuit board (PCB), which states where each component physically sits and how the copper tracks run between them.

Layout is not a tidying exercise. Track width must suit the current it carries, signal tracks must be routed so that a noisy motor line does not corrupt a sensor line beside it, and components that dissipate heat must be placed where the heat can escape. A mechanical CAD model of the enclosure is developed alongside, so that the board fits, the connectors line up with their openings, and there is a path for air.

The board is then manufactured by etching away the unwanted copper, drilling, and placing and soldering the components by machine, and the enclosure by CNC milling or injection moulding.

Follow that path once more and the shape of the whole chapter appears. A robot senses, decides and acts; the parts that do those three things are themselves designed on a screen, checked in simulation and made to a stated tolerance; and every property the finished machine has — its reach, its precision, its reliability, its safety near people — was fixed by choices made at that stage. The chapters that follow take each of these parts apart in turn.

1.7 Summary

What a robot is

A robot is a programmable machine that acts physically on its surroundings, working from stored instructions, from sensor information, or from both, and carrying out its task with little or no human intervention while it runs.
Three properties matter: it changes something physical, its behaviour can be altered without rebuilding it, and it works from instructions or sensing. It need not look human and need not be intelligent.
A fixed automatic machine differs from a robot in that its motion is set by its mechanism. A robot’s instructions specify where to move and what to do there, so the same machine can be reprogrammed for another task.
The word came from Karel Capek’s 1920 play R.U.R., built on the Czech word robota, meaning forced labour. Asimov’s Three Laws of 1942 are a device in fiction, not a standard any machine is built to.
The engineering line runs from Devol’s 1954 patent and the Unimate installed in 1961, through computer-controlled arms from 1969, to machines that sense and decide — rovers, floor cleaners and driverless vehicles.
Robots are used for work that is repetitive and must be identical, work that is dangerous, work needing precision beyond the hand, places a person cannot go, and assistance and care. Each reason carries a cost, and robots remain poor at anything unfamiliar.

How a robot works

Every robot has five parts: sensing, a control system, actuation with mobility or manipulation, an energy source, and a program. Remove any one and the machine stops working.
These five form the sense–decide–act loop: sensors report the state, the controller applies stored instructions and selects an action, actuators carry it out, and the next reading is taken from the situation just altered.
A sensor converts a physical quantity into a signal and nothing more. Some sensors look outward at the world, others inward at the robot’s own state, and a robot needs both.
A controller knows only what its sensors report and can only choose actions its program provides. Most robot failures trace back to one of these two limits.
The end-effector is the device at the free end of an arm that performs the task — gripper, torch, suction cup or instrument. The arm decides where the tool can go; the end-effector decides what can be done there.
Power comes through a cable or is carried on board. A cable gives unlimited running time but restricts movement; a battery frees the robot but adds mass that the motors must then move.
Programs sit at three levels: a fixed sequence taking no account of surroundings, rule-based reaction to sensor readings, and learned behaviour derived from data. Most working robots are rule-based.

Autonomous and semi-autonomous operation

Any task divides into three decisions — the goal, the plan, and the moment-to-moment motion. What distinguishes robots is how those decisions are shared between machine and person.
An autonomous robot carries out its task without human intervention during operation, planning and controlling its own motion. A semi-autonomous robot performs part of the task on its own and depends on a person for the rest.
Autonomy is about which decisions are delegated, not about the absence of people. Someone still sets the goal, writes the program, fixes the limits and can stop the machine.
Full autonomy becomes necessary where supervision is impossible — a Mars rover’s commands and replies are separated by roughly eight to forty-eight minutes.
Perception is the step that turns raw readings into a usable picture of the surroundings, and it is built by sensor fusion, combining sensors that fail under different conditions rather than trusting any one.
Artificial intelligence contributes at five points: perception and interpretation, decision-making, planning and path-finding, control and execution, and learning and adaptation. Only the last requires machine learning.
Autonomy gains freedom from fatigue, safety, scalability and reach. It costs quiet sensor failures acted on at full speed, unpredictable environments, harder questions of responsibility, and a system that must be secured.

Types of robots

Classifications are not exclusive; each sorts by a different property, and the same machine occupies one place in each.
By application: Industrial, Service, Medical, Military, Agricultural, Exploration, Domestic and Entertainment robots.
By control mechanism: Manual or Remote-Controlled, Semi-Autonomous, Fully Autonomous, and Tele-operated. Tele-operation differs from remote control in that the operator works entirely through the robot’s own sensors, making the communication link a critical component.
By mobility: Stationary, Mobile, Aerial, Underwater and Humanoid robots.
By mechanical structure: Articulated, SCARA, Cartesian, Cylindrical, Spherical or Polar, and Delta robots.
The work envelope is the complete set of points the end-effector can reach. It is fixed by the structure and cannot be programmed around, so choosing a structure is choosing the shape of the space you can work in.
SCARA arms are stiff vertically and slightly compliant horizontally, which lets a peg find its hole. Delta robots keep their motors on a fixed overhead frame so only light arms move, which is why they are so fast.
Hybrid and specialised types include Swarm, Bio-inspired, Soft and Modular robots. Each gains something by giving something up — predictability, simplicity, precision or a purpose-built frame.

