Chapter 4: Real World Robots
On the day of the school science exhibition, the corridor outside the Class 9 rooms had six tables in a row.
On the first table, a small round vacuum cleaner moved in slow circles across a mat. It bumped gently into a chair leg, paused, turned, and set off again. The second table had a large photograph of a car factory, where orange mechanical arms welded door frames one after another, hour after hour, without stopping. The third showed a surgeon seated at a console a few steps away from the operating table, moving two handles while thin instruments moved inside the patient.
The fourth table had a picture of an aircraft with no pilot inside it, flying alone over open ground. The fifth showed a life-sized dinosaur at a theme park, roaring and turning its head towards visitors. The last one showed a tractor ploughing a field with nobody in the driver’s seat.
Meera walked along the row and stopped at the end. Six machines. Six completely different shapes. Not one of them looked like the others. Yet the card above every single table carried the same word: robot.
The six tables are shown together in Fig. 4.0-1.
Diagram
Fig. 4.0-1 – six small sketch panels arranged in two rows of three, drawn in the same simple style; panel 1 a round floor-cleaning robot on a mat, panel 2 an industrial arm welding a car door on a conveyor, panel 3 a surgeon at a console with robotic instrument arms beside a patient, panel 4 a fixed-wing uncrewed aircraft in flight, panel 5 an animatronic dinosaur at a theme park, panel 6 a driverless tractor ploughing a field; each panel labelled with its field – Domestic, Industry, Medical, Defense, Entertainment, Agriculture
As you work through this chapter, you will find the answer to each of these five questions.
4.1 What Makes a Robot a Real World Robot
Think of an ordinary table fan. You switch it on and it spins. You switch it off and it stops. It does the same thing whether the room is empty or crowded, cool or hot. The fan never checks anything about the room. It only obeys the switch.
Now think of an air conditioner set to 24 °C. Inside it there is a small part that keeps measuring the room temperature. When the room is warmer than 24 °C, the cooling turns on. When it drops below, the cooling stops. The machine is now doing something extra — it is taking a reading and changing its own behaviour because of that reading.
A robot goes one step further. It takes readings, applies the rules stored in its program, and then moves a body of some kind to get a real job done — clean a floor, weld a joint, pick a tomato.
So every robot, whatever its shape, is built from three kinds of parts working as a loop. See Fig. 4.1-1.
Diagram
Fig. 4.1-1 – a simple three-box loop drawn as a circle of arrows; box 1 labelled “Sensors – take readings from the surroundings”, box 2 labelled “Controller – applies the rules in the program”, box 3 labelled “Actuators – motors and parts that move and do the work”; arrow from box 3 curving back to box 1 with the label “the surroundings have now changed”
A sensor is a part that measures something about the surroundings and turns it into an electrical signal — distance, light, temperature, weight. This is the robot’s input.
The controller is the small computer inside the robot that holds the program. It takes the sensor readings, applies the rules it was given, and works out which command to send next.
An actuator is a part that produces movement or a physical effect when it is given electrical power — a motor, a wheel, a gripper, a valve. This is the robot’s output.
The loop then repeats, many times a second. Readings come in, the controller applies its rules, the actuators move, and because the actuators have moved, the next set of readings is different.
The words “real world” matter here. A robot in a laboratory works on a clean, flat, well-lit table where everything is where it is expected to be. A real world robot has to manage a floor with slippers on it, a field with stones in it, or a patient who is breathing.
Who is in control
Two robots can look almost the same and still be controlled in completely different ways.
A teleoperated robot is worked by a person from a distance. The person watches a screen and moves a joystick or a pair of handles, and the robot copies those movements. The robot supplies the reach, the steadiness or the safety; the person supplies every decision.
An autonomous robot carries out the whole task on its own once it has been started, using only its sensor readings and its program. Nobody is steering it while it works.
Many robots sit between the two. A driverless tractor may plough a field on its own but stop and wait for a person the moment something blocks its path.
4.1.1 One Idea, Many Different Bodies
Once you know the sense–apply–act loop, the strange variety of robot shapes stops being strange. The body is not decided by fashion. It is decided by two questions: where does this robot have to work, and what does it have to touch?
A floor cleaner has to slide under a bed and a sofa, so it is flat and low. It has to turn in a tight corner without getting stuck, so it is round. A welding robot never has to travel anywhere, so it has no wheels at all — it is bolted to the factory floor and is built instead as a long jointed arm that can reach right around a car body. See Fig. 4.1-2.
Diagram
Fig. 4.1-2 – four robot outlines side by side, each with a one-line label under it explaining the shape; (a) low round floor cleaner, “flat, to fit under furniture”, (b) jointed arm fixed to a base, “no wheels, long reach”, (c) four-rotor aerial robot, “light, to stay in the air”, (d) tracked ground robot, “tracks, for loose and broken ground”
An aerial robot must be light, because every extra gram has to be lifted. A robot sent over rubble runs on tracks rather than wheels, because tracks spread its weight and do not sink in. A robot that must walk into a home and be spoken to is often given a head and two arms, so that people find it easy to face and talk to.
