Matic’s Robot Vacuum Learned to Listen. That May Matter More Than Another Humanoid Demo
Matic has evolved from a quirky premium robot vacuum into one of the most intriguing bets in home robotics, with new voice and gesture controls hinting at a future where household machines understand us as naturally as people do.
For most of the past decade, the robotics industry has been obsessed with a very particular vision of the future: a humanoid machine with two arms, two legs and enough artificial intelligence to walk into a home and behave like a mechanical housekeeper. It is a compelling image, and one that plays beautifully in demonstration videos, but it also asks engineers to solve almost every difficult problem in robotics at once.
Matic has taken a much less theatrical route. Founded in 2017 by former Google Nest engineers Mehul Nariyawala and Navneet Dalal, the company began with a far more mundane ambition: build a machine that could clean floors reliably, navigate the chaos of a real house and require as little human supervision as possible. Nearly a decade later, that narrow focus is starting to look less like a limitation and more like a strategy.
This week, Matic released one of its most consequential software upgrades yet. A new feature called Matic Cues allows owners to speak directly to the robot, ask it to follow them, direct it toward another room or simply point at a mess and tell it to clean there. The robot uses its cameras and onboard perception system to work out where the user is standing, interpret the direction of the gesture and identify the area being indicated.
It is an interface so natural that it almost feels anticlimactic. If you want another person to clean something, you do not open an app, locate a room on a digital floor plan and drag a rectangle over the offending crumbs. You point and say, “Can you get that?”
The significance of Matic Cues is that the robot is beginning to adapt itself to human behaviour, rather than demanding that humans adapt themselves to a machine. That may turn out to be a much bigger step than it initially appears.
Introducing Cues: Voice & Gesture control for Matic We raised $115M and spent 9 years to make the world’s first intuitive home robot Say you spilled coffee: Point to the spill and say, "Hey Matic, clean this" and it will hear, see, locate in 3D, and go clean on its own. Matic Show more
Building a Robot for the Messiness of Real Homes
Matic’s path to this point has been unusually long for a consumer electronics company, in large part because homes are hostile environments for robots in ways that are easy to underestimate.
A demonstration room can be controlled. A real house cannot. Furniture moves, lighting changes, doors are left half open, cables fall onto the floor, rugs curl at the edges, shoes migrate into hallways, mirrors confuse vision systems, pets create new obstacles, and children scatter objects into spaces that were empty five minutes earlier. Almost no two homes are arranged in quite the same way.
Matic says its founders originally expected to get a product to market in roughly three years, but the difficulty of building a robot capable of dealing with those edge cases stretched the development cycle far beyond that estimate. In the company’s account of its development history, it argues that a robotic policy capable of achieving 95% accuracy in controlled laboratory conditions can drop to around 60% when exposed to real homes.
That experience helped shape Matic’s distinctive hardware. Rather than using the low, circular form factor popularised by Roomba, the machine is taller and uses a five-camera vision system and an onboard Nvidia processor to build and continually update a three-dimensional understanding of the house as furniture and objects move.
The result is not necessarily the most powerful vacuum cleaner in raw suction terms, nor can its taller body reach beneath every piece of furniture. What Matic has focused on instead is autonomy: the ability to understand a home well enough to move around it without repeatedly trapping itself, consuming cables or requiring an owner to rescue it from under the dining table.
That distinction matters because consumer robotics has long suffered from a reliability problem disguised as a feature problem. Manufacturers have spent years adding stronger motors, larger batteries, better docks and increasingly elaborate mopping systems, yet the experience still falls apart if the robot gets stuck halfway through a job and sends its owner a notification asking for help.
After six months of testing, The Verge’s Jennifer Pattison Tuohy gave Matic a 9/10, describing it as the smartest robot vacuum she had tested and highlighting its ability to operate successfully inside a cluttered, multi-storey home. That kind of performance is less photogenic than a humanoid folding a shirt on a stage, but it may be a much better measure of progress in domestic robotics.