New age robotics systems

New age systems are built to cope with surroundings that are not controlled. Their defining features — built-in artificial intelligence, extended self-directed operation, collaborative working, learning, network connection and swarm coordination — all address variation.
Enabling technologies include artificial intelligence, machine learning, computer vision, the Internet of Things, cloud robotics, edge computing, 5G connectivity and advanced materials. Cloud processing offers power at the cost of delay; edge processing is immediate but limited, so real systems split the work.
Machine learning derives a decision rule from example data instead of being given one, which suits tasks nobody can specify. Its behaviour on unfamiliar input is not guaranteed, and it gives no warning when it is wrong.
Rule-based systems can be read, predicted and explained; learned systems cannot, but they cope with variation that rules cannot cover. The choice between them follows from how controlled the surroundings are.
In smart manufacturing: predictive maintenance from vibration, temperature and sound data; visual quality inspection of every unit; process optimisation from production records; and collaborative robots working beside people without fencing.
A cobot is made safe by limits on speed and force together with contact or proximity sensing. Those same limits cap its speed and payload, and raising them removes the property that allowed the fence to be left out.
In healthcare: models that flag regions in scans for a clinician to judge, surgical systems that scale and steady a surgeon’s movements without deciding anything, and powered devices for rehabilitation and continuous monitoring.
In autonomous vehicles: cameras, LIDAR, radar and ultrasonic sensors with overlapping coverage; localisation by GPS combined with SLAM; route planning and prediction of other road users’ behaviour; and conversion of the plan into steering, braking and acceleration. Driving automation exists in degrees, and many systems called self-driving are semi-autonomous.

From visualisation to creation

System visualisation is the graphical representation of a system’s components, structure, behaviour and interactions. It supports design understanding, debugging and testing, communication, documentation and optimisation, and in robotics it is the common language of the mechanical, electrical and software work.
Design and creation form a loop: requirement, design, simulation, prototype, test, revision. A fault found in the model costs a redraw, the same fault in a made part costs the part, and in a running machine it can cost the machine.
CAD modelling holds not only shape and dimensions but materials, tolerances, surface finishes and assembly relationships, which is why it can check interference, simulate load, compute mass properties and feed manufacturing directly through CAM.
Precision manufacturing produces parts to stated tolerances. A tolerance is the permitted variation in a dimension; without one a specification is incomplete, and slack in a gear train becomes positioning the controller cannot correct.
Processes are chosen by the tolerance required: CNC machining for accurate metal and plastic parts, laser cutting for flat profiles, three-dimensional printing for prototypes and complicated shapes, injection moulding for quantity.
Sensors are designed in CAD at micrometre scale and made by MEMS fabrication, photolithography, thin-film deposition and laser micromachining, with housings machined or moulded conventionally.
Actuators are modelled for motion, torque, stress and heat, then machined, spark-eroded or printed to tolerances that keep the drive aligned.
Control units are designed with EDA software as a schematic and then a PCB layout, where track width, signal separation and heat paths are decided, with the enclosure modelled alongside.

