Glossary

Every term this course defines, in alphabetical order. Each links to the module that introduced it, which is where it is explained in full, with something to play with.

A*: A search over a grid map that always looks next at the frontier cell whose distance so far plus its straight-line guess of the rest is smallest, so it reaches toward the goal instead of spreading everywhere, and still finds the shortest path. Said "A star". From Planning a path around obstacles
acceleration: How quickly a thing's motion is changing, in speed or in direction. Any change of motion takes a push, and the bigger the change and the shorter the time it happens in, the bigger the push. An instant change would need an infinite one. From Trajectories: getting from A to B smoothly
actuator: Anything that makes a robot move. On most arms every actuator is an electric motor, one per joint. From The parts of a robot
balance point: The one spot in a body through which its whole weight can be taken to pull straight down. Hold a thing up under that spot and it balances; anywhere else and it tips. From Legs and balance: why walking is hard
ball joint: A joint that lets a link turn every way about one fixed center, like a shoulder or a hip. Three numbers say where it is, so it is worth three degrees of freedom. From Joints: the ones that turn and the ones that slide
base of support: The patch of floor enclosed by whatever is touching it. A body stays up while the line straight down from its weight lands inside that patch, and topples when it lands outside. From Legs and balance: why walking is hard
behavior cloning: The simplest kind of imitation learning. A policy is adjusted until, given what the robot sensed at any instant of a demonstration, it gives back close to what the demonstrator did next. From Teaching by showing: imitation learning
behavior tree: A way of writing the same decisions as a list of behaviors in order of priority, checked from the top every trip round the loop, so that the robot runs the first behavior on the list whose condition holds; a behavior that takes time keeps its own count of when it is done. A rule like "if bumped, back up" is written once, at the top, instead of once for every state it could interrupt. From Deciding what to do: state machines
calibration: Putting a sensor's numbers against something already known, so they can be read. For an encoder, moving the joint to a known pose and writing down the count there; every later reading is measured from it. From Knowing where your own joints are: encoders
camera: A sensor that catches the light arriving from a wedge of directions in front of it and reports, for each of a grid of small patches across that wedge, how much red, green and blue light came in. It hands a robot a picture made of numbers, and nothing else. From Cameras: what a robot actually sees
cart-pole: A cart on a rail with a pole standing on it on a free pivot, kept up only by the cart moving underneath it. The standard test rig for balancing controllers, and the one this course uses. From Stability: when control goes wrong
caster: A wheel or ball that is not driven and not steered. It carries weight and follows whatever the driven wheels do. From Wheels: how rovers turn with no steering wheel
closed loop: Doing something by acting, measuring what happened, and using that measurement to decide the next action, over and over until the job is done. From Open loop, closed loop: the idea of feedback
collaborative robot: A robot arm built to work beside people with no fence. It is light, rounded and slow, and it stops when its motors feel a touch, so that the hit it could give is small and the push after it is short. From Safety: limits, stops, and why robots are slow near people
collision check: Asking, for one pose, whether any part of the robot would overlap anything, including another part of itself, by placing every link where the angles say and testing the shapes against each other. Planning for an arm is mostly collision checks, thousands of them. From Planning for arms: moving without hitting yourself
command: One line of a program that tells the robot to do one thing. It names the action, then in brackets what the action applies to, such as which joint and what angle. A command that takes time, like moving a joint, is finished only when the robot has done it, and the next command does not start until then. From Telling a robot what to do: commands, loops, timing
compliance: Giving way when pushed. A compliant joint is a soft one, so that something in its path moves it instead of being crushed by it. Robots that touch people, fragile things, or the ground are made compliant on purpose. From Stiff or soft: holding an egg without breaking it
computer vision: The work of turning a camera's grid of numbers into things and where they are, from finding a patch of one color to telling a cup from a hand. From Cameras: what a robot actually sees
condition: A test in a program that comes out true or false, with some lines that run only when it is true. Written as if, a test in round brackets, and the lines in curly brackets; when the test is false the lines are skipped and the program goes on after them. From Writing the sense-think-act loop yourself
