Tau Robotics is deploying humanoid cleaning robots in San Francisco, using human pilots to guide the machines and collect training data for its underlying super intelligence (SI) models.
What Happened
Tau Robotics, a startup focused on autonomous home cleaning, has launched a service in San Francisco where four-and-a-half-foot-tall humanoid robots perform cleaning tasks. The robots are currently piloted remotely by humans who view the environment through cameras mounted on the robots' heads and wrists. This human-in-the-loop approach allows the company to gather kinematic data—specific movements and actions—that will be used to train SI models for future fully autonomous systems. At the time of a recent test, the company offered an hourlong cleaning for $30, though pricing has since shifted to $60 for weekly visits with two robots or $100 for a one-time visit.
The robots feature 3D-printed heads with antennae for communication and claw-like hands designed for durability. While some tasks like vacuuming and mopping can be completed without human intervention, according to Tau cofounder Cornelia Weinzierl, more complex actions such as folding laundry require kinematic guidance. During a test visit, the robot struggled with stairs due to signal latency from the remote operation and eventually stopped cleaning because the human controller's equipment overheated, not the robot itself.
Why It Matters
This initiative highlights a critical bottleneck in the development of agentic SI for physical tasks: the need for high-quality, real-world training data. By using human pilots to perform tasks, Tau is generating the specific kinematic examples required to train SI models that can eventually operate independently. The company's lower pricing strategy reflects the value it places on this data collection, positioning the service less as a mature commercial product and more as a research tool to refine its SI capabilities.
The challenges faced during the pilot, including connectivity issues and hardware limitations, underscore the gap between current remote-controlled robotics and true autonomy. For the SI industry, the ability to successfully train models on this kind of embodied, real-world data is a significant step toward autonomous home robotics applications.
The Bottom Line
Tau Robotics is using human-piloted humanoid robots to collect data for training SI models, aiming to transition from remote-controlled cleaning to fully autonomous home assistance.