A robot doing a backflip is a very effective way to win the internet for an afternoon. A robot reliably moving boxes for an entire shift is harder to turn into a viral clip. It is also much closer to why businesses buy robots.
The scale of that practical work is enormous. The International Federation of Robotics reported in September 2026 that five million industrial robots were operating in factories in 2025. Annual installations passed 600,000, up 11% on the previous year.
Its separate service-robot report, released that month, describes almost 250,000 professional service robots shipped in 2025. Those figures cover different categories. They should not be mixed together as if every robot is a factory arm, a delivery cart and a humanoid at once.
The room is part of the software
Web developers can often fix a mistake by changing code and deploying again. A machine operating beside people has to handle a much less forgiving environment.
A box arrives at a slightly different angle. A sensor gets dirty. A wheel slips. Someone puts a trolley in a route that was clear a minute ago. A system that works in a carefully arranged demonstration may struggle when the room starts behaving like a real room.
That makes integration fascinating. The robot's body, sensors, control software and working environment all affect the result. Clever perception is useful, but so are predictable lighting, sensible storage layouts and a recovery process an operator can follow.
Count interventions, not applause
A useful deployment question is how often a person has to step in.
A robot that completes a task in ten seconds but needs rescuing every few attempts may create more work than it removes. A slower system that handles the ordinary variations and stops safely when it is unsure can be a better fit.
That suggests a different set of measures from the ones shown in a launch video:
- Completed tasks across a full shift, rather than the fastest single attempt.
- Human interventions and the time each recovery takes.
- Behaviour when a sensor or network connection fails.
- Maintenance effort and the availability of replacement parts.
These are engineering questions, but they are also product questions. Operators need clear status, understandable faults and a way to recover without improvising around a moving machine.
Useful can look ordinary
Humanoid designs are interesting because our buildings and tools were made for human bodies. That does not mean every useful robot needs legs or a face. A fixed arm, a wheeled platform or a specialised gripper may suit a job better.
The robot boom is worth watching through that lens. Which systems keep working after the camera crew leaves? Which tasks become safer or less repetitive for the people doing them? Which deployments can be maintained by the team that actually runs the site?
A reliable answer to those questions is more impressive than a backflip. Even if the backflip is pretty cool.
Cover: original AI-generated editorial illustration.