confidence: The share a recognizer gives the answer it chose, from 0 to 100%. It says how strongly the picture drives the network's sums toward that answer, which is not the same as the answer being true. From Seeing and recognizing: neural networks for perception
configuration space: A map of a robot's poses rather than of the room, with one axis per joint, so that every point on it is one set of joint angles and a move of the arm is a line across it. An obstacle in the room appears on it as the set of every pose in which some part of the robot would be inside that obstacle. From Planning for arms: moving without hitting yourself
controller: The computer that reads the sensors, decides what to do, and tells the actuators to do it. From The parts of a robot
current: The flow of electricity through a wire or a motor. A motor pushes harder when more flows, so its current is a measure of its push. From Touch and force: feeling the world
dead reckoning: Keeping track of where you are by starting from a place you knew and adding up every move you have made since, with nothing checking the sum against the world. From Odometry: where am I, and why wheels lie
degree of freedom: One independent way a thing can move — one number you have to give to say where it is. A robot arm usually has one per joint. From Degrees of freedom
demonstration: One recorded run of a person doing a job through the robot, by moving its arm or by driving it from a distance. What is kept is where every joint was, many times a second, from start to finish, and on a real robot what its cameras saw. From Teaching by showing: imitation learning
derivative: How fast something is changing right now. In a controller, the derivative is the joint's speed, and the push made from it leans against the movement, harder the faster the joint is moving. From Push, patience and a brake: what PID stands for
differential drive: Steering by wheel speed alone. Two wheels on the same line across the body, each with its own motor; the same speed on both goes straight, a difference turns toward the slower wheel, and opposite speeds spin the body on the spot. From Wheels: how rovers turn with no steering wheel
disturbance: Anything that pushes a machine away from what its commands would have done on their own, such as a slippery floor, a gust of wind, or a heavier load than expected. From Open loop, closed loop: the idea of feedback
domain randomization: Running a plan in many simulated worlds at once, each with its numbers shaken a little, such as the block's grippiness anywhere from a quarter to double, and accepting a plan only if it works in all of them. A plan that survives a crowd of made-up worlds usually survives the real one. From Simulation vs reality: why sim is used, and how it lies
drift: The slow wandering of a robot's belief about where it is away from where it really is, as small errors pile up with nothing to correct them. From Odometry: where am I, and why wheels lie
emergency stop: The red button on a robot and on the wall beside it that cuts the motors and puts the brakes on, whatever the program is doing. It needs a person to notice the danger and press it, and a surprised person takes longer over that than a fast arm takes to arrive. From Safety: limits, stops, and why robots are slow near people
encoder: The sensor inside a servo that reports how far its shaft has turned, as a whole number of steps around the circle. There is one per joint, and it is the only way the robot knows where that joint is. From Knowing where your own joints are: encoders
end effector: Whatever is fitted to the end of a robot arm to do the job. A gripper, a suction cup, a welding torch, a camera. The arm's job is to put it somewhere; the end effector's job is everything after that. From Grippers and hands: holding on
episode: One try, from the moment everything is put back at the start to the moment the try ends. A learner's practice is cut into episodes so that it meets the beginning again and again. From Learning by trial and error: reinforcement learning
error: The gap between where a machine wants to be and where its sensors say it is. In control, "error" means only that gap, not a mistake. From Open loop, closed loop: the idea of feedback
exploration: Doing something other than what looks best so far, on purpose, to find out whether it is better than it looked. A learner that never explores keeps the first habit that worked at all. From Learning by trial and error: reinforcement learning
feedback: A measurement of what a machine's actions have actually done, fed back into the decision about what to do next. From Open loop, closed loop: the idea of feedback
field of view: The wedge of the world a camera takes in at once. Anything outside it is not in the picture at all, however close it is. From Cameras: what a robot actually sees
filter: Anything that cleans up a sensor's readings before the robot acts on them. A moving average is the simplest filter there is. From Noise: why every sensor lies a little, and how averaging helps
force sensor: A sensor that reports how hard it is being pushed, in newtons. Fitted in a robot's jaw or at its wrist, it is how the robot feels what it touches. From Touch and force: feeling the world
forward kinematics: Working out where the hand is from the angle at every joint and the length of every link, stepping along the arm from the base outward. There is always exactly one answer. From Where is the hand? Forward kinematics
