AivexaNewsSearch
AI news for builders and product teamsChecked every hour

I watched a humanoid robot clean my apartment. It felt like a taste of the future.

Collected Nov 6, 2026

Tau Robotics, a San Francisco startup, has been running a humanoid-robot cleaning service that a MIT Technology Review reporter tried twice, most recently in September, and the experience captures both the promise and the present limits of household robotics. The company's long-term aim is fully autonomous cleaning machines. For now, each robot is driven remotely by a human operator who watches the room through cameras mounted on the robot's head and wrists. Every wipe of a counter and fold of a shirt is captured as training data that Tau says will teach its models to work independently later.

That data is central to the business, which helps explain the pricing. When the reporter booked, Tau charged $30 for a one-hour cleaning with a single robot. It has since moved to $60 for a weekly cleaning with two robots, or $100 for a one-time visit. The reporter's verdict on value was mixed: the robots are thorough but slow enough that watching the whole session is less efficient than simply cleaning yourself.

The hardware is deliberately unusual. Each robot stands about four and a half feet tall and is fitted with a 3D-printed head carrying antennae for communication with Tau's headquarters, plus two 3D-printed claw hands. More humanlike hands apparently did not hold up to the wear of cleaning, while the claws handle most tasks well. The reporter's robot wore a Tau Robotics shirt and grippy socks intended to protect hardwood floors.

A lot of the work is what Tau calls kinematic guidance. For tasks like making a bed or folding laundry, the human pilot performs the motions, and sensors transmit them to the robot, which copies the movement. These guided sessions are especially valuable as training examples, and the reporter notes that even if Tau's models cannot yet fold laundry on their own, learning from real-world demonstrations is a plausible route there. Cofounder Cornelia Weinzierl said the robot can already vacuum and mop without human help. This is worth parsing carefully: limited autonomous routines and teleoperated general cleaning are very different capabilities, and the service as experienced runs firmly on the second.

The reporter described the robots as closer to radio-controlled cars than to Waymo's self-driving vehicles, given how much human support they need. The first visit exposed where that support breaks down. The robot climbed the building's eight front steps with a drunk-looking stagger, each step sometimes followed by a backward stumble that erased progress without a fall. Weinzierl attributed the gait problems to signal delay from remote operation, saying the company spends a surprising amount of effort on timing and connectivity. The carpeted interior staircase defeated the robot outright, and Weinzierl carried it up the last few steps.

Inside, the robot stalled repeatedly. Weinzierl initially blamed heat from unseasonably warm weather, then learned from a colleague that high temperatures had shut down the controller's equipment rather than the robot itself. With no remote operator able to see, the robot could not continue, and the session ended before the booked hour. Weinzierl emailed an apology, comped the visit and offered another free cleaning. In September, a Tau representative stayed with the robot for the whole visit, though she says that is no longer necessary.

A second cleaning roughly a month later went far better, with the robot managing the stairs with little help and still staggering. It cleaned the kitchen credibly, skipping the stove, and handled the toilet, where the most striking moment came: the head fell off its cleaning wand three times and three times the robot reattached it by probing with the handle until it caught at the right angle. Rough edges remained. Taps were left dripping, a closed bottle of olive oil was knocked over, and a toppled air purifier was not righted.

Weinzierl's case for the service rests on labor economics: cleaning has high turnover and high injury risk, and she argues workers are hard to find and few want the job. She said Tau is also beginning to consider other physically demanding or dangerous applications, including mold remediation and home health care. The reporter raises the counterweight: if autonomous cleaning becomes cheap, cleaning jobs could disappear, and the United States has little policy in place for workers displaced by automation. The scale is still small, with only a couple hundred cleanings completed so far, and the volume of data needed to master the varied tasks of apartment cleaning is presumably vast.

Why it matters: Tau's service is a working preview of how embodied AI products may reach consumers before they are autonomous, selling subsidized remote labor while harvesting demonstration data. For robotics and product teams, the bottlenecks on display were not grasping or perception alone but teleoperation latency, thermal resilience of operator equipment and stair handling, which suggests infrastructure around the robot can decide whether a job succeeds. Teams planning similar services should expect beta-grade reliability, incident handling and pricing that reflects data value rather than labor cost. Whether that data converts into genuinely independent robots remains the open question, and by inference, the first useful signal will be how quickly Tau's autonomous task list grows beyond vacuuming and mopping.

Read at MIT Technology Review · AI

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

I’m not the sort of person who typically pays for a house cleaner. I’m too frugal, and I kind of enjoy putting on a podcast and going to town on my tub with a Magic Eraser. But when I saw an X post about a new humanoid-robot cleaning service in San Francisco, I immediately signed…