4.1.2 Why We Send Robots to Work
A robot is expensive to buy, install and maintain. So it is worth asking what kind of work is worth handing over to one. Four kinds of work come up again and again.
The first is work that repeats exactly. A car door needs the same weld in the same place, several hundred times a day. A person doing this gets tired, and quality slips a little as the shift goes on. A robot arm makes the last weld of the day the same way it made the first.
The second is work that is dangerous. Inside a mine, on the seabed, near a suspected landmine, or in a room full of fumes, sending a machine instead of a person removes the risk to life completely.
The third is work that needs more precision than a hand can give. Placing a component a fraction of a millimetre out of position will ruin a circuit board. A robot can hold that position again and again.
The fourth is work that must run for very long stretches. A robot does not need sleep or meals. It stops only for charging and maintenance, so a job can run through the night.
That last point is the honest limitation of every robot in this chapter, so it is worth saying plainly. A robot has no understanding of the place it is working in. It has readings, and it has rules. When something happens that the rules do not cover — a spilled bucket of water, a branch fallen across a row of crops, a patient who moves unexpectedly — the robot does not work out a sensible new plan. It carries on applying old rules to a new situation, which is exactly when robots go wrong.
This is why almost every robot you will meet in the next six sections has a person somewhere in the picture: watching it, checking its output, or ready to stop it.
4.2 Robots in Domestic Use
A home is a small place. It should be the easiest place in the world for a robot to work in. In fact it is one of the hardest.
A factory floor is planned. The same car body arrives at the same spot at the same moment, and nothing else is ever there. A home is the opposite. Chairs get pushed back. Slippers land in the middle of the room. A dog walks across the floor. Children leave a bag where the robot expected clear space. Nothing in a home stays where it was yesterday.
So home robots are built to do one narrow job and to survive being interrupted. That is the pattern to look for in every example below.
Cleaning the floor
The oldest and most successful home robot is the floor cleaner. The problem it solves is simple: sweeping a whole house is dull, it must be done every day, and it takes time nobody has.
A floor-cleaning robot is a flat disc a few centimetres high with two driven wheels, a rotating brush underneath, and a small dust bin inside. To work safely in a room it has never seen, it needs to answer two questions constantly — is something in front of me, and is the floor still there? See Fig. 4.2-1.
Diagram
Fig. 4.2-1 – underside and side view of a round floor-cleaning robot; label the two driven wheels, the small front castor wheel, the rotating side brush, the dust bin, the front bumper marked “bump sensor”, and three small downward-facing sensors along the front edge marked “cliff sensors”; add a side view showing the robot at the top of a stair with the cliff sensor beam pointing down into empty space
A bump sensor is a switch behind the robot’s front bumper. When the bumper is pushed in by an obstacle, the switch closes and the controller receives a signal meaning “something is there”.
A cliff sensor points down at the floor from the underside of the robot. It sends out an infrared beam and measures the reflection that comes back. Over a floor, the beam bounces back quickly. At the edge of a staircase, nothing bounces back, and the controller stops the drive motors before the robot goes over.
Cutting grass and cleaning pools
A robotic lawn mower does for a lawn what the floor cleaner does for a room, with one extra problem — a lawn has no walls. Nothing stops the mower from wandering onto the road.
The usual answer is a boundary wire: a thin insulated wire pegged around the edge of the lawn with a small current passing through it. The current creates a weak magnetic field, a coil in the mower detects that field, and the controller turns the mower back inside whenever the coil signal rises. Husqvarna’s Automower range is a well-known example. See Fig. 4.2-2.
Diagram
Fig. 4.2-2 – top view of a lawn with a boundary wire pegged around its edge and running around a flower bed in the middle; the charging station is drawn on one edge connected to the wire; a small mower is drawn inside with a curved arrow showing it turning back after reaching the wire
Robotic pool cleaners work on the same idea in three dimensions. They crawl along the floor and walls of a swimming pool, brushing the surface and pumping water through a filter bag to trap leaves and dirt.
Watching the house
A home security robot is a small camera on wheels. Instead of fixing many cameras to many walls, one moving camera patrols a route and streams what it sees to a phone. Appbot Riley is an example of such a home monitoring robot.
The camera image is processed to find movement between one frame and the next, and a large change triggers an alert to the owner.
Kitchen, laundry and pets
Three more home tasks have robots built for them, though these are far less common in ordinary homes than a floor cleaner.
In the kitchen, robotic arms mounted above a work surface can stir, pour and follow a stored sequence of movements for a recipe. Moley Robotics’ kitchen robot is the best known of these. It does not cook the way a person cooks — it replays a set of recorded arm movements, so every ingredient must be pre-measured and placed exactly where the sequence expects it.
For laundry, machines have been built that take clothes from a pile, flatten each piece and fold it. This turns out to be very hard, because cloth changes shape as soon as it is touched, and no two shirts land in the same position.
For pets, an automatic feeder holds dry food in a hopper and releases a measured portion at set times, so an animal is fed even when the family is away. PetSafe Smart Feed is one such feeder.
Table 4.2 pulls the domestic examples together and shows what each one senses and what each one moves.
Look down the middle two columns. Every one of these machines is doing the same three things — a reading comes in, the program’s rule is applied, a motor turns. Only the job at the end of the row changes.
