The Reality Behind Viral Humanoid Robots Is Far Less Impressive

Viral videos make humanoid robots look capable of jogging, making coffee, or tidying a living room, but robotics experts say the reality is much more limited. At a robotics conference in Boston, specialists said most of today’s humanoid machines are either remotely operated or restricted to narrow, preprogrammed tasks. Even Neo, promoted as the first consumer robot meant to transform home life, was shown being guided by a person with a controller off to the side.

The gap between the claims of figures such as Elon Musk, who presents Optimus as a breakthrough, and the technical state of the field remains wide. For robots to work independently in the real world, they need huge amounts of training data, and there is not enough of it. To fill that gap, robotics companies are now placing cameras everywhere, from people cooking at home to workers in textile workshops in India, in order to record human movement and teach machines how to behave.

Researchers say progress is real and artificial intelligence is speeding it up. New vision-language-action, or VLA, models let a robot connect what it sees through its camera with written instructions in real time. Robot hands are also improving, with sensors that can detect contact with human skin.

But safety remains the biggest obstacle. Unlike chatbots such as ChatGPT, a robot mistake in the physical world can cause serious injuries. Experts describe these systems as unpredictable and as a black box, saying even their creators do not always understand why they act the way they do. Industry figures warn they are still far from the safety level needed, especially around people, and that a truly versatile android is still years away. Companies including Hyundai and BMW are experimenting, but until a robot can handle changing tasks without a hidden operator or a preset route, what appears on screen is often just a sophisticated illusion.