1.8 Key Terms

Actuator — a device that converts a controller’s electrical command into physical motion, such as a motor.
AI — artificial intelligence; the field concerned with making machines perform tasks that would ordinarily be said to require human intelligence, such as recognising objects, planning and choosing under uncertainty.
Autonomous robot — a robot that carries out its assigned task without human intervention during operation, planning its actions and controlling its motion from its own sensing.
CAD — computer-aided design; software used to create precise digital models of objects, holding their shape, dimensions, materials, tolerances and assembly relationships.
CAM — computer-aided manufacturing; software that converts a CAD model into the instructions a machine needs to produce the part.
Cloud computing — processing and storage held on distant servers rather than on the robot, giving large capacity at the cost of a round trip and its delay.
CNC — computer numerical control; machining in which the cutting tool is guided by instructions derived from the digital model.
Cobot — collaborative robot; a robot arm designed to work in the same space as people, kept safe by limits on speed and force together with contact or proximity sensing.
Computer vision — the interpretation of camera images by a computer, to recognise and locate objects.
Controller — the processor that reads sensor signals, applies the stored program and issues commands to the actuators.
Deep learning — machine learning using artificial neural networks with many layers, suited to large and complicated data.
EDA — electronic design automation; software used to capture a circuit schematic and lay out a printed circuit board.
Edge computing — processing carried out on the robot itself, so that decisions are immediate but limited by the power on board.
End-effector — the device fitted at the free end of a robot arm that performs the task, such as a gripper, welding torch or surgical instrument.
GPS — global positioning system; satellite-based positioning, used with other sensing because it is lost in tunnels and degraded between tall buildings.
IoT — Internet of Things; the connection of machines to a network so that they report their state and receive instructions remotely.
IR — infrared; light just beyond the red end of the visible range, used by proximity and reflectance sensors.
LIDAR — light detection and ranging; a sensor that measures distance from the time taken by light pulses to return, swept across a scene.
MDF — medium-density fibreboard; a flat sheet material commonly cut to make robot chassis parts.
MEMS — microelectromechanical systems; mechanical structures etched into silicon at micrometre scale, used to make sensors such as accelerometers.
Microcontroller — a single chip carrying a processor, memory and input-output pins, used as the controller of a small robot.
ML — machine learning; the part of artificial intelligence in which a system derives its own decision rule from example data instead of being given one.
New age robotics systems — robots built to cope with surroundings that are not controlled, using artificial intelligence, network connection, learning and collaborative operation.
NLP — natural language processing; the interpretation and generation of human language by a computer.
PCB — printed circuit board; the board that physically holds electronic components and the copper tracks connecting them.
Perception — the processing step that turns raw sensor readings into a usable picture of the robot’s surroundings and its own position.
Precision manufacturing — production of components to tight tolerances using controlled processes, so that parts conform closely enough to be interchangeable.
Reinforcement learning — a form of machine learning in which a strategy is arrived at by trying actions and being rewarded or penalised for the outcome.
Robot — a programmable machine that acts physically on its surroundings, working from stored instructions, from sensor information or from both, with little or no human intervention while it runs.
SCARA — Selective Compliance Assembly Robot Arm; an arm stiff in the vertical direction and slightly compliant horizontally, used for fast assembly of parts fitted from above.
Semi-autonomous robot — a robot that performs part of its task on its own but depends on a person for the remaining decisions, typically the goal or the approval of critical actions.
Sensor — a device that converts a physical quantity, such as distance, light, temperature or force, into an electrical signal.
Sensor fusion — the combining of readings from several sensors, which fail under different conditions, into one estimate of the situation.
SLAM — simultaneous localisation and mapping; building a map of the surroundings while working out one’s own position within it at the same time.
Soft robot — a robot built from flexible materials, which conforms to what it holds and is safer in contact with a person, at the cost of precision.
Swarm robotics — the use of many small, simple, identical robots that coordinate through local rules rather than one large capable machine.
System visualisation — the graphical representation of a system’s components, structure, behaviour and interactions, used to understand, test, communicate, document and improve it.
Tolerance — the amount by which a manufactured dimension is allowed to differ from its stated value.
Work envelope — the complete set of points a robot’s end-effector can reach, fixed by the arrangement of its links and joints and the limits of travel at each joint.

1.9 Objective Questions

Which one of the following is essential before a machine can be called a robot?

It resembles the human form.
Its behaviour can be changed without rebuilding it.
It contains artificial intelligence.
It moves on wheels or tracks.

A robot built with three straight sliding axes at right angles to one another has a work envelope in the shape of:

a hollow cylinder around the base
part of a hollow sphere
a rectangular box
a shallow dome below the base

Assertion (A): A tele-operated robot cannot continue its task if the communication link fails. Reason (R): In tele-operation the operator cannot see the machine and works entirely from information returned by the robot’s own sensors.

Both A and R are true, and R is the correct explanation of A.
Both A and R are true, but R is not the correct explanation of A.
A is true but R is false.
A is false but R is true.

Assertion (A): The response of a learned system to an input it has never met can be predicted by reading its program. Reason (R): In machine learning the decision rule is derived by the machine from example data rather than written out by an engineer.

Both A and R are true, and R is the correct explanation of A.
Both A and R are true, but R is not the correct explanation of A.
A is true but R is false.
A is false but R is true.

Match each sensor in Column A with the quantity it converts into a signal in Column B. | Column A | Column B | |—|—| | a. Ultrasonic | i. Rate at which the body is turning about its axes | | b. Gyroscope | ii. Time taken by light pulses to return, in many directions | | c. LIDAR | iii. Whether contact has been made | | d. Touch | iv. Time taken by a sound pulse to return from a surface |
Fill in the blanks.

