Shared Autonomy Over Teleoperation

Pure teleoperation breaks the humanoid business model because labor cost rises almost one for one with each robot instead of getting spread across the fleet. The only scalable setup is shared autonomy, where the robot does most of the task itself, a human steps in only on edge cases, and every intervention becomes training data that makes the next robot and the next shift more autonomous.

• In practice, a teleop only fleet means staffing like a call center attached to every robot. That destroys the core pitch of humanoids, which is replacing repetitive labor with a machine that can work across multiple shifts without adding matching headcount.
  1 sacra 2 sacra
• The winning pattern across autonomy markets is not continuous remote driving, but remote assistance. Waymo uses human agents to give contextual help while the vehicle stays in control, and warehouse and delivery robotics follow the same model because one human can cover many machines only when autonomy handles the base case.
  1 sacra 4 waymo 6 agilityrobotics
• This is why humanoid companies talk so much about data flywheels. Figure is pushing fully autonomous full body control, Agility frames Digit around reducing constant intervention, and NVIDIA research is explicitly aimed at replacing large volumes of teleop data with simulation and world models.
  2 sacra 3 sacra 5 figure 7 nvidia

The market is heading toward fleets where teleop becomes a rare exception layer, not the product. Companies that use humans mainly to catch failures and feed retraining loops will widen their cost advantage fastest, because each software improvement lifts gross margin across every deployed robot at once.