Tau Robotics deploys 4.5-foot humanoid units for $60 weekly SF apartment cleaning under remote kinematic control

Tau Robotics dispatched two 4.5-foot units equipped with 3D-printed heads, antennae, and claw hands to a San Francisco residence. Pilots at headquarters used head- and wrist-mounted cameras for real-time oversight while kinematic guidance transmitted human motions for bed-making and folding. The robots collected video and joint data from every completed task. Stairs required manual assistance, exposing current mobility limits.

Performance data shows vacuuming and mopping executed without intervention, while laundry and bed tasks still demanded full kinematic teleoperation. Each successful cycle produces imitation-learning pairs that Tau feeds into its training pipeline. At $60 weekly for two units the service undercuts conventional cleaning yet remains slower than human labor, confirming the data-gathering phase still dominates economics.

The deployment mirrors Waymo’s early teleoperation-to-autonomy trajectory but targets indoor manipulation rather than navigation. Imitation learning from real household trajectories directly addresses the sim-to-real gap documented in recent robotics benchmarks. Success hinges on scaling these datasets before competitors with larger fleets capture equivalent interaction records.

Operational next steps require stair-climbing reliability and end-to-end autonomy on at least three additional tasks within the next billing cycle to justify price increases and reduce pilot hours.