frame: A point to count from and three directions to count along, usually forward, left and up. Every position is a set of numbers measured in some frame, and the same place gets different numbers in different frames. From Frames: where "left" is depends on who you ask
friction: The force that resists two touching surfaces sliding across each other. Its size is the squeeze between the surfaces times a number that says how grippy the pair is, near 1 for rubber on most things, about a fifth for hard plastic on hard plastic, and far smaller for anything wet or polished. From Grippers and hands: holding on
frontier: The edge of a spreading search. The cells that have been reached and given a distance from the start but not yet looked at themselves. The search grows by taking one cell off the frontier and adding its neighbors. From Planning a path around obstacles
gain: How much push a controller gives for each unit of error. In proportional control it is the one number you can change, and it works like the stiffness of a spring pulling toward the target. From Proportional control: push harder the further you are
gait: The pattern in which a walker lifts and places its feet, repeated step after step. From Legs and balance: why walking is hard
gear ratio: How many turns the motor makes for one turn of what it drives. A ratio of 4 gives four times the torque at a quarter of the speed. Nothing is gained for free. From Motors: force, speed, and gears
generalization: Doing the right thing in a situation that was not among the examples. A policy made from demonstrations generalizes between the things it was shown, and cannot be trusted beyond them. From Teaching by showing: imitation learning
grip force: How hard a gripper's jaws squeeze what is between them, in newtons. It comes from the jaw motor's torque, and it is what friction has to work with. From Grippers and hands: holding on
gripper: An end effector that holds things by closing on them. Most have two jaws driven by one motor, and hold on by squeeze and friction. From Grippers and hands: holding on
grounding: Tying a word to the actual thing or action it stands for, here and now. The word "block" is grounded when the robot knows which object on the table it means and where that object is. From Robots that understand language, and what is real today
heading: The direction a robot's nose points, as an angle on the floor. From Odometry: where am I, and why wheels lie
heuristic: A quick guess at how far a cell still is from the goal, made without looking at the map, used to steer a search toward the goal. The usual one is the straight-line distance, which is never more than the true distance left. From Planning a path around obstacles
hinge: A joint that lets two links turn around one line and do nothing else. One number, an angle, says where it is. Engineers call it a revolute joint. From Joints: the ones that turn and the ones that slide
hover: Holding still in the air, with the lift exactly matching the weight and shared so that no corner is lifted more than another. From Flying: four motors and a lot of math the robot does for you
imitation learning: Building a robot's behavior out of demonstrations instead of writing it down. A person shows the job several times, and a rule for doing it is made from the recordings. From Teaching by showing: imitation learning
impedance control: Controlling a joint by choosing how it should feel when something pushes back on it. The joint is given a spring of chosen stiffness toward its target, with a damper beside it, and the push it makes on whatever stops it is the stiffness times how far short it is held. From Stiff or soft: holding an egg without breaking it
integral: A total that keeps adding up over time. In a controller, the integral of the error is a running total of how far off the joint has been and for how long, and the push made from it grows for as long as any gap lasts. From Push, patience and a brake: what PID stands for
inverse kinematics: Working out the joint angles that put the hand at a chosen place. The opposite direction from forward kinematics, and unlike it, there may be two answers, a whole family, or none. From Getting the hand there: inverse kinematics
inverted pendulum: Anything balanced above the point that holds it up, so that a lean pulls it further over instead of back. A broom on your palm, a pole on a cart, a person standing, a rocket on its engine. Keeping one up is a job that never ends. From Balancing something that wants to fall
joint: The place where two links meet and can move relative to each other. Most joints on an arm are hinges. From The parts of a robot
joint limit: The furthest a joint can go in each direction — the two ends of its travel. On a robot, the motor is told never to go past them. From Joints: the ones that turn and the ones that slide
label: The right answer attached to a training example, such as the words "red block" attached to a picture of one. A network can only ever answer with the labels it was trained on. From Seeing and recognizing: neural networks for perception
lag: How far behind the world a filtered reading is. An average of the last two seconds of readings describes, roughly, where things were a second ago. From Noise: why every sensor lies a little, and how averaging helps
