When you watch a humanoid robot perform a task on video, the most important question — more important than what the robot is doing, more important than how fast — is: is a human operating it right now?
The answer is almost always yes, partially. The honest answer is rarely given.
There is a spectrum here, not a binary. At one end is full autonomy: the robot perceives the world, plans an action, and executes the action without human intervention. At the other end is pure teleoperation: a human wearing motion-capture or holding controllers maps their movements onto the robot, the way a puppeteer operates a puppet. In between is a wide middle ground — hybrid systems where autonomy handles the easy parts and teleoperation kicks in for the hard ones, or where a human supervises and corrects in real time.
Marketing material almost never sits anywhere on this spectrum. It just shows the robot doing the thing. The viewer fills in the autonomy gap from context: a robot folding laundry must be autonomous (it's not). A robot walking from point A to point B is probably autonomous (it usually is). A robot performing a delicate manipulation task is somewhere unclear (it almost always involves teleoperation, even when it looks autonomous).
Why does the distinction matter? Because a robot that can do a thing with a human pilot is a piece of impressive hardware. A robot that can do that same thing without a pilot is a piece of impressive hardware and a piece of capability that scales. The former needs a human per robot. The latter needs no humans, ever, after deployment. Those are different products. They have different costs, different applications, and different paths to profitability.
When you read about a humanoid that "folded a shirt" or "made a sandwich" or "interacted with attendees at our event," the question to ask is which part was autonomous and which part wasn't. Some companies — 1X is the most prominent example — have started saying so on the record. Most have not.
The Reality Score on a robot profile reflects this asymmetry. A robot whose demonstrations are accompanied by a clear autonomy disclosure scores higher on "transparency" than one whose demonstrations are not. A robot whose claimed capabilities have been demonstrated without teleoperation in front of independent observers scores higher on "demonstrated capability" than one whose capabilities exist only in edited promotional footage.
The teleop-vs-autonomy split is also why mass production timing is so hard to predict. Manufacturing a robot at scale is a problem the field knows how to attack. Making a robot autonomous enough to be useful without a human pilot is the problem that is genuinely unsolved. Companies that scale up production before solving autonomy end up shipping expensive pilotable hardware. Companies that solve autonomy first end up shipping fewer units but useful ones. Both paths are visible in the current roster.
There is one more thing worth knowing: even fully autonomous robots almost always require human review. Autonomy in the field is not the absence of humans; it is the absence of a human on the controls. There may still be a person watching the camera feed and pressing a button when the robot gets confused. This isn't dishonest — it's how every autonomous system, from cars to factory robots, currently works. The question is how often the button gets pressed, and how invisible the button is in the marketing material.
When in doubt, assume teleop. When the company is specific about which task is autonomous and which is hybrid, give them credit. When the company is not specific, assume the demo is mostly teleop, and reflect that in your reading of the Reality Score.