From teleoperation to autonomy: how we hand a task over
Every 1to1 robot ships with an operator in the loop. Here is the exact process we use to move a task from human control to full autonomy without losing the audit trail.
Why both modes, always
Autonomy is not a switch you flip on a delivery date. It is earned one task at a time, in one environment at a time. That is why every platform we build supports direct teleoperation and full autonomy from the day it arrives.
An operator takes control through a low-latency link with force feedback, and the robot enforces its own motion limits in firmware regardless of what the console asks for. Nothing an operator does can push a unit outside its safety envelope.
The handover process
A task starts fully teleoperated while we record the sensor state and the operator's decisions. Once the Physical AI runtime can reproduce those decisions within tolerance, the task runs autonomously with the operator supervising. Only after a clean run of statistically meaningful length does supervision drop to exception handling.
Because the runtime logs every autonomous decision against the sensor state that produced it, a safety engineer can replay any session frame by frame. That replay is what makes autonomy defensible in a hospital or a licensed terminal.
What this means economically
You do not wait for perfect autonomy to see a return. One operator supervising a dozen machines already changes the arithmetic of a night shift, and the ratio improves every month the fleet runs.