large language model: A neural network with billions of weights, trained on a large share of what people have written, to predict what comes next in a text. It can take sentences worded in very different ways as the same request, and can break a job into steps. From Robots that understand language, and what is real today
latency: The time between something happening and a controller acting on it, from the sensor reading it, through the wire and the computer, to the motor responding. Every real controller acts on slightly old news. From Stability: when control goes wrong
lidar: A rangefinder that sends a flash of laser light and times its return. The flash is a pencil-thin line, so the number belongs to one exact direction; spun round on a motor, one lidar draws a floor plan of a room as dots. From Measuring distance: sonar and lidar
lift: The four thrusts of a quadrotor added together. They all point the same way, straight up out of the top of the body, so the lift points wherever the body's top does. From Flying: four motors and a lot of math the robot does for you
link: One rigid piece of a robot's body. It does not bend; joints at its ends move it as a single lump. From The parts of a robot
localization: Working out where you are on a map you already have, from what your sensors report, and keeping that answer right as you move. From Maps, and knowing where you are on one
loop: A part of a program that runs the same commands again and again. A counted loop runs them a set number of times; a loop with no count runs them until something stops it. From Telling a robot what to do: commands, loops, timing
loop rate: How many times a second a loop goes round. The robot notices the world only when its loop comes round, so the loop rate sets how far the world can change before the robot can respond. From Writing the sense-think-act loop yourself
map: What a robot carries of the place it works in, usually a grid of small cells each marked wall or floor, so that from any spot on it the robot can work out what its sensors ought to see. From Maps, and knowing where you are on one
mass: How much stuff a thing is made of, measured in kilograms. It is the same on Earth, on the Moon and in orbit, and it is what makes a thing hard to get moving and hard to stop. From Gravity and momentum: why arms sag and overshoot
message: One packet of information sent on a topic, such as five distances or a pair of wheel speeds, made whenever the sending node has something new. A node that listens to the topic gets every message sent on it, and does its job each time one arrives. From How real robot software is organized: nodes and messages
model: The description a simulation works from, its list of numbers for one robot and the things around it. A model of a block says how big, how heavy and how grippy the block is, and each of those numbers can be wrong. From Simulation vs reality: why sim is used, and how it lies
momentum: What a moving thing carries by moving, its mass times its speed. Taking it away needs a force applied for a time, and the more momentum, the more force or the more time. From Gravity and momentum: why arms sag and overshoot
motion planning: Path planning for a robot with joints, done in configuration space, so that the route found is a sequence of poses that keeps the whole robot clear of everything, not only its hand. From Planning for arms: moving without hitting yourself
moving average: The average of a sensor's last few readings, worked out again every time a new one arrives, the newest reading coming in and the oldest dropping out. From Noise: why every sensor lies a little, and how averaging helps
network weight: One of a neural network's adjustable numbers, called a weight for short. It says how much, and in which direction, one input counts toward one of the network's sums. Everything a network has learned is in its weights. From Seeing and recognizing: neural networks for perception
neural network: A formula built from many small sums stacked in layers, each sum multiplying its inputs by adjustable numbers called weights. Nobody writes what it computes. Its weights are set by training on examples. From Seeing and recognizing: neural networks for perception
no-load speed: The fastest a motor can turn under its own power, which it reaches only with nothing to push against. At that speed it has no torque left to give. From Motors: force, speed, and gears
node: One small program on a robot that does one job, such as reading a sensor or working out wheel speeds, and runs on its own, at its own pace, alongside the others. A robot's software is many nodes running at once. From How real robot software is organized: nodes and messages
noise: The random wobble in a sensor's readings, a little high one time and a little low the next, with no pattern to it, so that the same distance never reads quite the same twice. From Noise: why every sensor lies a little, and how averaging helps
odometry: Working out how far a robot has moved, and which way, from how far its wheels have turned. The word means "road measuring". From Odometry: where am I, and why wheels lie
open loop: Doing something by following a plan made in advance, without checking along the way whether it is working. From Open loop, closed loop: the idea of feedback