4.3 Robots in Industry
Imagine you have to draw the same circle, in the same place, on ten thousand sheets of paper. Your first hundred circles would be good. By the thousandth your hand would ache, and by the five thousandth the circles would be drifting away from the mark.
That is the problem industry has. A factory does not need one door welded well. It needs the same weld, in the same twelve places, on every door that comes down the line for years. Human hands are wonderfully skilled, but they are not built to repeat a movement to the millimetre all day long.
This is the gap the industrial robot fills, and it is why factories were the first place robots were used seriously.
The workhorse: the robot arm
Most industrial robots are not vehicles. They are arms bolted to the floor, made of rigid metal links joined by motor-driven joints. The arm does not travel; the work is brought to it. See Fig. 4.3-1.
Diagram
Fig. 4.3-1 – side view of a six-joint industrial robot arm bolted to a heavy base; label the base, each joint with a curved arrow showing which way it turns, the links between the joints, and the tool at the far end labelled “end effector”; show three small tool drawings alongside – a welding torch, a two-finger gripper and a paint spray gun – with the caption note “the same arm, different end effectors”
The tool at the far end of the arm is called the end effector. It is the part that actually does the job. Fit a welding torch and the arm welds. Fit a gripper and the same arm picks and places. Fit a spray gun and it paints. The arm is general; the end effector makes it specific.
The quality that makes such an arm valuable has a name. Repeatability is how closely a robot returns to exactly the same position each time it is sent there. A typical factory arm has a repeatability of a few hundredths of a millimetre — far finer than a person can hold by hand, and, more importantly, it does not get worse as the shift goes on.
How much difference does it make?
The gain from a robot is easiest to see as a simple count of finished pieces.
Two things are worth noticing about that answer. Only one third of the gain comes from the robot being three times faster per piece; the rest comes from it running through hours when the workshop would otherwise be shut. And the sum says nothing about the cost of buying and installing the cell, which is what actually decides whether a workshop buys one.
Moving goods around the plant
Not all factory work happens at one spot. Materials have to travel — from the store to the line, from the line to the packing area, from packing to the loading bay.
Stacking finished boxes onto a pallet in a neat pattern is called palletizing, and it is a standard job for a robot arm with a vacuum gripper. Carrying those pallets across the building is a job for a different kind of machine.
An autonomous mobile robot (AMR) is a wheeled robot that finds its own way across a floor while carrying goods. It builds a map of the warehouse using a laser scanner, works out its position on that map from the walls and racks it can see, and plans a path around whatever is in the way.
Checking, inspecting and building
Robots do three more kinds of industrial work that are worth knowing about.
The first is quality control and inspection. A camera is mounted above the line and photographs every piece as it passes. The controller measures features in the image — a hole’s diameter, the length of an edge, whether a component is missing — and compares each measurement with the limits stored for a good part. Anything outside the limits is pushed off the line. A person checking by eye would tire and miss faults; a camera measures the ten-thousandth piece exactly as carefully as the first.
The second is inspection and maintenance in hard places. Pipelines, boiler tubes, tall chimneys and offshore platforms all have to be checked from inside or from underneath, in places that are cramped, hot or flooded. Crawling robots carry cameras through pipes. An autonomous underwater vehicle (AUV) — an uncrewed submarine that follows a stored survey path on its own — inspects underwater oil pipelines and cables, where a diver could work only for a short time and at real risk.
The third is mining and construction. In mines, robotic drilling rigs and remotely driven loaders keep people away from unstable rock and dust. On building sites, machines have been built to lay bricks, spray concrete and print building sections layer by layer.
Two industries deserve a special mention because they need the arm’s precision more than its strength. In electronics manufacturing, pick-and-place machines set tiny components onto a circuit board and solder them — components far too small to place by hand at any useful speed. In textile and garment factories, robots handle cutting and fabric feeding, though cloth remains difficult for the same reason it defeats a folding machine: it changes shape the moment it is touched.
4.4 Robots in the Medical Field
Hold your hand out in front of you, as still as you can, and look at your fingertips. They are not still. Everyone’s hand carries a small, constant shake, and it grows when you are tired or tense. For most work that shake does not matter. For work inside a human body it matters a great deal.
Surgery has a second, harder problem. To reach an organ, a surgeon has traditionally had to make an opening large enough for two hands and a clear view. The operation itself may take an hour; recovering from the opening can take weeks.
Minimally invasive surgery is the answer to that — operating through a few small cuts, each about the width of a pencil, using long thin instruments and a camera passed through them. It heals far faster. But it is awkward, because a long rigid instrument through a small hole moves the wrong way: push the handle left and the tip goes right, and the instrument cannot bend once it is inside.
This is exactly where a robot helps.
The surgical robot
A surgical robot system has two separate parts standing a few metres apart in the same room. See Fig. 4.4-1.
Diagram
Fig. 4.4-1 – two connected parts drawn side by side; on the left a console with a seated surgeon looking into a viewer and holding two hand controls, labelled “surgeon’s console”; on the right a cart with four slim jointed arms leaning over a patient on a table, three holding instruments and one holding a camera, labelled “patient-side cart”; a thick arrow from console to cart labelled “hand movements sent as signals”, and an arrow back labelled “camera view”
The Da Vinci Surgical System is the best-known example of this design and is used in hospitals around the world, including in India.
