The humanoid safety paradox: the more normal the more dangerous!
The more normal we make a humanoid behave, the more people expect it to behave normally. They walk closer. They turn their back to it. They trust it like they would trust a colleague.
But a humanoid is still a robot. It is a computer on two legs, and computers go wrong. A sensor glitches, the software freezes, a joint does something nobody planned. When that happens next to a person who has stopped paying attention, it can be dangerous.
The better humanoids get, the more serious safety has to become.
Borrowed from aviation
Aviation is the safest way to travel. It did not get there with better pilots. It got there by making it safe to admit mistakes.
In the 1990s the industry adopted what it calls a just culture. A pilot or mechanic who reports a near-miss is protected, by law in the EU since 2014. The report goes into a system. The system gets fixed. The next crew does not repeat the error. The only thing that gets punished is deliberately breaking a rule or hiding what happened.