oscillation: Swinging back and forth past the target, again and again. The swings of an arm arriving at its star are an oscillation; so is a pole rocking on its cart. From Stability: when control goes wrong
overshoot: How far past its target a moving thing travels before it is stopped. From Gravity and momentum: why arms sag and overshoot
particle filter: A way of localizing with a crowd of guesses about where the robot is. Every guess is moved as the wheels say the robot moved, weighed by how well the map would explain the latest sensor readings from there, and the crowd is redrawn from its best-fitting members, so the guesses pile up where the robot is. From Maps, and knowing where you are on one
path planning: Working out a route from where the robot is to where it wants to be, on its map, before it moves. On a grid map the route is a chain of floor cells, each touching the next, from the start cell to the goal. From Planning a path around obstacles
PID control: Control by adding three pushes together, proportional, integral and derivative. P pushes toward the target, harder the further away; I adds patience, pushing harder the longer a gap has lasted; D is a brake, pushing against movement. From Push, patience and a brake: what PID stands for
pixel: One patch of a camera's grid, the smallest piece of the picture, reported as three numbers, the amounts of red, green and blue light that landed on it. From Cameras: what a robot actually sees
policy: A robot's rule for turning what it senses into what it does. It can be written by a person, as everything that steered a robot earlier in this course was, or made from demonstrations. From Teaching by showing: imitation learning
potential field: A way of writing a reactive rule as pushes and pulls added up. The goal pulls the robot toward it, everything the sensors see near pushes it away, harder the nearer, and the robot moves the way the sum points, as a ball rolls downhill on a landscape with the goal at the bottom and a hill on every obstacle. From Dodging things on the fly
program: A list of commands, written down in advance, that a robot's computer carries out one after another from the top. The robot does what the list says, in the order it says, and nothing the list does not say. From Telling a robot what to do: commands, loops, timing
proportional control: A closed-loop rule that pushes in proportion to the error. Twice the gap gets twice the push, and no gap gets no push at all. From Proportional control: push harder the further you are
protective stop: A stop the robot triggers itself, from its own sensors, with nobody pressing anything. A beam of light broken by someone walking in, or a joint that stops short while its motor is still pushing, stops the arm. From Safety: limits, stops, and why robots are slow near people
quadrotor: A flying robot with four propellers on fixed arms, one at each corner of a cross, steered by nothing but how hard each propeller spins. From Flying: four motors and a lot of math the robot does for you
radian: A unit of turning. Something going round a circle has turned one radian when it has traveled one radius's length along the circle; that is about 57 degrees, and 6.3 radians make a whole turn. From Wheels: how rovers turn with no steering wheel
rangefinder: A sensor that sends something out, times how long it takes to come back, and reports one number, the distance along its line to the first thing that sent the signal back, or that nothing came back within its reach. From Measuring distance: sonar and lidar
rapidly exploring random tree: A planner that grows a tree from the start by throwing random points on the map and growing a short branch toward each from the nearest tip, dropping any branch that would cross a wall, until a branch comes within reach of the goal. RRT for short. Its route is crooked and not the shortest, and it needs no grid. From Planning a path around obstacles
reactive control: Steering a robot by a rule that turns this instant's sensor readings straight into a command for the motors, run every trip round the sense-think-act loop, with no map, no plan and nothing remembered from the trip before. From Dodging things on the fly
reality gap: The difference between what a simulation predicts a robot will do and what the real robot does. It comes from numbers the simulation was given wrongly and from things it leaves out altogether. From Simulation vs reality: why sim is used, and how it lies
reinforcement learning: Learning what to do by trying things and keeping what scores well. Nobody shows the robot the job or writes down how to do it. It acts, is handed a score, and adjusts, thousands of times over. From Learning by trial and error: reinforcement learning
resolution: The smallest change a sensor can tell apart. For an encoder it is one tick; a smaller turn leaves the count where it was. From Knowing where your own joints are: encoders
reward: The score a learning robot is handed as it goes, bigger when things are going the way its builder wants. It says how well things went, never what to do, and it is the only description of the job the robot ever gets. From Learning by trial and error: reinforcement learning
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. From What makes something a robot?
