Humanoid robots have started rolling off a production line in Guangzhou. The notable fact is not the capability of the machines, which has been extensively documented in demonstrations for years. It is the shift from prototype to mass production, which moves the question from what is possible to what it costs per unit.
We have covered robots learning through simulation. Here is the industrial stage, and it should be read without getting carried away.
What mass production changes
A demonstration proves that something is possible. A production line proves that it is reproducible at a given cost, which is information of an entirely different nature.
Three consequences follow.
Price becomes a parameter. As long as a robot is a prototype, its cost is meaningless. Produced at scale, it becomes comparable to an annual salary, to a machine tool, to an investment that can be depreciated.
Data accumulates. This is probably the most important effect in the medium term. Deployed robots generate data from real-world use, which improves the models, which improves the next robots. This is the loop we described in relation to acceleration through loops.
The supply chain takes shape. Motors, sensors, gearboxes: volume production brings down the cost of components, which brings down the cost of the next robots.
What remains very far from resolved
We need to be precise about the limitations, because demonstration videos systematically conceal them.
Dexterity. Walking and carrying a crate are largely solved problems. Manipulating varied, soft, slippery or fragile objects is not. The human hand remains out of reach, and it is the hand that performs most tasks.
The unexpected. A structured industrial environment, where everything is in its place, is infinitely simpler than a cluttered warehouse or a home. Real deployments happen in the former case.
Energy autonomy. A bipedal robot consumes a great deal just to stay upright, before it even starts working. Operating times between charges remain short.
There is a genuine technical debate that is rarely mentioned. The human form is difficult to design, costly and unstable. A fixed arm or a wheeled robot does the same work far better and far more cheaply. The argument in favour of the humanoid is that it fits into a world already designed for humans — stairs, handles, tools. It is a serious argument, and it assumes that the extra cost of the form is lower than the cost of adapting the environment. For a new factory, that assumption is dubious. For an existing building, it holds up.
What this implies for employment
This is the question everyone asks, and it deserves a cautious answer.
We reported that after three years of software AI, nine out of ten executives see no effect on employment. Robotics will run into the same obstacles, amplified: it requires capital, reorganisation, maintenance, and machines break down.
What seems reasonable is to expect deployment first in dangerous, repetitive and physically gruelling tasks, in structured environments. That is where the cost-benefit ratio is most favourable, and it is also where replacement is least contestable.
What to take away
This shift to mass production is a real milestone, and it says nothing about when you will pass a robot in the street.
What to watch is not the number of units produced but two more mundane figures: the unit cost and the uptime rate under real conditions. A machine that costs three times an annual salary and works half the time interests no one. The same figures reversed would transform an industry.
It is the same discipline we apply to models: demonstrated capability says nothing about real-world use, and it is the unglamorous indicators that decide.