Why an impressive demo isn't the same as reliable work
A demo video is, by nature, the best take: it can be scripted, run in a controlled environment, filmed multiple times until one attempt goes smoothly, and edited. None of that is necessarily dishonest — demos are a legitimate way to show what a robot's underlying capability looks like — but it means a demo tells you almost nothing about failure rate, about how the robot handles the countless small variations of a real, messy environment, or about how often a human has to step in to reset, assist, or take over. A robot that can fold one shirt perfectly on camera is a very different claim from a robot that can fold shirts reliably, unsupervised, for eight hours a day, every day, in a facility that depends on it.
Unscripted, sustained deployment is the harder and more meaningful bar, because real facilities are messy: lighting changes, objects are placed slightly differently than expected, other workers and equipment move unpredictably around the robot, and the robot has to keep working reliably hour after hour without novel failures piling up. That gap between 'did it once, on camera, under ideal conditions' and 'does it every day, without much human intervention, in a live facility' is the single most important thing to watch for when evaluating any humanoid robotics claim.