2The parts of a robot

Look at your own arm and name what it is made of. There are bones, which are stiff and do not bend. There are places where two bones meet and can move — your shoulder, elbow, wrist and knuckles. There are muscles that pull the bones around those places. There are nerves that tell your brain where everything is. And there is a brain.

A robot arm is made of exactly the same five kinds of thing, with less poetic names. This module is a tour of them on a real robot: the SO-101, a small arm that people build for a few hundred dollars and that research labs use to teach robots new tricks. The one in the box is drawn from the same files the physical arm is built from. The tour goes in a fixed order, and each stop has something to find on the arm.

Links: the bones

A link is one rigid piece of a robot's body. It does not bend or stretch; whatever happens to one end of a link happens to the other. The arm below has four: the base, which is bolted to the table and never moves; the upper arm; the forearm; and the gripper at the end, which is the arm's hand.

They form a chain from the base out to the gripper, and it helps to think of the chain as flowing that way: the base end is upstream, the gripper end downstream. If the base is still and the elbow moves, the forearm swings but keeps its shape. Everything downstream of a moving joint moves as one lump; everything upstream stays put.

Loading the simulation. The robot runs in your browser, so the first one takes a few seconds to arrive.

ResetPauseSlowLabels

This is easier on a larger screen, but it works here too.

Joints: where the bones meet

A joint is the place where two links meet and can move relative to each other. The arm has six, one per slider, and each is a hinge: it lets its two links turn around one line, the way a door turns around the line through its hinges, and nothing else.

Each joint has a plain name here. Five of them move the arm. The shoulder turn swings the whole arm left and right on the base. The shoulder lift tips the upper arm forward and back. The elbow bends the forearm. The wrist bend tips the hand up and down, and the wrist twist rolls it around the forearm's length. The sixth, the gripper jaw, does not move the arm at all; it swings one jaw against the other so the hand can hold something.

Actuators: the muscles

Bones and hinges make a skeleton, and a skeleton does not move by itself. An actuator is anything that makes a robot move. Some work with air pressure or hydraulic fluid, but on most arms, and on this one, every actuator is a small electric motor, exactly one per joint.

On the SO-101 each motor is a black box the size of a matchbox, called a servo. Inside are the motor itself, which spins fast but weakly, and a stack of gears. A small wheel turning a bigger wheel turns it more slowly but pushes harder, the trade you make by shifting a bicycle into a low gear for a hill, and the stack turns a fast, weak spin into a slow, strong one. Out of the box pokes a short rod that turns: the output shaft. The box bolts to one link and the shaft to the next, so when the shaft turns, the joint turns. Under "Show: Motors" the elbow motor is marked where it sits, on the upper arm at the elbow, and dragging the Elbow slider swings the forearm from it. There is a third thing in the box, and it belongs to the next stop.

Sensors: the nerves

When you close your eyes you still know where your hand is, because nerves in your joints report their angles. A robot has no such sense unless it is built in. A sensor is a part that measures something and reports it as a number: an angle, a force, a distance.

The sensor this arm relies on most is hidden. Inside every servo, on the output shaft, is a small device that measures how far the shaft has turned and reports the angle a few hundred times a second. That is how the arm knows where its own elbow is.

The third thing in the servo's box uses it: a small circuit that reads the sensor, compares the angle with the one it was told to hold, and nudges the motor until the two agree. Ask a servo for forty degrees and it gets there and stays there on its own, because it can feel where it is.

The numbers next to the sliders in the box are those sensor readings ("Show: Sensors" marks where each one sits). The slider is what you asked for; the number is where the joint actually is. In this simulation the sensor is exact, and the motors are strong enough that the two agree almost perfectly. A real sensor is off by a fraction of a degree, and a real motor loses a little to the weight of the arm; later modules make a great deal of both.

The controller: the brain

Each servo can hold an angle, but something has to decide what the angles should be, so that six of them add up to reaching for a cup. That something is the controller: the robot's computer. It might be a chip the size of a stamp, a small board tucked into the base, or a laptop on the bench connected by a cable. What makes it the controller is the job, not the size.

Its loop is short and never stops: read every sensor, compare what the arm is doing with what it should be doing, send each motor a new angle, go round again, many times a second. It is the read-compare-nudge loop each servo runs for one joint, done for the whole arm at once. Every robot in this course runs some version of it, and Part 3 is about the "compare" step.

What you found

You looked for a computer on the arm and found none. That is right: the real SO-101 has no controller of its own. A cable runs out of the base to a laptop on the bench, and the laptop is the controller. It reads the angle back from every servo, decides what the angles should be next, and sends them down the cable.

The simulated arm on this page has no laptop, so who is its controller? You are. Each slider sets the angle one servo is told to hold, which is exactly the instruction a controller sends. Move a slider and you are doing the controller's job by hand, one joint at a time. The simulated servo then does its part: it reads its sensor and nudges its motor until the readout matches what you asked for.

So the five parts, once more, in the order you met them: links are the rigid pieces; joints are where two links meet and can move; actuators make joints move, and here they are motors; sensors measure what happened; and the controller reads the sensors and commands the motors. Every robot you will meet in this course, arms and rovers and drones and legs, is these five things arranged differently. When something looks baffling, find the five and it will usually stop being baffling.