AI gets a body: why Robotics 2.0 starts now

Lukas M. Ziegler likes to call himself a robotics evangelist. Not because robots are a fun toy, but because he is convinced that the next major AI wave will not take place on a screen, but in the physical world. Artificial intelligence can now write texts, generate code, plan trips and accelerate research. But the real breakthrough, he argued during the Holland High Tech Networking Event, will come when AI gets a body. “Physical AI”, or embodied AI, is no longer a hobby topic for technical nerds, but a strategic field for young engineers, companies and institutions. Because "Labour is becoming scarce, demand for production and logistics keeps rising, and robots are no longer merely executing tasks - they are beginning to understand them."

Ziegler started his talk with a simple question: why robotics, and why now? His answer was not primarily technological, but societal. Europe is facing a demographic shift that is unfolding slowly, yet is already being felt everywhere. Demand for production, logistics and services continues to rise, while labour is becoming scarcer. “For the first time in history, human labour is truly becoming a luxury,” he said. We do not want to order less, produce less or wait longer for our parcels. On the contrary: the expectation remains that everything should be delivered tomorrow. “Something has to fill that gap. I believe that something is robotics.”

According to Ziegler, there is a second layer to this: if robots are to play a larger role, a solid technical foundation is needed to maintain, service and safely operate them. Robotisation therefore does not simply mean less manual labour; it also creates a new demand for technical talent. The third factor is the sheer scale of the economy behind logistics. Ziegler referred to the global e-commerce market, which he said is heading towards 7 trillion dollars. “Every dollar of that represents a package that needs to be checked, sorted, shipped and delivered.” Automation, he argued, prevents the system from collapsing under its own weight.

Unimate

Yet his story did not begin with humanoids or AI models, but with Unimate, the classic industrial robot that carried out repetitive tasks at General Motors. That first robot revolution revolved around precision, repeatability and fixed programmes. Robots were strong, reliable and predictable, but also inflexible. Ziegler contrasted that with Robotics 2.0: robots that learn from data, recognise patterns, adapt and are able to work with intention. “In Robotics 1.0, we taught the robot what to do. In Robotics 2.0, we expect the robot to learn by itself.”

According to Ziegler, we are no longer standing at the edge of a new era. We have already entered it. “We are no longer approaching Robotics 2.0,” he said, “we have entered a new era.” The difference is not only better hardware, but above all, data. Companies are building world models, collecting images, movements, and actions, and training robots to handle variation. This also changes the role of system integrators. They no longer merely install and maintain robots; they are close to the customer, the production environment and the industrial data needed to make robots truly smart.

Ziegler showed that cognitive robotics is already visible in many different settings. In oil, gas, chemicals and energy, autonomous or tele-operated robots can inspect dangerous installations. Quadrupeds such as Anybotics’ ATEX-certified robots can enter environments where humans would be at risk, collect data and detect anomalies. In other examples, the value lies less in legs or arms and more in software: robots that translate inspection data into business insights, or autonomous security systems that patrol sites, checking for open gates, intruders, licence plates or holes in fences. The underlying message: robots are not only hardware; they are also software.

Tetris

He then broadened the picture to robot arms, logistics and recycling. Robots can load trucks as if they were playing Tetris, building walls of boxes in real time and using space more efficiently. In recycling, vision systems turn images of waste streams into data, after which materials such as metal, cardboard or aluminium can be classified. Even a robot chef from Cambridge fitted into his story: a system that analyses videos of chefs, derives recipes and tries to reproduce their actions. The example was partly light-hearted, but the lesson was serious: video is valuable training material. Robots can learn from images how the physical world works.

The most charged part of the presentation concerned humanoids. Ziegler acknowledged the hype, but did not avoid the critical question: why should robots look like humans at all? He is sceptical about domestic use. Who wants a 60-kilogram machine in the house that cannot yet be fully trusted? In industry too, he said, the business case is often still weak today. During the Q&A, he was explicit: at the moment, he sees “no real payback” for humanoids in many industrial applications. Sometimes they are mainly a PR instrument, a way for companies to show that they are innovating.

Still, he did not dismiss humanoids. His advice: do not start with the dream of a single robot that can do everything, but with one task. Successful companies often begin with a narrow application, such as moving totes in logistics or welding in a shipyard. Only then do they expand horizontally to similar tasks. The promise of a general-purpose robot that can wash dishes, weld, sort goods and work in warehouses is attractive, but the road towards it is far more complex. Hype can even be useful, Ziegler added later: robotics has long struggled to attract capital and talent. Humanoids open the door to more attention, more investment and more young people choosing robotics.

Cross-embodiment

Three technological breakthroughs could further accelerate the field, according to Ziegler. The first is cross-embodiment: one type of robot brain capable of controlling different forms, from a robot arm to a robot dog or a humanoid. The second is dexterous manipulation: the combination of hardware and software that enables robots to learn fine motor tasks that are obvious to humans but extremely complex for machines. The third is sim-to-real: reducing the gap between simulation and reality, so that robots can learn to deal with scenarios in virtual environments before being deployed physically.

So what is needed to scale robotics for real? Ziegler named four conditions: industrial data, reliability, edge models and integration. There are enough demos and enough lab data, but real production data is often missing. Safety and reliability engineering are crucial if robots are to enter the factory floor. Edge AI is needed because robots cannot always wait for the cloud; they must be able to respond locally, quickly and safely. And integration ultimately determines whether a robot becomes useful in an existing production environment.

For Europe and the Netherlands, Ziegler sees opportunities precisely because of the strong industrial knowledge, system integration and manufacturing base already present here. Dependence on Chinese robot components, actuators and supply chains is increasingly being discussed internationally. European companies can play a role in components, production, integration and safe applications. But that requires talent. His closing message to the hightech audience was therefore strikingly simple: lower the barrier to robotics, start early, and make the field visible and attractive. “Make manufacturing and robotics sexy again,” he said. Not because robots will simply replace people, but because the next industrial wave requires people who can give AI a body.