Tau launches $30 humanoid cleaning in San Francisco with humans at the controls

Why it matters

Tau is using a paid home service to acquire training data before its humanoids are autonomous. The model could shorten the path from lab demonstrations to useful robots, but it depends on customers accepting continuous recording and human teleoperation inside their homes.

Alexander Koch (@alexkoch_ai), co-founder and CEO of Tau Robotics, launched an invite-only humanoid cleaning service in San Francisco on July 28th, charging $30 for a one-hour visit while a human operator remotely controls the robot.

https://x.com/alexkoch_ai/status/2082135074615763251?s=46

Koch announced the service in a thread on X, posting footage he said was shown at normal speed. Tau is opening access gradually through a neighborhood-based waitlist, with staff transporting each humanoid to the customer's home.

Koch founded Tau in 2024 with Cornelia Weinzierl. His background is in robot learning rather than residential services: Koch studied computer science at the Technical University of Munich and co-authored a 2023 ICRA paper on teaching robotic manipulation skills through language-conditioned reinforcement learning. On his personal site, Koch describes his goal as building "a general AI for robots" and says he has spent nearly a decade working on scalable robot-learning algorithms.

Tau's cleaning offer puts that research inside paying customers' homes before the underlying system can complete the work independently. The distinction is explicit in Tau's service terms: the robot is controlled in real time by a trained operator working from Tau's San Francisco facility. A customer or another authorized adult must remain home throughout the visit.

The $30 cleaning is also a data transaction

Tau's price is tied directly to the data it collects. Its service privacy policy says every cleaning is continuously recorded through cameras mounted on the robot's head and wrists. Tau also captures depth information, robot telemetry and the commands sent by the operator.

That material is used to train the vision-language-action models Tau hopes will eventually perform the work without a human controller. The robot's microphones are disabled, according to the policy, although the remote operator can speak through its speaker. Tau says faces are not blurred before footage enters its training pipeline.

Customers can request deletion of raw footage and remove it from future training sets. Tau also warns that data already used to train a model cannot be extracted from the model's parameters. Unless a customer requests deletion, visit footage and telemetry can be kept indefinitely.

The terms make clear how much operating complexity remains behind the apparent simplicity of a $30 robot cleaner. Tau staff deliver the machine, an employee controls it from another location, and the customer supplies a working internet connection. Pets must be secured, fragile items put away and potential hazards disclosed. Bedrooms and bathrooms are excluded unless the customer opts in.

Tau advertises house, deep and office cleaning, including wiping surfaces, vacuuming, taking out trash and clearing clutter. It has not published the number of humanoids available, the number of homes it expects to serve or how many robots one operator can supervise. Those figures will determine whether the service can expand beyond tightly managed training deployments.

Real homes become the training environment

Tau's launch follows another San Francisco attempt to sell humanoid labor as a consumer service. Gatsby said in a May announcement that it had completed a residential humanoid cleaning on May 14th and was charging a flat $150 per visit. Gatsby positions itself as a booking and distribution layer that can use robots from multiple manufacturers. Tau is building the robot intelligence and using the cleaning visit to gather its own training data.

That difference explains Tau's unusually low introductory price. Each booking gives Tau a structured demonstration of a human controlling a humanoid through kitchens, offices and other spaces that are difficult to reproduce in a laboratory. The customer receives an hour of cleaning; Tau receives video, spatial data, operator actions and records of failures in a real home.

Tau's own website has previously emphasized autonomous data collection and minimizing dependence on teleoperation. The cleaning service starts from the opposite end of that path, using human control to produce the examples needed to train greater autonomy. Tau is hiring on-site engineers in San Francisco to work on humanoid hardware and teleoperation interfaces, alongside roles focused on world models and reinforcement learning.

Funding database Dealroom lists a $3 million 2024 seed round from Pebblebed and HF0. Pebblebed's portfolio identifies Koch and Weinzierl as Tau's founders and describes Tau as building toward general-purpose humanoid robots.

The cleaning launch gives Tau a way to test that thesis with customers while the robots still depend on people. Its near-term product is a remotely operated cleaning visit. The longer bet is that enough recorded visits can teach the humanoid to work alone.