What humanoid robots are actually for syllabus
Count the cuts
A robot that can do a task reliably can be filmed doing it in one take. This is not a gotcha; it is simply what reliability means, and it is why the length of the longest unbroken shot is the most informative number in any demo.
What editing conceals, and what it doesn't
Cuts are not proof of dishonesty. Films are edited; so are product videos, for length and for boredom.
What a cut does is remove your ability to distinguish two very different worlds: one where the robot did the task, and one where it did the task on the eleventh attempt. Both produce identical footage once edited. So the question is never "did they cut" but "what is the longest continuous segment showing task-relevant work, and what happens at its boundaries".
The same logic applies to speed. Footage played at 2× with no on-screen indication is a claim about capability that the video does not support.
Analyse one demo, frame by frame
Pick any humanoid robot demonstration from the last year. Watch it once normally, then again with the scrubber, and record:
- Total runtime, and total runtime showing the robot doing task work.
- Number of cuts during task work.
- The longest unbroken take, in seconds.
- Whether playback speed is stated anywhere on screen.
- Whether teleoperation is disclosed, and where — in the video, the caption, or nowhere.
Do not pick the worst example you can find. Pick one you were impressed by; that is the one worth understanding.
The longest unbroken take
In seconds. This is the strongest claim the video actually supports: the robot did this much, continuously, at least once, on the day of filming.
Forty seconds, unbroken, no speed-up
A demo contains a genuine 40-second unbroken take of the robot folding a towel at normal speed. What does that establish?