The robot is not the hard part. The factory is.
Why humanoid robots in industry succeed or fail less on an AI model than on integration with machines, safety and production.
The robot is not the hard part. The factory is. This view informed my post from the end of May 2026. The occasion was news from the two weeks before, and I want to explain it in more detail here.
What was reported?
In those two weeks, several industrial and software companies reported plans and trials with humanoid and autonomous robots in manufacturing and logistics. Humanoid robots moved one step closer to real industrial environments.
The simple headline is: humanoid robots are coming. I think it is incomplete.
Which question is harder than the headline?
The harder question is: can such a robot hold up in a factory? A factory is not a demo stage. It has poor lighting. Dust. Vibration. Safety zones. Older machines. Old protocols. Operator habits. Time pressure because of downtime. Missing data. Unexpected special cases. And expectations of cost-effectiveness that are not swayed by the attention the topic attracts.
Each of these points is known on its own. Together they make up the environment in which a robot has to work. On a stage, a robot can impress. In the factory, it becomes visible whether it fits that environment.
Is that a problem of the AI model?
Not only. I do not see AI for machines only as a question of the model. I see it as an integration problem in the embedded field.
The visible part is the robot. The value comes from the setup around it:
- machine vision,
- motion,
- sensors,
- actuators,
- real-time control,
- edge AI,
- industrial communication,
- safety logic,
- connection to existing production systems.
Each of these building blocks has to work together with the others, and all of them together with the machines that already stand in the hall.
Where is it decided whether a project succeeds?
This is where many projects will succeed or fail. Not in the press release. Not in the demo video. But during a Monday morning shift, when the robot has to work alongside real machines and real people under real production conditions.
This thought shifts the view. Anyone judging a robot often looks at what it can do. For a factory, the question is a different one: what can it do next to what already runs there?
What follows for businesses?
This suggests a sequence for businesses considering robotics or AI in production. First the environment: light, dust, vibration, safety zones, existing machines and protocols. Then the integration: how sensors, control, communication and safety logic work together with what is already there. Only after that the question of which robot or which model fits.
In my view, this order is the reason why, in the debate about humanoid robots, I look less at announcements than at the question of how a robot is integrated into an existing production process process. That is where the work is, and that is where it is decided whether a project makes it into production.
Why is this a question for embedded development?
Because the list of building blocks shows where the work lies. Machine vision, sensing, real-time control and industrial communication are topics at home in the embedded world. For me the conclusion is: success does not depend on the model that controls the robot alone, but on whether the whole system made of all these parts copes with the factory.