Why humanoid robots are not just hype, but a breakthrough in robotics

In the Russian internet, humanoids are hardly ever discussed. Well, until I start. I'm starting.

Time and again, corporations like Nike, Boeing, Adidas pour tens of millions into “tin” robots. It often ends with not always, but frequently dismantling lines and slave exploitation hiring unskilled workers. Why humanoid robots might change the situation.

Universality vs Specialization. The New Paradigm.

• MP3 player

• GPS navigator

• Alarm clock

• Camera

Here's a far from complete list of things that smartphones have replaced (or they're still used, but rarely and situationally)
Although each of them does its job better than a smartphone and can be cheaper (individually), I have never seen anyone voluntarily give up their smartphone in favor of such a zoo. To me, we are currently on the brink of a revolution in robotics, in terms of consequences no less significant than the invention of the iPhone – effective humanoid robots.

Moreover, androids are already used in car assembly and logistics. So the technology has already matured to "first successes in production," which is already serious. Although it still lags behind industrial robots, which have been in use for half a century.

Many criticize humanoids for their apparent inefficiency, but miss the point: their strength is not in surpassing specialized machines in a single task, but in universality and compatibility with a world built for humans

"Project Glory" from Nike — how NOT to do robotization.

I already made a post like this before, in digest form. But I realized it's better to make a post as "an opinion." I’ll also add information from the comments to my previous post. Special thanks to @baldr for the active discussion.
Key points: Nike, Adidas, and other giants wanted to automate their shoe production. And they collided with reality

• Inflexibility of robots: if the shoe sole slightly differs from the last one, the machine might just apply glue in the wrong place. While a human worker would quickly adapt.

• Diversity problem: it turned out to be too difficult to work with different types of shoes.

• Design vs. technology: Nike's business is built on designers, whose work was hindered by the limitations of robots. Either design or robots – one or the other.

Conclusion: The project failed. Millions were spent, but the goals were not achieved. Factories closed, and the focus shifted back to Asia. The takeaway from this case: trying to implement highly specialized robots into a complex, ever-changing human process is a path to failure.

The solution to the problem, or the two pillars of the humanoid robot revolution

Why universal humanoid robots can be deployed where specialized non-humanoid ones failed.

1st Pillar: Infrastructure – the untouchable King.

• Boeing remodeled half the facility and spent several years only to fail at robotization.
  There's no need to remodel the facility for androids.

• Maintaining a zoo of robots is harder than maintaining a unified one.

• Managers often underestimate the difficulties—which can lead to huge losses.

This sharply lowers the entry barrier. Initial investments are cheap and safe.

2nd Pillar: Training data. Humans are the best teachers for robots.

Versatility is impossible without neural networks, and neural networks need data. It's most efficient to collect data from human form.

• You can collect from simulation, but simulation is always simplified (meaning inaccurate) physics. A robot trained in simulation is guaranteed to only work in simulation.

• You can collect through "experience," and by the robot's own mistakes in the real world. But for "from scratch" training, this doesn't work.

• Collecting data from humans is expensive and slow, but it's very valuable.

This literally gives humanoid robots a cheat-level advantage.

But before, neural networks were much worse, and humanoid robots weren’t capable of useful autonomous work. Both Japan’s Miim in 2009 and the 18th-century "Chess Automaton" were only good for show.

Analysis of some counterarguments

• Myth: Human mechanics are imperfect and shouldn't be used in robots
  Fact: Androids don't actually copy human mechanics—they (usually) have no joints, no teeth, no skin. What matters is the form factor and compatibility.

• Myth: Robots move very slowly
  Fact: 3.5 seconds per package is actually faster than an average human.It's slower than seasoned professionals, but faster than beginners.

• Myth: They'll never be used in serious manufacturing
  Fact: Car production at BMW’s largest factory more than qualifies as "serious manufacturing."

• Myth: They'll never pay off—the cost is higher than the salary of two migrants for 30 years
  By my quick calculations, <$1 per hour per robot is noticeably cheaper than migrant labor in the US (and elsewhere). Not including, of course, the engineers’ salaries. But HR isn’t included either.

It’s clear that there are tasks where humanoid robots are unnecessary. They shouldn't be used instead of a washing machine. They should be used to stuff clothes into it.

Not everything will fit, I’m open to discussion in the comments. If tekkix is inconvenient, feel free to message me privately on Telegram

Conclusion: A robot that doesn’t make production change.

The smartphone didn’t make us carry a gadget backpack, it fits in our pocket. An android doesn’t force factories to rebuild and break technical processes. Therefore, I am confident that in 10-15 years, there will be more androids than people.