The complete set of points a robot’s end-effector can reach is called its ____________.
The amount by which a manufactured dimension is allowed to differ from its stated value is called its ____________.
Combining readings from several sensors into a single estimate of the situation is called ____________.

Write the full form of each of the following: SCARA, LIDAR, MEMS, PCB, SLAM.

1.10 Check Your Concepts

Define a robot. Give one reason why an ordinary washing machine is not counted as one, even though it runs a stored programme on its own.
Differentiate between an autonomous robot and a semi-autonomous robot, and give one example of each.
Explain why a robot is normally fitted with several different sensors rather than with one very accurate sensor. Support your answer with two examples of sensors that fail under different conditions.
Explain what is meant by the work envelope of a robot. State why a structure with three straight sliding axes would be a poor choice for a task in which the tool must reach over and around an obstruction.
A small workshop wants a machine that can stand at a bench beside two workers, with no fencing, and be reprogrammed for a different product every week. Justify whether a collaborative robot or a fenced industrial arm should be selected, giving three reasons, and state one limitation that your choice brings with it.
A rover is to work on a planetary surface where a command sent from Earth and the reply to it are separated by tens of minutes. Explain why the machine must be autonomous rather than driven from Earth, describe the points at which artificial intelligence contributes to its operation, and state two costs that the mission accepts by giving the machine this autonomy.

1.11 Diagram, Circuit & Output Reading

A student draws the operation of a robot as: sensors → controller → actuators → movement, and stops there, leaving out the arrow that returns from the surroundings to the sensors.

State what this omission implies about how the machine would behave.
Name one real machine, from anywhere in this chapter, whose operation is correctly described by the student’s incomplete diagram, and say why.

A packaging line must pick light packets from a moving belt and place them into cartons, at about 120 packets per minute. Each packet has a mass of 200 g. Three robots are available. | Robot | Structure | Work envelope | Maximum payload | Speed of the pick-and-place cycle | |—|—|—|—|—| | P | Cartesian | Rectangular box, 600 mm × 400 mm × 200 mm | 5 kg | Moderate | | Q | Delta | Shallow dome, 1000 mm across and 250 mm deep | 1 kg | Very fast | | R | Articulated | Rounded volume, reach 900 mm | 10 kg | Moderate |

State which robot suits the task, and give two reasons drawn from the table.
Give one reason why each of the other two is unsuitable here.

A robot fitted with a single ultrasonic sensor at the front drives reliably in an open room, but it collides with a heavy cloth curtain, and it also collides with a plain wall whenever it approaches the wall at a steep angle.

Explain each of the two failures in terms of what the sensor actually measures.
State one change to the robot’s sensing that would prevent both, and say why it works.

The sensor coverage of a driverless vehicle is arranged as follows: radar in a long narrow cone ahead; a camera in a wide cone ahead; LIDAR sweeping the full circle around the vehicle; ultrasonic sensors covering only about two metres at the front and rear bumpers.

Name the sensor that covers a point 30 m directly behind the vehicle.
Explain why the region directly ahead is deliberately covered by more than one type of sensor.
Name one condition that weakens the camera and one that weakens the LIDAR.

1.12 Case-Based Questions

A school robotics club is building a robot to run along a corridor, stop when someone steps in front of it, and carry a small tray of laboratory samples between two rooms. The parts available are: an acrylic chassis, two motors with wheels, a castor, a motor driver module, a controller board, an ultrasonic sensor, a gyroscope, and a lithium polymer battery pack.

Identify the five parts of a robot in this build, naming which item on the list performs each.
The club must decide how much the robot does for itself. Taking the goal, the plan and the moment-to-moment motion in turn, state which of the three you would leave with the machine and which with a person, giving a reason in each case.
State two precautions the club must follow in handling and charging the battery pack, and give the reason for each.

A workshop makes steel brackets in batches of about fifty. The product changes every week, each bracket has a mass of 2 kg, and the only free space is a bench shared with two workers. The team is considering three options. | Option | Description | Change-over between products | Space required | |—|—|—|—| | 1 | A dedicated fixture that performs one fixed sequence | Rebuild required | Small | | 2 | A fenced industrial arm | Reprogramming, plus re-teaching positions | Large, fencing needed | | 3 | A collaborative robot at the bench | Reprogramming, often by guiding the arm by hand | Small, no fencing |

Recommend one option for this workshop and justify it against all three of the constraints stated.
State two things the team must check before deciding that the chosen machine may work beside people without fencing.
A custom mounting plate is needed to hold the workpiece, and the holes that locate the bracket must be within a tenth of a millimetre of their stated positions. Name a suitable manufacturing process for this plate, and give the reason for your choice.