saturation: When a controller asks for more than the motor can give. The motor gives all it has, and until the request falls back within reach, changing the controller changes nothing. From Proportional control: push harder the further you are
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. From What makes something a robot?
sensor: A part that measures something and reports a number to the robot's computer — an angle, a force, a distance. From The parts of a robot
sim-to-real transfer: Taking a plan or a controller that works in simulation and getting it to work on the real robot. Most of the work is measuring the real world's numbers, putting them into the simulation, and testing again, on the real robot, before trusting it. From Simulation vs reality: why sim is used, and how it lies
simulation: A computer working out, step by step, what a described world would do. The description is a list of numbers, such as each part's shape and mass, how grippy each surface is and how strong each motor is, and the rules are physics. The simulation is only as true as the numbers it was given. From Simulation vs reality: why sim is used, and how it lies
skill: Something a robot can already do from start to finish when asked, such as picking a thing up or opening a drawer. A skill may be a program somebody wrote or a policy made from demonstrations. Either way it exists before anyone asks for it. From Robots that understand language, and what is real today
slide joint: A joint that lets two links slide along one line and do nothing else. One number, a distance, says where it is. Engineers call it a prismatic joint. From Joints: the ones that turn and the ones that slide
software limit: An angle written into the controller that a joint's target may never pass, wherever the program tries to send it. It keeps the arm out of a place, such as where a person stands, at the cost of the arm never being able to work there. It has to be set short of that place, because a moving arm swings a little past its target before it stops. From Safety: limits, stops, and why robots are slow near people
sonar: A rangefinder that sends a click of sound too high to hear and times the echo. Cheap and slow, and its click spreads into a cone, so it sees a wide patch but cannot say where in the patch the echo came from. From Measuring distance: sonar and lidar
stable: A controlled system is stable when a knock dies away on its own, each swing smaller than the last, and it comes back to where it was. From Stability: when control goes wrong
stall torque: The most torque a motor can give, which it gives only when standing still. A load that asks for more stalls it, and a heavier one pushes it backward. From Motors: force, speed, and gears
stance: The part of a step during which a foot is on the ground, carrying weight and pushing the body along. From Legs and balance: why walking is hard
state: The handful of numbers that say everything about where a system is and how it is moving right now. For the cart-pole there are four, the pole's lean and how fast it is changing, and the cart's position and speed. From Balancing something that wants to fall
state machine: A way of writing a robot's brain as a short list of states, each one thing the robot does, with rules for leaving each state for another. The robot is in exactly one state at a time, does what that state says, and every trip round the sense-think-act loop checks only whether one of the rules leaving its current state has come true. From Deciding what to do: state machines
steady-state error: The error that is left once everything has stopped moving. Under proportional control, it is whatever gap is needed to earn the push that holds the load. From Proportional control: push harder the further you are
stiffness: How hard a spring pushes back for each unit it is bent. For a joint under impedance control, the push it makes for each degree it is held away from its target. A stiff joint holds its position against anything; a soft one gives. From Stiff or soft: holding an egg without breaking it
stopping distance: How far a moving robot travels between being told to stop and standing still. Its momentum has to be taken away first, so the faster it was going, the further it goes. From Safety: limits, stops, and why robots are slow near people
strain gauge: A thin zigzag of metal foil glued to a part that bends a hair under load. Stretched, the foil passes electricity a little less easily, and that change is read as a force. The sensing element inside most force sensors. From Touch and force: feeling the world
swing: The part of a step during which a foot is in the air, on its way to where it will land next. From Legs and balance: why walking is hard
tactile sensor: A patch of many tiny pressure sensors, like a skin, that reports not only how hard something presses but where, and in what shape. From Touch and force: feeling the world
