1What makes something a robot?
A machine that notices, decides and does — over and over. Switch each one off and see what is left.
This assumes nothing. About 6 minutes of reading.
A washing machine spins, heats, pumps and beeps. It has a motor, a computer and a program. Nobody calls it a robot. A robot vacuum cleaner has a motor, a computer and a program too, and everybody does. What is the difference?
Watch the vacuum for a minute. It rolls forward, meets a chair leg, and turns. It did not have the chair leg in its program; it found it. Its bumper is a sensor, a part whose only job is to notice something and report it, and the bumper reported a bump. The machine noticed, decided to turn, and turned. Then it did all of that again for the next chair leg. The washing machine never does anything like it. It runs its program from the top, the same way every time, whatever is happening around it.
That noticing is the whole difference, and it has a name. The sense-think-act loop is the cycle a robot runs without stopping: take in what the sensors report, decide what to do about it, do it, and go round again. Sense: the bumper is pressed. Think: a pressed bumper means turn. Act: the wheels turn the machine. Then straight back to sense. The loop runs many times a second, and every trip round it can come out differently, because the world outside is different.
So here is a working definition. A robot is a machine that runs the sense-think-act loop on its own: it notices what is around it, decides, and acts, and keeps doing so without a person in the loop. The washing machine acts without sensing much. A remote-controlled car only acts: the sensing and the thinking both happen in your head, and the car carries out what you decided. A robot does all three by itself.
The best way to understand a loop is to break it. The box below has a small wheeled robot in a walled pen, a yellow post called the beacon that it is trying to reach, and a brown box in the way. Its brain is a short script: head for the beacon, and if the distance sensors on its nose see something close ahead, steer round it. It opens paused, so nothing happens until you press Play. Two switches under the box let you cut its senses or its thinking. Under them, one line says what the brain is doing: how far the beacon is, how near the closest thing is, and what the wheels are being told. The line below that is the raw sensor readings: five distances in meters, from far left to far right, with a dash where a sensor sees nothing in range, and whether the nose is pressed against anything.
This is easier on a larger screen, but it works here too.
What you saw
With everything on, the rover set off toward the beacon, and the "ahead" number fell as the box came closer. At about forty centimeters, the brain's rule fired: something close ahead, so steer round it. The rover swung away, the number climbed again, the rule stopped firing, and the brain went back to heading for the beacon. Nothing about the box was in the script. The rover found it, every time round the loop.
With sensing off, the brain still ran, but it was fed nothing. Its rule is "head for the beacon", and heading for something means turning toward it; with no reading of where the beacon is, there is nothing to turn toward, so the script's turn came out as zero and the wheels got the plain instruction "go". The rover went in a straight line until the wall stopped it, and then kept pushing. That is a machine acting without sensing, like the washing machine with wheels. It is not stupid; it is blind. Every decision it made was correct for the world it was told about, which was no world at all. A different script might have stopped instead; this one was written to go.
With thinking off, the sensors kept reporting: the same five distances, over and over, because nothing moved. Sensing without thinking is a set of readings that nobody acts on. The rover sat still until you turned the brain back on.
Take away either one and what is left is not a robot. Sense and act with no thinking is a sensor with wheels. Think and act with no sensing is a program on a timer. The loop needs all three, and it needs them in a loop: sense, decide, act, and sense again, because the world just changed.
Drawing the line
The definition sorts most machines quickly. A thermostat senses the temperature, decides the room is cold, and switches the heating on: a tiny loop, so in a thin sense it is a robot, and some people would say so. A dishwasher runs its cycle whatever is inside it: no loop. A drone holding still in a gust senses that it has drifted and corrects: a robot. A drone you fly by hand is a remote-controlled aircraft, and the robot is you.
The line is blurry at the edges, and that is fine. The useful question is never "is this officially a robot" but "where is its loop, and how much of the world does it sense?" A robot vacuum senses a bump, and a drop ahead like the top of a stair. A self-driving car senses a street. The loop is the same shape; what differs is how much goes into sense and how much the think step has to work out.
Two honest notes
First, "think" here is a small word for a small thing. The brain in the box is a few lines of rules: head that way; if something is close, turn. It does not understand posts or boxes. Later modules build the think step up from rules to plans to learned behavior, and none of it is magic; it is more of the same, done better.
Second, the rover knows where the beacon is because this simulation hands it the direction, as if the beacon were broadcasting a signal it could pick up. A real robot has to work that out from a camera or a map, and later modules on sensing and on maps are about how. The distance sensors on its nose, though, are real: they measure straight-line distance to whatever is in front of them, and that is exactly what a cheap robot vacuum uses.
You now have the one idea everything else in this course hangs off. Every module from here is about one part of the loop, done on a real robot: what the sensors are and how they lie, how the thinking is done, how the acting is done, and how the three are wired together.
Watch
How Robots Sense, Plan, and Act — World Science Festival (1:56 of it, starting at 0:10)
- Why this video
- A working roboticist, Chad Jenkins of Brown University, says the loop in his own words over footage of real research robots doing it — sensing a scene, choosing among possible actions, moving. It is the same three steps as the words, said by someone whose job is building them.
- What to watch for
- He calls the middle step "plan" where this module says "think": same step, and his version says what it is — the robot trying out possible sequences of actions in its head, the way a map app tries routes. And at the very end, the line that matters: "this loop goes around over and over and over."
Words introduced
- sense-think-act loop
- The cycle a robot runs without stopping — take in what its sensors report, decide what to do, do it, and go round again.
- robot
- A machine that runs the sense-think-act loop on its own — it notices the world, decides, and acts, and keeps doing so without a person in the loop.
Where next
Suggested next: 2. The parts of a robot