Two of those steps deserve a closer look, because they are what a hand alone cannot do.
Removing the shake is straightforward for a controller: a tremor is a small, fast, back-and-forth movement, and the program simply does not pass movements of that size and speed on to the instruments. Only the slower, deliberate movement gets through.
Scaling is just as simple and just as useful. The controller multiplies the surgeon’s movement by a fraction before sending it on.
That is the whole trick. The surgeon works with big, comfortable, controllable hand movements, and the instrument inside the patient makes small, fine ones.
Helping people move again
Not every medical robot is for the operating theatre. Some are worn.
An exoskeleton is a powered frame strapped to the outside of a person’s legs or arms. Motors at its hips and knees drive the frame through a walking movement, taking the person’s limbs with it. For someone relearning to walk after a spinal injury or a stroke, this allows many more steps of practice than two therapists lifting by hand could ever manage. The ReWalk exoskeleton is one such system.
Powered artificial limbs work from the other side. Small sensors on the skin pick up the faint electrical signals produced when the wearer tenses the remaining muscles in the limb, and the controller drives motors in the hand or foot to match. The person moves the limb by using the same muscles they always did.
Distance, laboratories and beams
Three more uses complete the picture.
The first is reaching patients who are far away. A telepresence robot is a screen and camera on a wheeled base that a doctor drives from another city, speaking to the patient and the nursing staff through it. The VGo telepresence robot is used this way. Where a specialist is hundreds of kilometres from a small hospital, this turns a journey into a video call that can move from bed to bed.
The second is laboratory and diagnostic work. Testing thousands of blood or tissue samples means opening tubes, drawing an exact volume of liquid and moving it to the right well of a tray, over and over. A pipetting robot does this all day with the same volume every time, which is both faster and more consistent than by hand, and it keeps laboratory staff away from infectious material. The same automation speeds up pharmaceutical research, where huge numbers of candidate compounds have to be prepared and tested.
The third is aiming a beam. In radiosurgery, a radiation beam must be delivered to a tumour while sparing the healthy tissue around it. Mounting the beam source on a robot arm allows it to be fired from many directions in turn, so the beams cross at the tumour and each one passes through different healthy tissue on the way in. The CyberKnife system works in this way.
Robotic endoscopes and catheters are steered along the body’s own passages and blood vessels to reach places no straight instrument can. And in medical colleges, robotic manikins that breathe, blink and respond to drugs let students practise a procedure many times before they ever meet a patient.
4.5 Robots in Defense
Almost every use of robots you have read about so far saves time, money or effort. Defense is different. Here the thing being saved is usually a person’s life, and the reasoning is blunt: if a task might kill whoever does it, send a machine that can be replaced instead of a person who cannot.
That single idea explains most of the robots used by armed forces and by the police units that handle dangerous objects.
Looking without going
The first job is simply seeing. Before anyone can be sent into an area, someone has to know what is there — how a road is blocked, whether a building is occupied, where a fire is spreading after an accident.
An uncrewed aircraft does this from above. It carries cameras instead of a pilot, flies a route planned in advance, and sends its pictures back to a ground station. Because there is no person on board, it can stay in the air for many hours and fly over ground nobody could safely cross. The MQ-9 Reaper is one such large uncrewed aircraft used for aerial surveillance.
On the ground, small tracked robots do the same job at close range. The iRobot PackBot is a well-known example: it is light enough to be carried in a backpack, and it is driven into a building or a tunnel while its operator watches the camera view from a safe distance outside.
Underwater, the same idea appears again. Uncrewed underwater vehicles such as the REMUS 600 survey the seabed and harbours, following a planned path and recording what their sonar finds.
The ground robot up close
Look at the working parts of a small ground robot and you will recognise every one of them from the earlier sections. See Fig. 4.5-1.
Diagram
Fig. 4.5-1 – side view of a small tracked ground robot; label the rubber tracks, the small front flippers that help it climb steps, the camera on a short mast, the jointed manipulator arm with a gripper at its end, the antenna, the tether spool at the rear, and the battery pack; alongside, draw an operator sitting behind a low wall a long distance away holding a control unit with a screen
Search and rescue uses exactly this machine with a different purpose. After a building collapse or an earthquake, a robot small enough to crawl into a gap under broken concrete can carry a camera and a microphone to a place no rescuer could reach, and tell the team where to dig. The REMOTEC ANDROS F6A is used for work of this kind.
Mine clearance is the heaviest version of the job. A landmine buried in a field will not go away on its own, and finding one by hand is slow and dangerous work. Remotely driven machines such as the DOK-ING MV-4 are built low and armoured, and are driven across the ground from a distance so that any explosion damages a replaceable machine.
The delay problem
There is one difficulty that grows the further away the operator sits. Signals are fast, but they are not instant, and a control path that goes up to a satellite and back is a very long path.
Half a second sounds small. It is not. It means the picture on the operator’s screen always shows where the robot was half a second ago, so a robot driven at any speed is being steered on old information. See Fig. 4.5-2.