teleoperation: Driving a robot from a distance while it does the job, often by moving a second arm of the same shape that the robot copies joint for joint. It is the usual way demonstrations are recorded. From Teaching by showing: imitation learning
threshold: The number a condition compares a reading against, such as 0.4 meters. On one side of it the program does one thing, on the other side another, so where the threshold sits decides how early the robot reacts. From Writing the sense-think-act loop yourself
thrust: The push a propeller gives by throwing air one way and being pushed the other. Spin it faster and the thrust grows. From Flying: four motors and a lot of math the robot does for you
tick: One step of an encoder's count, and so the smallest turn it can report. The SO-101's servos cut a full turn into 4,096 ticks, so one tick is 0.088°. From Knowing where your own joints are: encoders
time of flight: How long a rangefinder's click or flash takes to fly to the object and back. Halve it, multiply by the speed of sound or light, and that is the distance. From Measuring distance: sonar and lidar
tool frame: The frame that rides on the robot's hand, or on whatever tool the hand holds. Its forward is wherever the hand is pointing right now. From Frames: where "left" is depends on who you ask
topic: A named channel that messages travel on, such as /distances. A node sends on a topic without knowing who is listening, and any node can listen to a topic without the sender knowing, so a topic is the only thing two nodes have to agree on. From How real robot software is organized: nodes and messages
torque: Turning force. How hard a motor twists its shaft, or how hard gravity twists an arm about its joint. Measured in newton meters, written N·m, the twist of a one-newton pull at the end of a one-meter lever. From Motors: force, speed, and gears
training: Setting a neural network's weights by showing it examples whose right answers are known, over and over, and nudging every weight a little toward the right answer each time. From Seeing and recognizing: neural networks for perception
trajectory: A plan for where something should be at every moment on its way from one place to another. The path through space and the timing along it, together; two trajectories can follow the same path at different timings. From Trajectories: getting from A to B smoothly
transition: One rule of a state machine, an arrow on its diagram. It names the state it leaves from, the condition the robot must sense for it to fire, and the state it goes to. The transitions leaving a state are checked in order, and the first whose condition holds is the one that fires. From Deciding what to do: state machines
trot: The gait in which diagonal pairs of legs move together, front-left with rear-right and front-right with rear-left, one pair on the ground while the other swings. From Legs and balance: why walking is hard
tuning: Choosing the sizes of a controller's pushes, by trying a size, watching what happens and adjusting. Too small and a push does not do its job; too big and it fights the others. From Balancing something that wants to fall
twist: A push that turns a body instead of moving it along. A twist does not set an angle; it sets the body turning, and the body keeps turning after the twist has gone until another twist stops it. From Flying: four motors and a lot of math the robot does for you
unstable: A controlled system is unstable when a disturbance grows on its own, each swing bigger than the last, until something falls, hits a limit or breaks. From Stability: when control goes wrong
vision-language-action model: A neural network that is given a camera's picture and a sentence, and gives back the robot's next short stretch of motion. It is a policy trained by behavior cloning on demonstrations that each came with a sentence describing the job. From Robots that understand language, and what is real today
waypoint: A place a trajectory must pass through on its way. The plan decides how the hand gets from each one to the next. From Trajectories: getting from A to B smoothly
weight: The pull of gravity on a mass. On Earth, about ten newtons for every kilogram; less where gravity is weaker, and nothing in orbit. From Gravity and momentum: why arms sag and overshoot
workspace: Every place a robot can put its hand, taken all together. A solid shape in the air around the robot, set by how long its links are and how far its joints can turn. Also called the work envelope. From Reach: where the hand can and cannot go
world frame: The frame that never moves. On an arm bolted to a table it sits at the base, and everything around the robot is described in it. From Frames: where "left" is depends on who you ask