A robot can move by itself.
That does not mean nobody helped.

Autonomy is not a universal badge attached to a robot. It describes who performs which decisions and actions for a defined task, under stated conditions. The same robot may be autonomous for route navigation, teleoperated for manipulation and manually reset after an exception.

A source that only says autonomous does not tell us the task boundary, supervision model or amount of human support. Those details must remain separate instead of being compressed into one label.

A nearby worker does not automatically make every cycle assisted. Conversely, a remote operator can materially control the work without standing beside the robot. Good reporting defines which human actions count as normal workflow, supervision, assistance, intervention or recovery.

Training must also be separated from evaluated operation. Teleoperating a robot to demonstrate a task may be part of data collection; it is not automatically an intervention during a later autonomous run.

A useful autonomy disclosure needs counts and time. State the operating mode for each task phase, the number and duration of interventions, who supervised how many robots, and whether assisted output is included in the headline result.

The denominator must travel with the rate. Ten interventions across ten attempts describe a very different system from ten interventions across ten thousand attempts, even though the intervention count is identical.

Imagine 100 defined task attempts: 80 finish without help, 12 finish after a remote correction, 5 finish under direct teleoperation and 3 do not finish. This is a hypothetical example, not a result for a robot in the database.

The workflow produced 97 completed tasks, but it did not achieve 97% autonomous completion. A transparent report preserves all four outcomes and states whether the 12 corrections count as interventions, how long they took and what triggered them.

HumanoidUptime records the operating mode a source actually supports. A product claim, a teleoperated demonstration, supervised task execution and measured autonomous-time share remain different evidence. Words such as deployed, working or autonomous do not fill in a missing mode breakdown.

When a source discloses mixed autonomous and teleoperated execution but gives no shares, the record says exactly that. When intervention logs are not public, the intervention rate stays undisclosed rather than being treated as zero.

For Apollo at Mercedes-Benz, operator-confirmed teleoperation describes how employees transferred task knowledge during training. It does not establish how often people intervened during autonomous execution.