Diagram
Fig. 4.5-2 – a simple path diagram; operator with control station at bottom left, an upward arrow to a satellite at the top labelled “command up”, a downward arrow to the robot at bottom right labelled “command down”, then a matching pair of arrows for the returning picture; a caption box noting “total path ≈ 144 000 km, delay ≈ 0.48 s”
This is the main reason defense robots are given some ability to work on their own — an aircraft that follows a stored route without being steered, or a ground robot that stops by itself when its sensors show an obstacle. The further away the operator, the more the machine must handle between commands.
Carrying, and the hardest question
Not all of this work is dangerous. Armies also move enormous quantities of supplies, and legged and wheeled carriers have been built to follow a squad over ground too rough for a truck, carrying loads that soldiers would otherwise carry on their backs. The Boston Dynamics LS3, a legged support machine designed to carry heavy loads for troops in the field, was built for exactly this.
Finally, some defense robots carry weapons and are operated remotely, so that the person firing is not exposed to return fire.
4.6 Robots in Entertainment
Every robot so far has been judged on results. Was the floor clean? Was the weld in the right place? Did the patient recover?
Entertainment robots are judged on something else entirely: did it look right to the person watching? A theme-park dinosaur that welds nothing and cleans nothing is a complete success if a child believes it turned its head to look at her. This changes what the engineers work on. Speed and output matter less. Appearance, timing and safety around a crowd matter far more.
Animatronics
Animatronics is the building of figures — animals, dinosaurs, characters — that move by motor power to look alive.
Under the skin, an animatronic figure is a rigid frame with joints, much like an industrial arm, but shaped as a neck, a jaw or an eyelid instead of a work arm. See Fig. 4.6-1.
Diagram
Fig. 4.6-1 – cutaway side view of an animatronic head; label the rigid inner frame, a small servo motor at the neck joint, two servo motors linked by thin rods to the eyelids and the jaw, the control wires running down the neck, and the flexible skin drawn as a dotted outline over the whole frame
Some figures do add a small amount of sensing on top — a sensor that starts the sequence when someone comes near, or a camera that lets a character turn towards a visitor. But the performance itself is nearly always stored in advance, because a show must run the same way forty times a day without surprises.
Robots behind the camera
Film-making uses robots in a quieter way, where the audience never sees them.
The clearest example is aerial photography. A camera carried under an aircraft shakes with every gust and every correction the aircraft makes, and shake ruins a shot. The answer is a gimbal — a motorised mount holding the camera in a frame of pivots.
Inside the gimbal is a small sensor that measures how fast the mount is tilting. Whenever the aircraft tips, the sensor reports the tilt and the controller drives the gimbal motors the opposite way by the same amount, so the camera keeps pointing where it was. See Fig. 4.6-2.
Diagram
Fig. 4.6-2 – a four-rotor aerial robot shown tilted nose-down; underneath it a camera held in a three-ring gimbal, with the camera drawn still level; small curved arrows on the gimbal rings labelled “motors turn the opposite way”, and a small box on the gimbal labelled “tilt sensor”
The same idea, on the ground, is a camera on a robot arm. Because the arm has excellent repeatability, it can repeat a camera movement exactly — which is what lets a film crew shoot a scene twice, once with an actor and once without, and combine the two shots perfectly.
Robots on stage, in arcades and in classrooms
Live theatre uses robots in two ways. Some appear as characters. Many more stay out of sight, moving scenery, opening trapdoors and swinging lights on cue — jobs that used to need a crew hauling ropes, and that now have to hit the same mark on the same beat every night.
Amusement centres and arcades use robotic games as attractions in themselves, and robot competitions — machines built by teams to race, fight or play football against each other — have become events that people come to watch. Robots are also combined with virtual reality and augmented reality systems, where a physical machine gives a player something real to feel while the headset supplies the scene around it.
The last use is the one nearest to you. Educational robot kits — LEGO MINDSTORMS is the best known of them — package a controller, a few motors and a few sensors with parts to build with, so that a student can assemble a machine and program it in an afternoon.
These kits matter more than they look. Every idea in this chapter — a reading arriving, a rule being applied, a motor turning — becomes obvious the first time you build a machine that drives forward until its sensor sees a black line and then stops.
4.7 Robots in Agriculture
A factory is built to suit its robots. A field is not built at all.
Farming is where robots meet the least forgiving conditions of any field in this chapter, and it is also where they are spreading fastest — because farm work is heavy, it is seasonal, it has to be finished in a short window of good weather, and there are fewer hands available to do it every year.
Driving in straight lines
Ploughing, sowing and spraying are all done by driving up and down a field in parallel strips. Doing this by eye is surprisingly hard. A driver cannot see exactly where the last pass ended, so the usual habit is to overlap generously — better to cover a strip twice than to leave a gap.
The Global Positioning System (GPS) removes the guesswork. A receiver on the tractor picks up signals from satellites and computes its own position on the Earth’s surface. The controller compares that position with the line it is supposed to be following and steers to stay on it. See Fig. 4.7-1.
