How Did Robot Vacuums Learn to Find Their Way Around a Room?

1. The First Home Robot Vacuum Already Sensed the Room With Sound

Stanford University's AI100 report, a long-running study of artificial intelligence, says that in 2001, after many years of development, the Electrolux Trilobite became the first commercial home robot. It was a vacuum-cleaning robot with a simple control system for avoiding obstacles, plus some navigation [1].

A design museum in Munich, which holds one in its collection, adds the details. The Trilobite sent out ultrasonic waves and listened for the echo, much as bats find their way [2]. Ultrasound is sound pitched too high for people to hear. The longer an echo takes to come back, the farther away the thing it bounced off.

The museum also notes that the flat disc shape of the Trilobite is still the standard shape for robot vacuums. Its name and its scale-like vent slits were inspired by trilobites, sea animals that lived hundreds of millions of years ago [2]. So the "first" robot vacuum already used a sensing idea that many later machines still follow: send out a signal, read what comes back.

2. How the Roomba Moved Around With Only a Tiny Brain

According to the AI100 report, iRobot introduced the Roomba a year after the Trilobite, in 2002. It cost about a tenth as much. It ran a "behavior-based" controller, which means it reacted to what its sensors told it with simple built-in actions, and it used only 512 bytes of memory (RAM). That is about enough to hold a short sentence. The report says the most intelligent thing it did was to avoid falling down stairs [1].

The Smithsonian's National Museum of American History describes the Roomba as a machine that rolls on wheels and reacts to its surroundings using sensors and computer processing. When it bumps into an obstacle, or detects an infrared beam that marks a line it must not cross, it changes what it is doing [5].

One way to picture it is walking through a dark room with your hand out: you touch a wall, turn, and keep going. The Roomba did not draw a map first. A simple design like that may be one reason it could be sold at a much lower price. This is a simple picture to help you imagine it, not a claim about every detail of the code.

3. When Did a Roomba First Use a Camera to Build a Map?

In September 2015, iRobot announced the Roomba 980. The company's press release calls it the first Roomba to combine adaptive navigation with visual localization [3]. The trade publication eeNews Europe explains the idea: the 980 used the company's vSLAM technology, which works with a low-resolution video camera. The robot creates visual landmarks in its map so it does not lose track of where it is or where it has been [4].

SLAM stands for "simultaneous localization and mapping." In plain words, the robot draws a map and works out where it is on that map at the same time. The AI100 report notes that this technology had been around for about twenty years before robot vacuums used it, and says it lets the robots build a full 3D model of a house as they clean and cover floors more efficiently [1].

Japanese tech reporting on the Japan launch gave one number: the 980 could clean up to about 185 square meters (roughly 2,000 square feet), more than four times the maximum for the older 800 series, according to the Japanese distributor [6]. That figure comes from a launch announcement, so treat it as the maker's claim rather than a lab measurement.

4. Maps Came Before and After the Bump-and-Turn Roomba

Putting the pieces together gives a timeline that is not a simple staircase:

• 2001: the Trilobite senses the room with ultrasound [2].
• 2002: the Roomba uses simple reactions and very little memory, at a much lower price [1].
• 2015: the Roomba 980 adds a camera and landmark map [3][4].

So the idea of a robot knowing its surroundings came early, and the idea of a robot succeeding with a very small brain came right after. Comparing these two machines, the Trilobite used ultrasound, while the Roomba 980 used a camera. The AI100 report links the improvement to cheap processors and memory getting much better after the year 2000 [1].

The sources used here do not compare how well the Trilobite's ultrasound and the 980's camera actually cleaned, so this article cannot say which was more accurate. They also do not cover other companies' robots, which is a good question to explore on your own.

5. Try It: A Pencil-and-Dice Game That Compares Bumping and Mapping

You can feel the difference with paper. Draw a 5 by 5 grid and put an eraser on the top-left square. Roll a die: 1 means up, 2 right, 3 down, 4 left, and a 5 or 6 means roll again. If the move would go off the grid, stay put and roll again. Mark each square you enter, and count the real moves until all 25 squares are marked.

Now try a planned path: go left to right across the first row, step down, go right to left across the next row, and so on. That covers every square in exactly 24 moves. The dice version usually takes more, because it keeps revisiting squares. Real robots are far more complicated than this game, so your results only describe this game, not how well any actual robot cleans. Try the dice version a few times and notice how much the count changes.

Sources