Robot vacuum cleaners: how they work and navigate your home

An idea older than the first working prototype

The idea of a vacuum cleaner that works on its own didn’t start in a lab; it started in a novel. In 1956, science fiction writer Robert Heinlein described a device in The Door into Summer that roamed the house all day “looking for dirt.” A year later, in the real world, engineer Donald Moore patented several robotic home appliances, including an automatic sweeper: Whirlpool considered the idea but decided against building it after a 1959 demonstration.

It would take until 1985 for a product to actually reach the market: Tomy’s Dustbot, described as the first robot with built-in suction. Electrolux showed off a prototype of its Trilobite on TV as early as 1996, but the actual commercial launch didn’t come until 2001, making it the first true commercial robotic vacuum cleaner in history. The Trilobite, though, suffered from serious navigation problems and was pulled from the market, an unlucky pioneer that arrived before the technology could really make it work.

Meanwhile, at MIT, an engineer named Joe Jones had already been working on the idea since 1989, pitching early prototypes to two different companies without finding any commercial interest. Only in 1999, inside the fledgling iRobot, did the project take real shape: “Scamp,” the oldest documented prototype, was born, and S.C. Johnson funded its development before eventually stepping away.

The focus group that changed the product

The turning point came in the summer of 2001, and it didn’t come from a lab but from a group of ordinary consumers. The original design of what would become the Roomba was, in practice, a motorized carpet sweeper: it didn’t actually vacuum. When focus group participants found that out, their willingness to pay for it collapsed: people who’d first said $200 now offered $100. That same session also revealed who the product’s first real audience would be, not the tech enthusiasts the team had expected, but busy moms managing a household, a very different market segment from the one they’d designed for.

From there, the team went back to the drawing board to add real suction, and in September 2002 iRobot put the first production version of the Roomba on sale.

Practical example: the most concrete constraint the team had to solve was energy. The robot had a total power budget of 30 W, and could only afford to devote about 10% of it, roughly 3 W, to actual suction, versus about 1,200 W for a conventional vacuum cleaner of the time. The fix wasn’t a stronger pump, impossible on that little power, but a suction opening only a couple of millimeters wide, sealed by two rubber vanes that work like a squeegee on hard floors.

The sensors: how a robot “sees” without eyes

A robot vacuum doesn’t have vision the way we think of it, but it adds up data from several sensors to figure out where it is and what’s around it.

Sensor	What it does
Ultrasonic time-of-flight (ToF)	Measures distance to obstacles in millimeters, regardless of surface color; detects pets or toys to avoid collisions
Short-range ToF	Tells floor types apart based on the reflected signal, slowing the robot down when moving from carpet to hardwood
Cliff detection	Spots drop-offs, like the top of a staircase, to prevent falls
Inertial measurement unit (IMU)	Measures roll, pitch, and yaw of the robot’s movements, helping it tell whether it’s covering the same area twice
Pressure sensors	Estimate how full the dustbin is based on a drop in airflow
Thermistors	Monitor the temperature of the motherboard, motors, and brushes, pausing the robot if it detects overheating

In newer, higher-end models, manufacturers add dedicated sensors to recognize specific objects: iRobot’s PrecisionVision technology, for instance, is built to spot and avoid power cords, shoes, and other typical household obstacles. Recognizing and classifying an object on the fly like that, rather than just measuring its distance, is a classic machine learning task: the system is trained on huge numbers of real-world examples before it’s ever set loose in someone’s home.

Random bounce, LiDAR, vSLAM: three eras of navigation

This is where the difference between a budget robot and a high-end one becomes most visible, and it’s also the point where most misconceptions come from.

The earliest robots, including the first generations of the Roomba, navigate using what’s called random bounce: a set of fairly simple algorithms — spiral, follow the walls, change angle after hitting something — combined with wheel rotation and IMU data to blindly estimate where the robot is, a technique known as dead reckoning. It works, in the sense that the floor ends up mostly covered, but without a real map of the room.

The shift came in 2010, when Neato Robotics launched the XV-11, the first consumer model to use LiDAR navigation: a laser that measures distances and lets the robot build an actual map of the space, cleaning it with systematic, straight-line paths instead of random ones. In 2015 came a second route to the same goal, camera-based mapping, or vSLAM (visual simultaneous localization and mapping), adopted by both Dyson and iRobot in that year’s flagship models. Processing camera images in real time to figure out “where am I and what’s around me” is the same kind of problem that neural networks tackle in plenty of other computer vision applications.

Both technologies, LiDAR and vSLAM, share the same weak spot: they struggle in dimly lit spaces, and under a low table or sofa they lose effectiveness for lack of a clear line of sight or a good signal reflection.

Who makes robot vacuums today

For years, “robot vacuum” and “Roomba” were nearly synonymous: iRobot created the category in 2002 and, by May 2022, was reporting more than 40 million robots sold worldwide. The picture today looks very different.

In August 2022, Amazon announced it would acquire iRobot for $1.7 billion, but the deal was blocked by US and European regulators concerned that Amazon could use its e-commerce weight to limit competition. After the deal collapsed, in January 2024, iRobot laid off a third of its staff and suspended research and development; co-founder and longtime CEO Colin Angle left the company. On December 14, 2025, iRobot filed for bankruptcy, and its assets were taken over by Picea, the Chinese partner that had already been manufacturing its products.

Behind this story lies a bigger shift: China has been the world’s largest market for robot vacuums since 2020, with local companies that, according to IEEE Spectrum, invested two to three times more than iRobot did in research and development. It’s no coincidence that iRobot’s European market share has fallen to 12%, and is still declining. The robot vacuum remains a growing category, in short, but today several manufacturers and several navigation technologies compete for it, not a single pioneering brand.

Practical limits and maintenance

Even the most advanced model remains a device with physical limits that have nothing to do with how sophisticated its navigation is.

Thick rugs are still a real obstacle: robots struggle to cross them and often cut power to manage it. Areas under low furniture remain a blind spot even for LiDAR and vSLAM, which lose effectiveness there. If someone physically picks up and moves the robot while it’s working, it loses track of its position and has to head off in a random direction to reorient itself, not unlike having to start over after losing GPS signal in a car. And hair or elastic bands tangled in the brushes can overload and overheat the motors, which means periodic manual cleaning is unavoidable.

Even the most reliable sensors have their own odd practical limits. In 2016, a wave of reports, later picked up by the international press, described robots rolling over pet accidents left on the floor and spreading them across the room during a cleaning cycle, a concrete reminder that systematic floor coverage, however orderly it looks on paper, isn’t enough on its own to guarantee a clean result in every circumstance.