Diagram
Fig. 4.7-1 – top view of a rectangular field with a tractor and wide sprayer boom shown mid-pass; on the left half draw wavy, unevenly spaced passes with wide shaded overlap strips, labelled “steered by eye”; on the right half draw straight, evenly spaced passes with thin overlap strips, labelled “steered by GPS”; mark the boom width and one overlap strip clearly
That 4560 m² is chemical bought and paid for, sprayed onto ground that had already been treated. Multiply it by every field and every spraying round in a season and the waste becomes the reason the guidance system pays for itself.
Guided tractors can also work through the night, which matters when a crop must be sown or harvested within a few days.
Telling a crop from a weed
Spraying a whole field to kill weeds treats the crop as well. A better method is to spray only where a weed actually is — but that means the machine must first tell one green plant from another.
A weeding robot carries a camera over the row and lights the ground evenly so the picture does not change with the sun. Its program has been trained on many photographs of the crop and of the weeds that grow with it, and for each plant it sees, it computes which of the two the plant more closely matches. When the answer is “weed”, a small nozzle sprays that plant alone, or a blade cuts it out. See Fig. 4.7-2.
Diagram
Fig. 4.7-2 – camera view looking down at two neat crop rows; the crop plants sit evenly spaced along the rows and are outlined with plain boxes, while three differently shaped plants growing between the rows are outlined with dashed boxes; a nozzle drawn above one dashed box with a short spray cone aimed only at that plant
Watching, picking and milking
Three more jobs complete the picture of a robotic farm.
The first is watching the crop. In precision agriculture, the aim is to treat each part of a field according to what that part needs, instead of treating the whole field alike. Aerial robots fly over the crop with cameras and sensors and record where the plants are dense or thin, where the soil holds water and where it has dried. Ground robots take soil samples. The farmer then waters and fertilises the weak patches rather than the whole field.
The second is picking the crop. Harvesting soft fruit is the hardest robotic job on a farm. A tomato or a strawberry bruises if it is squeezed, hides behind leaves, and only some of the fruits on a plant are ripe on any given day. A picking robot judges ripeness from colour, works out where the fruit is from two camera views, and closes a soft gripper around it with just enough force to hold. It works slowly compared with a person, which is why it is used where pickers are hard to find rather than because it is faster.
The third is milking and feeding animals. A milking robot handles one cow at a time. The cow chooses when to walk in, and a tag it wears identifies it to the controller, which looks up when that particular cow was last milked. Sensors locate the teats, the cups are attached, and the yield is recorded against the cow’s own record. Systems such as the GEA DairyRobot work in this way. The cow is milked when it is ready rather than at two fixed times a day, and any drop in a cow’s yield shows up in the records early.
Once the crop leaves the field, sorting and packing lines take over — cameras grade fruit by size and colour, and arms place the graded fruit into trays.
Notice what every one of these machines has in common. None of them is doing something a farmer cannot do. Each one is doing a small part of the work more evenly, or at a time when nobody is available to do it at all.
4.8 Looking Across the Six Fields
Six fields, and not one machine in common between them. A floor cleaner and a milking robot share no parts, no shape and no purpose. Yet everything in this chapter has been built on the same three-step loop: a sensor takes a reading, a controller applies the rules in its program, an actuator moves.
So it is worth asking what actually changes from one field to the next. Three things do.
What changes: the workplace
The first is how predictable the place is.
A welding cell is the most controlled workplace in this chapter. The same part arrives at the same spot, clamped the same way, under the same light, with no people inside the fence. A robot there can succeed with very little sensing at all — it can simply repeat stored positions.
A field after rain is the opposite. Nothing is where it was last week, the light changes hourly, and every plant is a slightly different shape. A robot there needs cameras, position sensing, and rules for situations that could only be guessed at when it was built.
The other four fields sit between these two. See Fig. 4.8-1.
Diagram
Fig. 4.8-1 – a horizontal scale with “highly controlled workplace” at the left end and “unpredictable workplace” at the right end; place six small labels along it in order – Industry, Entertainment, Medical, Domestic, Agriculture, Defense; below the scale draw two arrows, one pointing left labelled “working alone is easier here”, one pointing right labelled “a person stays closely involved here”
This scale explains something that surprises many students. The most impressive-looking robots — the surgical system, the bomb-disposal machine — are usually the ones that work least on their own. That is not a failure of engineering. It is the sensible answer to an unpredictable workplace where being wrong is very costly.
What changes: the person’s place in the loop
The second difference is how much a person is doing while the robot works.
At one end, nobody is involved at all: a floor cleaner runs at night in an empty house. In the middle, a person sets the machines up and then supervises several at once, as in a warehouse or a milking shed. At the other end, a person is working the machine second by second, and the robot moves only while they move — the surgical console and the bomb-disposal controller both work this way.
What changes: what counts as doing the job well
The third difference is the measure of success, and it is easy to miss.
In industry, success is countable — pieces per day, faults per thousand. In medicine, success is the patient’s recovery, and a slower operation is perfectly acceptable if it heals better. In entertainment, success is whether the audience believed it. On a farm, success is measured across a whole season rather than a shift.
Table 4.5 sets the six fields side by side.
