Tau Robotics cleans San Francisco homes with humanoid robots piloted remotely

A San Francisco startup, Tau Robotics, is putting humanoid robots to work cleaning real apartments in the city — one of the first paid, consumer-facing deployments of general-purpose humanoids. Grace Huckins, a writer at MIT Technology Review, booked the service twice and described both the promise and the rough edges of the experience.

How the service works

The company's stated goal is fully autonomous domestic robots, but the current machines are not autonomous. Human operators pilot them remotely, working from camera feeds captured by sensors on the robots' heads and wrists. Pricing reflects an early experiment: when Huckins first booked, a one-hour cleaning with one robot cost $30; the service now charges $60 for a weekly cleaning with two robots, or $100 for a one-time visit.

According to MIT Technology Review, the low price makes sense because the real product is data. Every wiped counter and folded garment becomes training material for the AI models Tau hopes will eventually run these machines without human help. The hardware is deliberately practical: each robot stands about four and a half feet tall, with a 3D-printed head carrying antennae for communicating with Tau's headquarters and a pair of 3D-printed claw hands. More humanlike hands apparently were not durable enough, while claws handle most household chores surprisingly well. The robots wear company T-shirts and grippy socks to protect hardwood floors.

Humans in the loop

Difficult tasks such as making a bed or folding laundry currently rely on what the industry calls kinematic guidance: the human operator performs the motions, sensors transmit them, and the robot imitates the movement. These guided demonstrations are reportedly the most valuable data Tau collects. Cofounder Cornelia Weinzierl told MIT Technology Review that the robots can already vacuum and mop without any human assistance. Huckins's verdict is that, for now, the machines sit closer to radio-controlled cars than to autonomous robotaxis.

Two visits, mixed results

The first cleaning was rocky. The robot arrived in a minivan and struggled up the eight exterior steps to the building, climbing haltingly and frequently lurching backward, though it never fell. Weinzierl attributed the awkward gait to the signal delay between the robot and its remote operator, noting that her team spends significant time tuning timing and connectivity.

A carpeted staircase inside defeated the robot entirely, and Weinzierl had to carry it up the final steps. The visit then ended early after the robot repeatedly stalled. The initial explanation was overheating in unusually hot weather — but the failure turned out to be in the human operator's viewing equipment, not the robot itself. Without the pilot watching, the robot could not work. Weinzierl apologized, comped the visit and offered a free return session, all before learning her customer was a journalist.

The second visit, roughly a month later, went far better. The robot managed the stairs with minimal help, cleaned the kitchen (though it avoided the stove) and even scrubbed the toilet — reattaching the head of its cleaning wand three times by prodding with the handle until it seated correctly. It still left taps dripping, knocked over a closed bottle of olive oil and toppled an air purifier it could not right.

Why it matters

Tau has completed only a couple hundred cleanings so far, and the volume of data needed to master the varied, messy tasks of apartment cleaning is likely enormous. Even so, this is a rare commercial test of general-purpose humanoids in ordinary homes rather than factories or warehouses, and the teleoperation-plus-data-collection playbook mirrors the trajectory that took robotaxis from novelty to routine service.

The economic stakes are real. Huckins notes that if robots learn to clean autonomously and cheaply, people could lose their livelihoods while US policy for AI-driven unemployment lags far behind. Weinzierl's counterargument is that cleaning work already suffers from high turnover and high injury risk, and that qualified labor is scarce. Tau is also weighing other physically demanding or hazardous jobs for future deployments, including mold remediation and home health care. For now, the service is a beta: slow, clumsy in places, and charming enough that early customers seem willing to forgive its mistakes.