The appeal of humanoid robots is simple: our homes, factories and warehouses are designed for the human body, so a robot shaped like a person could, in theory, work anywhere we do. In practice, almost every part of the human body is hard to copy.

Balance and walking

Walking on two legs is a constant controlled fall. The robot must sense its position hundreds of times per second and adjust its joints to stay upright, even on uneven floors or when bumped.

Hands

Human hands have many degrees of freedom and dense touch sensing. Robot hands need small, strong actuators and tactile sensors to pick up both a heavy box and a fragile glass without dropping or crushing them.

Power

Motors, computers and sensors drain batteries quickly. Running a full work shift, and recharging without human help, is still a major challenge.

Learning skills

Many robots learn in simulation, where they can practice millions of times safely. Transferring those skills to the real world, known as the sim-to-real gap, is difficult because real friction, lighting and objects never match the simulation exactly.

Safety

A heavy machine moving near people must detect contact and stop instantly. Proving that a robot is safe around workers is as important as making it capable.

Cost

Even if every technical problem were solved, a humanoid still has to make economic sense. Each one contains dozens of precision motors, gearboxes, sensors, cameras and a powerful onboard computer. To replace or assist a human worker, it has to be affordable to buy, cheap to maintain and reliable enough not to stop a production line. Industrial robot arms became common because they do one task, bolted in place, millions of times without failing. A walking general-purpose robot has many more ways to break.

Why not wheels?

A fair question is why legs at all. Many warehouse robots roll on wheels, and some humanoid designs put a human-like upper body on a wheeled base. Wheels are simpler, more stable and more efficient on flat floors. Legs win on stairs, steps, cluttered floors and spaces such as homes and older factories that were never designed for machines. Companies betting on full humanoids are betting that one body able to go anywhere a person goes will be worth the extra difficulty.

Why progress is speeding up

Better AI models for vision and planning, cheaper actuators and large-scale training in simulation are moving humanoids from lab demos toward pilot jobs in factories and logistics. Researchers are also training robot control policies on large collections of demonstrations, recorded from people operating robots remotely, in a similar spirit to how language models learn from huge amounts of text.

What to watch for

Polished demo videos are easy to make: a task can be filmed many times, sped up or performed with a person controlling the robot remotely. When you see a humanoid announcement, look for a few things:

  • Is the robot acting on its own, or teleoperated? Honest companies say so on screen.
  • Is the video at real speed?
  • How many hours does it work in a real customer’s site, and how often does a person have to step in?

Hours of unassisted, paid work in real facilities are the measure that matters. That number, more than any demo, will show when humanoids are ready to leave the pilot stage.

What we think

Humanoid robots are making real progress, but they are still in the pilot stage, and the gap between an impressive video and a reliable worker on a real factory floor is large.

Our view is to judge the field by hours of useful, unassisted work rather than by demos. When companies start publishing those numbers from customer sites, and the numbers are good, that will be the signal that humanoids are ready for everyday jobs.