Read the last column downwards. The same machine idea appears three different ways — running alone, replaying a sequence, or copying a person’s hands — and which one is chosen depends far more on the workplace than on how advanced the robot is.
What has not changed anywhere
For all the variety, one limitation runs through every row of that table, and it is the same one stated back in Section 4.1.
None of these machines understands the place it is working in. It has readings, and it has rules. The welding arm does not know it is building a car. The milking robot does not know it is dealing with an animal. The surgical arm does not know there is a patient. Each of them computes an output from an input, extremely well, within the situations it was built for.
That is why every robot in this chapter has a person somewhere near it — setting it up, watching it, checking its work, or ready to switch it off. The fence around the welding cell, the surgeon at the console, the operator behind the wall, and the farmer who tests a weeding program on his own crop first all exist for that one reason.
So the useful question to ask about any robot you meet is not “how clever is it?” It is two plainer ones: what exactly is this machine measuring, and what happens when something it was never given a rule for walks into its path?
Having seen what robots do, the next step is to open one up and look properly at the blocks it is built from.
4.9 Summary
- Every robot, whatever its shape, works as a loop of three parts: sensors take readings, a controller applies the rules in its program, and actuators move to do the work.
- A real world robot is one built to do a useful job in an actual working place — a home, a factory, a hospital, a field — rather than on a clean laboratory bench.
- Robots look completely different from one another because the body is decided by where the robot works and what it must touch, not by any fixed idea of what a robot should look like.
- A teleoperated robot is worked by a person from a distance and copies that person’s movements. An autonomous robot carries out the whole task on its own, from its sensor readings and its program, once it has been started.
- Work is usually handed to a robot when it repeats exactly, is dangerous, needs more precision than a hand can give, or must run for very long stretches.
- Domestic robots do one narrow job in an unpredictable home. A floor cleaner uses bump sensors to find obstacles by touching them, and cliff sensors to stop at a stair edge; simpler models turn at random after each bump, while costlier ones build a map with a laser scanner.
- Other home robots include lawn mowers kept inside a boundary wire, pool cleaners, patrolling camera robots, kitchen arms that replay a stored cooking sequence, and timed pet feeders.
- The industrial workhorse is a jointed arm bolted to a base, carrying a changeable tool called an end effector — a welding torch, a gripper or a spray gun.
- Repeatability is how closely a robot returns to exactly the same position each time. It is the quality that makes an arm valuable, and it does not fall off as the shift goes on.
- A welding arm has no camera; it reproduces taught positions, so the part must be clamped in exactly the right place every time.
- Industry also uses robots for palletizing, for carrying goods with autonomous mobile robots, for camera-based quality inspection, for inspecting pipelines and underwater structures, and in mining, construction, electronics and textiles.
- Industrial arms are kept inside fenced cells with light curtains, because an arm cannot sense that a person has walked into its path.
- In medicine, a surgical robot is teleoperated. It filters out the surgeon’s hand tremor and scales the movement down, so the instrument tip inside the patient moves far less than the surgeon’s hand. The surgeon controls every movement.
- Medical robots also include exoskeletons and powered limbs for rehabilitation, telepresence robots that let a specialist see distant patients, pipetting robots for laboratory testing, robot-aimed radiotherapy beams, steerable endoscopes and catheters, and training manikins.
- Defense robots exist mainly to keep people out of danger: uncrewed aircraft and underwater vehicles for surveillance, small tracked robots for examining dangerous objects, machines for search and rescue and for mine clearance, and carriers for supplies.
- A long control link adds a delay, so the operator always sees where the robot was a moment ago. This is one reason defense robots are given some ability to work between commands.
- Entertainment robots are judged on how they look to an audience. An animatronic figure replays a recorded list of joint positions in step with the sound, and a dancing legged robot plays back a routine written by people while computing its own balance.
- Film-making uses a gimbal — a motorised camera mount that turns against the aircraft’s shake using a tilt sensor — and robot arms that repeat a camera move exactly.
- Farms are the least predictable workplace: uneven ground, changing light, and plants that differ from each other and change every week.
- Farm robots use satellite positioning to drive accurate parallel passes and cut wasteful overlap, cameras to tell a crop plant from a weed so that only the weed is sprayed, aerial and ground sensing for precision agriculture, soft grippers for picking fruit, and milking systems that identify each cow by its tag.
- Across all six fields the loop stays the same. What changes is how predictable the workplace is, how closely a person stays involved, and what counts as doing the job well.
- No robot in this chapter understands the place it works in. It has readings and rules, so it fails when something happens that its rules never covered — which is why a person is always somewhere in the picture.
4.10 Key Terms
- Actuator — a part that produces movement or a physical effect when it is given electrical power, such as a motor, a wheel, a gripper or a valve; it is a robot’s output.
- Animatronics — the building of motor-driven figures of animals, creatures or characters that move so as to look alive.
- Autonomous — describes a robot that carries out a whole task on its own from its sensor readings and its program, once it has been started, with nobody steering it.
- Autonomous mobile robot (AMR) — a wheeled robot that carries goods across a floor, finding its own position on a map it has built and planning a path around obstacles.