WIRED magazine just gave a 10/10 rating to a piece of hardware for the first time in a decade. It is a robot vacuum. Meet Matic. You do not use an app to control it. You literally treat it like a pet. You can point at a coffee spill and say "Hey Matic, clean this." You can even Show more
Introducing Cues: Voice & Gesture control for Matic We raised $115M and spent 9 years to make the world’s first intuitive home robot Say you spilled coffee: Point to the spill and say, "Hey Matic, clean this" and it will hear, see, locate in 3D, and go clean on its own. Matic
The Numbers Are Starting to Matter
By January 2026, Matic said it had shipped more than 6,000 robots, which had collectively cleaned more than 110 million square feet and travelled over 80,000 miles inside customers’ homes. At the same time, the company announced another $60 million in funding, intended to expand manufacturing and continue development.
Since then, the installed base has continued to grow. Matic now says more than 10,000 robots have shipped, while its live counter has passed 305 million square feet cleaned.
Robotics improves through exposure to reality. Every deployed machine encounters some new combination of objects, surfaces, lighting conditions, room layouts and human behaviour, and that accumulated experience helps expose the edge cases that engineers cannot reliably recreate in a lab.
A robot might encounter a black rug that looks like a hole in the floor, a mirror that appears to extend the room, a dangling laptop cable that shifts position from one cleaning run to the next or a pet bowl that has inexplicably moved from one end of the kitchen to the other. None of those examples is individually remarkable, but robotics is largely a discipline of surviving millions of unremarkable exceptions.
That means Matic’s growing footprint inside real homes is becoming an increasingly important asset. The company is not simply selling vacuum cleaners; it is accumulating operational experience in one of the most difficult environments robotics companies can choose to enter.
A Robot That Understands “Clean This”
The most visible feature in this week’s update is gesture control, but Matic Cues is really a broader attempt to remove the smartphone from the interaction loop.
According to Matic’s August 13 release notes, owners can now tell the robot to clean a room, vacuum a particular carpeted area, mop a hard floor, return to its dock, go to the sink or follow them through the house. More complicated requests can be combined into a single spoken instruction, allowing the robot to perform different jobs in different rooms without forcing the user to configure each task separately.
The pointing system is particularly interesting because it combines several capabilities that previously existed more independently: speech recognition, visual perception, body tracking, spatial mapping and autonomous navigation.
The Verge tested Matic Cues shortly before launch and found that it generally worked well, although the precision of the selected cleaning area could depend on the angle and height from which the user pointed. Even with those limitations, the interaction suggests a different model for controlling machines around the home.
Matic’s technical approach also reflects the current realities of AI hardware. The wake phrase, sound localisation and interpretation of the pointing gesture are handled locally on the robot. Once activated, the spoken command itself is currently sent anonymously to Google’s Gemini API for language interpretation, giving Matic support for more than 75 languages and considerably more flexible natural-language commands.
The company says it ultimately wants more of that language processing to happen locally as well, potentially by moving the most commonly used commands onto smaller models capable of running on the robot itself.
For now, the hybrid architecture is a reasonable compromise. The robot handles much of the privacy-sensitive job of seeing and understanding the physical home locally, while relying on cloud AI for the more computationally demanding problem of interpreting unconstrained human language.
Software Is Becoming the Product
Perhaps the most important part of Matic’s evolution is that many of these capabilities are arriving without new hardware.
Over the past year, Matic’s software releases have steadily improved the robot’s ability to deal with cables, mirrors, stairs, room boundaries, furniture, clothing and other objects left on the floor. Pet-waste detection has also entered beta, one of those unglamorous robotics capabilities that becomes extremely important the first time a vacuum cleaner gets it wrong.
The software release immediately preceding Cues made room-to-room path planning up to two times faster, improved Matic’s ability to determine when it was genuinely stuck while mopping, refined room boundaries and added live cleaning information to the iPhone Lock Screen and Dynamic Island.
Then Cues added voice, gestures and more natural interaction to robots that were already sitting in customers’ homes. Matic says the feature was enabled across all 10,000-plus Matic robots shipped to date.
That is a notable change in the economics of consumer robotics. Traditionally, household appliances improve primarily when the owner replaces them. A more powerful dishwasher or smarter vacuum arrives as a new model, and the previous generation remains essentially frozen at the capabilities it had on the day it left the factory.
