In recent years, there have been massive advances in hardware and software in the technology development surrounding human-like robot designs. Companies are outdoing themselves with promises about how many of these robots they will produce in series, when and what they will then be able to do. China in particular is pushing ahead. More than 120 companies are focusing on the development of humanoid robots. China experts like Georg Stieler project that around 10,000 humanoids will be built in China this year, mainly for research and entertainment. The combination of mass and speed with which developments are being driven forward should definitely make German companies sit up and take notice. Humanoids between desire and reality Interest in humanoid robots is also great in Germany. They could act as universal helpers in environments designed for people, taking on tasks flexibly and independently and meeting the challenges of demographic change and the shortage of skilled workers. But the distribution still falls short of expectations, fueled by PR, individual flagship projects primarily in automobile production and remote-controlled live demos. In practice, even purchasing a humanoid robot is often more complicated than expected. Inquiries to leading manufacturers show that very few humanoids are actually available for purchase. In addition, companies face legal, safety-critical and technical challenges when using these new types of robots. This is where application-oriented developments on humanoids come into play. What actually is a humanoid robot? A generally valid definition is currently still being found. This would be important for end users, but also system integrators or robot manufacturers themselves, in order to be able to better answer questions about safety and economic efficiency. A suggestion: A humanoid robot is characterized by two arms, an upper body and the ability to move on its own two feet or using a mobile platform. It optionally has a head, a gripper or sensors for detecting the environment. Similar to cobots, the small, compact robot arms, the following applies: There are no safe robots, only safe robot applications. This means that users are not only allowed to look at the humanoid robot itself, but also the entire application, the humanoid robot system. This also includes safety components, tools, workpieces and its operating environment.Application potential and hurdles from the perspective of German companiesIn view of demographic change, interest in humanoids is also great in Germany. In a study financed by the Baden-Württemberg Ministry of Economic Affairs as part of the AI progress center “Learning Systems and Cognitive Robotics” in cooperation with the VDMA, over 100 companies from production and logistics provide insights into desired application scenarios for humanoids. This includes material handling, machine loading and gripping complex objects. Humanoids could stand out from existing solutions with their flexibility combined with mobility. This would be particularly interesting for integration into existing plants (brownfield) in order to keep costs low. Despite promising prospects, those surveyed are skeptical about the technical possibilities. For now, many consider humanoids only suitable for tasks where accuracy and process speed are less critical. In addition, 60 percent believe that legs are not essential and prefer wheel-driven platforms or two-arm robots. This assessment is also reflected in current developments: Many manufacturers such as Humanoids (Great Britain), Neura Robotics (Germany) or Galbot (China) place the humanoid upper body on a wheel-driven platform in order to remove complexity. A central concern of those surveyed is the not yet established implementation of safe humanoid robots that can work together with people. There are also economic concerns. Fields of action for application development Building on this knowledge, the focus is on the following research fields in order to make humanoids more ready for industrial use: Economic efficiency In addition to the acquisition costs, humanoid robots are expected to have one-off engineering and setup costs for a task in the next few years. To get started, you should focus on applications that are technically less complex and at the same time have scaling potential for multiple robots. Better utilization of individual robots also increases profitability. Methods such as an automation potential analysis offer decision-making aids with regard to technical options and investments. Fraunhofer IPA is also currently developing a benchmark for humanoids, a catalog of criteria in order to be able to compare robots based on their technical key figures and equipment. This benchmark also helps end users evaluate the cost-effectiveness and select the right robot based on criteria such as master data, capabilities, energy consumption, IT security, functional safety and cleanroom suitability. SafetyThe use of humanoid robots poses significant risks of collision, especially when close to people, as current safety technologies often do not meet the standards of stationary collaborative robots. In the “KMUmanoid” research project, also funded by the Baden-Württemberg Ministry of Economic Affairs, extensive safety tests were carried out with the humanoid Unitree G1. In the worst case scenario, it can affect people with forces of up to 500 N. This significantly exceeds the permissible forces for operation close to humans and can lead to serious injuries. The ISO TC 299 Working Group 12 is dedicated to the topic of the safety of humanoids under the title “Safety requirements for dynamically stable industrial mobile robots (with legs, with wheels or other types of locomotion)”. It was launched at the beginning of 2025 and representatives from all manufacturers of humanoid robots, mainly based in Europe and America, are active in it. However, a new standard is not expected until 2028. In general, the use of humanoids should always be accompanied by security experts. It is currently advisable to combine several security measures in stages. Data availability and programming of humanoid robots Potential users primarily expect flexibility in the execution of tasks from humanoids, which excludes classic, rigid programming. Instead, technologies that enable autonomous task execution are necessary. Real data is the key to this because it opens up learning processes based on imitation or reinforcement learning as well as foundation models and thus expands the degree of autonomy of the robots. The topic of data factories, which China has already pioneered, is also crucial for Germany: In these, extensive application-related data sets are generated and made available for new applications. An example is Open In addition, research is being carried out on vision-language-action models (VLAM) (e.g. Google Gemini Robotics), which give robots a better understanding of their environment and tasks based on various sensor data and control the robot accordingly. When it comes to the topic of humanoids, the dynamics are enormous. At the same time, there are still many open questions to be solved. Flexible multi-purpose robots in the context of the much larger field of “embodied AI” are the clear goal. This is also evident from the federal government’s high-tech agenda presented at the end of October. In this context, it will be very exciting to see what the “AI robotics booster” announced there for 2026 will have to offer. Simon Schmidt Simon Schmidt is head of the “Automated Assembly, Manufacturing and Intralogistics Systems” division at the Fraunhofer Institute for Production Engineering and Automation IPA.Image: Private Dr. Werner Kraus Werner Kraus has headed the “Automation and Robotics” research area at the Fraunhofer Institute for Production Engineering and Automation IPA since 2019. In this role, he develops technologies and applications related to industrial and service robots for companies of all sizes and industries. A particular focus is on AI-based solutions for cognitive robotics. Image: Fraunhofer IPA
Go to Source