- Autonomous underwater vehicle (AUV) — an uncrewed underwater craft that follows a stored survey path on its own, used to inspect pipelines, cables and the seabed.
- Boundary wire — a thin current-carrying wire pegged around the edge of a lawn, whose magnetic field a robotic mower detects so that it turns back instead of leaving the lawn.
- Bump sensor — a switch behind a robot’s bumper that closes when the bumper is pushed in, telling the controller that an obstacle has been touched.
- Cliff sensor — a downward-facing sensor that measures the reflection of a beam from the floor, so the robot stops when no reflection returns at a stair edge or drop.
- Controller — the small computer inside a robot that holds the program, takes in sensor readings, applies the stored rules and sends out commands to the actuators.
- End effector — the changeable tool fitted at the far end of a robot arm, such as a welding torch, a gripper or a spray gun, which does the actual job.
- Exoskeleton — a powered frame strapped to the outside of a person’s limbs, whose motors drive the limbs through a movement, used in rehabilitation.
- Gimbal — a motorised camera mount that measures how fast it is being tilted and turns its motors the opposite way, so the camera keeps pointing steadily.
- Global Positioning System (GPS) — a system in which a receiver computes its own position on the Earth’s surface from signals sent by satellites.
- Minimally invasive surgery — operating through a few small cuts using long thin instruments and a camera, instead of one large opening, so that healing is faster.
- Precision agriculture — treating each part of a field according to what that part needs, using sensor and camera measurements, rather than treating the whole field alike.
- Repeatability — how closely a robot returns to exactly the same position each time it is sent there.
- Robot — a machine that takes readings from sensors, applies the rules stored in its program, and moves actuators to carry out a real task.
- Sensor — a part that measures something about the surroundings, such as distance, light or temperature, and turns it into an electrical signal; it is a robot’s input.
- Teleoperated — describes a robot that is worked by a person from a distance, copying that person’s movements and stopping when the person stops.
4.11 Objective Questions
The part of a robot that measures something about the surroundings and turns it into an electrical signal is the
- actuator
- sensor
- end effector
- controller
A floor-cleaning robot’s downward-facing sensors are there mainly to stop it from
- bumping into a chair leg
- running out of charge
- falling down a staircase
- tangling its brush in a cable
Fill in the blank: The closeness with which a robot arm returns to exactly the same position each time it is sent there is called its ______.
Fill in the blank: A robotic lawn mower is kept inside the lawn by a ______ pegged around its edge.
State whether true or false: A surgical robot of the kind used in keyhole surgery carries out the operation on its own once the surgeon has started it.
Match each field in Column A with the robot in Column B that is used in it.
Column A Column B a. Domestic i. Welding arm on a car assembly line b. Industry ii. GPS-guided tractor c. Medical iii. Roomba floor cleaner d. Agriculture iv. Da Vinci Surgical System A surgical system is set so that the instrument tip moves 1 mm for every 4 mm the surgeon’s hand moves. The surgeon moves a hand control 20 mm. In one line, state how far the instrument tip moves.
4.12 Concept Questions
Define an actuator, and name one actuator found in a robotic lawn mower.
Differentiate between a teleoperated robot and an autonomous robot, giving one real example of each.
A welding arm on a car line has no camera. Give the reason why every car door must therefore be clamped in exactly the same position before the arm begins work.
Explain, with three reasons, why an open field is a harder place for a robot to work in than a factory floor.
A field is covered by a sprayer in 15 passes, which meet at 14 boundaries. Each pass is 300 m long. Steering by eye leaves an overlap of 0.4 m at every boundary; satellite guidance leaves 0.1 m. Calculate the area sprayed twice in each case, and the difference between the two.
A thick layer of dust settles over the downward-facing sensors of a floor-cleaning robot so that no reflection reaches them from the floor. Describe what the robot will do when it is switched on, and explain why it behaves that way.
4.13 Case-Based Questions
A school buys a floor-cleaning robot for its library and sets it to run at 6 p.m., after the library closes. The library has a flat tiled floor, rows of reading tables, and a raised reading area reached by three steps at one end. The robot is a basic model with a bumper and downward-facing sensors, and no laser scanner.
- Name the two kinds of sensor the robot depends on in this room, and state exactly what each one measures.
- Describe the path the robot will take between the reading tables, and explain why it will not go neatly around each table leg.
- Explain what will happen when the robot reaches the top of the three steps, and state one precaution the cleaning staff must take before turning the robot over to clear its brush.
A farmers’ group buys a weeding robot that sprays only the plants its program marks as weeds. The program was trained on photographs of fields in another country. Before using it on the whole farm, the group tests it on eight plants in one row and records the results below.
Plant Marked by the robot as What the plant actually was 1 Crop Crop 2 Weed Weed 3 Crop Weed 4 Weed Crop 5 Weed Weed 6 Crop Crop 7 Weed Crop 8 Crop Crop - From the table, state how many plants the robot marked wrongly, and how many of the group’s crop plants would have been sprayed.
- Explain why a program trained on photographs of fields in another country may make these mistakes on this farm.
- Suggest one change to how the robot is set up that would reduce the harm done when the program is uncertain about a plant, and give the reason for your suggestion.