During Automation Experience, Marco van der Hoeven, editor-in-chief of Rocking Robots, and Richard Kuijpers of Smart Robot Solutions delivered a joint keynote on the rise of humanoid robots in the manufacturing industry. Van der Hoeven outlined trends and the figures behind the hype, while Kuijpers showed how companies are working with these machines in practice, and how expectations often far exceed reality.
Van der Hoeven, who has spent more than 25 years as a journalist watching technological hypes come and go, argues that none has been as profound as the current convergence of AI and humanoid robotics. He sketched a brief history: the first industrial robot was deployed at General Motors in 1961, in the automotive industry. It is no coincidence that this same sector is where various humanoid robots are now being scaled up.
The figures circulating around global humanoid robot production should, according to Van der Hoeven, be taken with a grain of salt. Research agencies use varying methodologies, countries report differently, and it is often unclear whether the numbers refer to sales, deployment, or pure production. Estimates for annual output range roughly between 13,000 and 22,000 units, a considerable spread. What is clear, however, is the regional breakdown: China is currently well ahead, American companies are at the bottom of the rankings, and Europe holds a middle position.
This gap has not gone unnoticed in Europe. At the European Robotics Forum, held this year in Stavanger, industry association euRobotics put forward what has since become known as the Vienna Convention: a European standard designed to give the region a more level playing field relative to American and Chinese robots, partly with an eye to European autonomy and sovereignty. Francesco Ferro, president of euRobotics, is leading this initiative. Progress is also visible elsewhere: German robotics manufacturer Neura, which partly produces humanoid robots, received an investment of €1.4 billion, and UK-based company Humanoid was given the opportunity to deploy robots in a real factory environment at Schaeffler.
Not Just a Colleague
One underestimated factor is the interaction between humans and robots. At demonstrations, people tend to watch with fascination, but on the shop floor, a humanoid robot must be treated seriously as a machine: weighing 70 to 80 kilograms and in motion, a robot is not automatically safe and cannot simply be deployed alongside people or left to stop abruptly. Videos are already circulating online of robots making awkward lunges, a clear signal that these are serious machines whose weight and form must be factored into their design.
That form varies considerably depending on the application, as not every humanoid robot is intended for the factory floor. In Australia, models are being used in care homes for social interaction with elderly residents, while Realbotix supplies realistic robots that serve as receptionists at companies including Ericsson. This illustrates just how diverse the humanoid robot market has become.
High Expectations, Cautious Practice
Richard Kuijpers picked up the thread with a look at what his company, Smart Robot Solutions, encounters daily in the requests it receives. The company operates across a wide range of robot applications, from hospitality to reception robots, and now receives inquiries specifically about humanoid robots on a weekly basis. Customers often expect a robot to get straight to work, expectations that Kuijpers says are largely shaped by science fiction films. In his view, the technology is at a stage comparable to the early Pentium processors: the beginning of something big, but with significant steps still to be taken.
How to Teach a Robot to Walk, Grasp, and Decide
Kuijpers devoted the bulk of his presentation to how humanoid robots are trained. Robots essentially arrive “empty” from the factory, he explained: they can walk, but still need to learn how to perform specific tasks on a production line. Training begins in a digital, simulated environment in which the robot’s weight and physics are replicated. What is learned there can be transferred to the physical robot with only minor deviation, far more efficiently than six months of training on the actual shop floor, and this is a method being actively explored worldwide.
Other training approaches also exist. Employees can wear an exoskeleton whose movements are directly transferred to the robot, for example when harvesting strawberries. Another method involves simply filming human actions, after which a model uses that footage to imitate the behaviour, significantly faster than fully programming a robot from scratch.
Programming in languages such as Python works well for a robot arm with six degrees of freedom, but a humanoid robot requires 80 to 100, compared to roughly 160 in a human being. That is no longer manageable with simple code and requires AI systems that can derive movements from images and video, right down to recognising the effect of gravity and weight in a photograph of a tomato in a pan, translated into the control of a robot arm. The most advanced models go a step further still, capable of making real-time decisions based on spoken instructions and their own perception, for instance, whether to jump onto a chair or a table.
Be Prepared
In practice, humanoid robots are already running around the clock in Chinese factories, including independently swapping out their own batteries. They carry out simple tasks such as moving objects from A to B, but also more complex operations. Kuijpers estimates current productivity at 14 to 30 percent of that of a human, a result of limited training data. That situation will improve quickly, he predicts, to the point where robots become faster and smarter than people. His advice to companies: start building experience now, even if the technology is not yet mature, precisely so you are ready for what is coming.