Robotics combined with modern AI creates the possibility of something different: a physical machine whose behaviour can continue evolving long after the hardware has been purchased. The chassis may remain the same while the robot gets better at seeing, navigating, understanding instructions and responding to its environment.
That makes the comparison with smartphones unavoidable. The first meaningful iPhone was not important because of what it could do on launch day alone, but because its combination of hardware, sensors, software and connectivity created a platform on which capabilities could accumulate.
A Different Route to Rosie the Robot
This is where Matic becomes more interesting than the premium robot-vacuum category it currently occupies.
Much of Silicon Valley is betting heavily on humanoids, based on the perfectly sensible argument that human environments were designed around human bodies. A robot with hands, arms and legs can theoretically use stairs, open doors, operate appliances and manipulate the countless objects found in homes and workplaces.
The difficulty is that this approach requires extraordinary competence across almost every major field of robotics simultaneously.
A humanoid must balance, move safely around people, recognise objects, manipulate them with sufficient dexterity, understand language, plan complex sequences of actions and recover gracefully when something goes wrong. It must also do all of this at a price that consumers or businesses can justify, with enough reliability that people are comfortable sharing physical spaces with it.
Matic is attacking the same long-term problem from the opposite end.
Instead of trying to build the general-purpose domestic robot first, it began with one constrained but genuinely useful task. In solving that task, it has had to develop many of the underlying capabilities that a more general home robot will eventually require: spatial awareness, persistent mapping, object recognition, autonomous navigation, language understanding and, increasingly, an ability to interpret human intent.
The broader ambition is not something outsiders have had to infer. In a detailed discussion of Matic’s robotics roadmap, co-founder Mehul Nariyawala said the company had never intended merely to reinvent the robot vacuum. Floor cleaning was the first step toward the much larger goal of a “Rosie the Robot” for the home, with perception coming first and manipulation following later.
That sequence is important. Nariyawala compares Matic’s approach with the way a child develops: first learning to perceive and navigate the environment, then gradually developing the ability to manipulate objects, and only later taking on more complex physical tasks.
Seen from that perspective, the vacuum cleaner looks less like the final product and more like a rolling perception platform that happens to earn its keep by cleaning the floor.
The obvious question is what happens when future machines inherit the intelligence Matic has already spent years building.
A cleaning robot does not need arms. The next generation of domestic robots very well might.
Home Robotics May Arrive Gradually
There are good reasons not to get carried away.
Matic remains an expensive product, currently priced at $1,245, and The Verge reports that the price will rise to $1,495 on September 9. Its physical design imposes limitations that software cannot remove, and its use of cloud processing for natural-language commands introduces a dependency into a system that has otherwise made local computation and privacy important selling points.
More fundamentally, there is an enormous gap between understanding that a sock is lying on the floor and possessing the dexterity to pick it up, identify where it belongs, carry it there and put it away.
Domestic manipulation is still one of the hardest unsolved problems in robotics.
But Matic is interesting precisely because it offers a glimpse of how the industry might get there without waiting for one miraculous breakthrough. The machine first learns to map the home, then to recognise rooms and surfaces, then to avoid objects, then to distinguish between obstacles, then to respond to human language and gestures. Each capability makes the next one more useful because they are being layered onto the same persistent understanding of the physical environment.
Over time, the distinction between “robot vacuum” and “home robot” begins to blur.
A machine that knows the layout of your house, recognises objects, understands spoken language, knows what you are pointing at and can move autonomously from room to room already possesses a surprisingly large part of the intelligence required for domestic robotics. What it lacks is not necessarily understanding so much as physical capability.
The future of robotics in the home may not arrive as a $30,000 humanoid walking through the front door and instantly performing every household chore. It may emerge much more gradually, through useful machines that earn their place by doing one job well and then become more capable as their perception, intelligence and hardware improve.
For years, robot vacuums have been treated as the slightly goofy underachievers of the robotics world: useful appliances, perhaps, but hardly evidence that science fiction was arriving.
The first truly important home robot may not look anything like a person. It may simply be the machine that already knows where your kitchen is, understands what you mean when you point at the floor and, crucially, does what you asked without needing to be rescued halfway through.
Maya Singh translates capability into ROI: products, adoption curves and the workflows that actually change how teams work.
