In recent years, scenes once confined to science fiction—robots walking on two legs, carrying conversations, assisting in human environments—have begun to unfold in research centers and technology demonstrations. Headlines announce a coming “robot revolution,” and start-ups promise commercial humanoid systems by 2026. Yet a closer look suggests something more measured is taking place: not a sudden upheaval of work, but the start of a long transition from experimentation to carefully controlled industrial use.
“Humanoid robots are fascinating because they share our physical form,” says Dimitra Anastasiou, a researcher at the Luxembourg Institute of Science and Technology (LIST). “But what people often forget is that robots are not always physical or tangible things. A robot can also be software, a program, an algorithm, a cognitive system that guides decisions and interactions. The intelligence behind the robot is what truly matters.”
From fiction to human-like machines
The word robot itself dates back to a 1920 Czech play and comes from robota, meaning forced labor. Today, robots are understood as machines or systems designed to perform tasks that are repetitive, dangerous or complex. A humanoid robot adds another layer: it is shaped like a human, with a head, torso, arms and legs, allowing it to function in environments designed for people. These robots combine sensors, artificial intelligence, language models and physical manipulation to interact with objects and humans in increasingly natural ways.
What has reignited interest in humanoid robots is not simply better mechanics, but the convergence of robotics with powerful artificial intelligence. Large language models and advanced perception systems now allow robots to understand spoken instructions, reason about tasks and adapt to dynamic surroundings. For the first time, interaction no longer depends solely on programming lines of code, but on conversation and context.
This technological shift explains why several companies have already planned commercial deliveries. China has begun large-scale humanoid robot production, and in the US, some companies opened full-scale humanoid robot factories. Still, these developments should not be mistaken for immediate mass deployment. Most systems will initially operate in tightly supervised pilot projects, performing limited and clearly defined tasks.
Safety, reliability and understanding humans
“The real challenges are not whether a robot can walk or talk,” says Dimitra Anastasiou. “They are about safety, reliability over time, understanding human intentions and integrating robots into real industrial workflows.”
Dimitra Anastasiou, together with Ben Gaffinet and Yannick Naudet, explores this pragmatic view in their paper A Cognitive Social Robot in Manufacturing. Their work does not imagine humanoids as replacements for factory workers, but as intelligent interfaces that support and guide them. The researchers focus on a social robot called QT, developed by LuxAI, and place it in a simplified manufacturing scenario: assisting a human participant in assembling a Lego product.
QT does not manipulate the pieces itself. Instead, it communicates through speech, gestures and eye contact, using a large language model enhanced by Retrieval Augmented Generation to consult assembly instructions and provide context-aware feedback. In future versions, the robot will be connected to a so-called Human Digital Twin, a software representation of the worker’s state and actions. This digital twin can anticipate mistakes—such as choosing the wrong component or tool—and prompt the robot to issue a warning before damage or injury occurs.
“In this setup, the robot is not the worker,” explains Yannick Naudet, scientific coordinator at LIST. “It is the cognitive bridge between the digital system and the human operator.”
Insights from user studies
The results of their exploratory user study were revealing. Most participants described the interaction as engaging and even enjoyable. Yet only a small number assembled the product exactly as intended. Many struggled with ambiguous language, terms like “on top,” “layer” or unclear orientation instructions. The difficulty was not the robot’s intelligence, but the subtle complexity of human communication.
These findings underline a central lesson for the future of humanoid robotics: success depends as much on language, feedback and trust as on hardware and algorithms. The researchers at LIST note that once robots interact with humans, they inherit all the ambiguities of human language and behavior. Designing that interaction becomes as important as designing the machine.
Industry 5.0: robots as collaborators, not replacements
This perspective fits squarely within the vision of Industry 5.0, which emphasizes collaboration between humans and intelligent systems rather than pure automation. In this model, robots whether physical humanoids or invisible software agents support human decision-making, reduce risk and free workers to focus on creative and complex tasks. The robot becomes less a replacement and more a colleague, or even a guide.
Deploying humanoid robots safely will require new standards for certification, clearer rules for accountability and careful attention to worker acceptance. Regulation will need to address not only machines on factory floors, but also the software systems that monitor, predict and intervene in human actions.
It is tempting to frame humanoid robots as the next great industrial revolution. But current evidence suggests something quieter and more deliberate is underway. These systems are leaving the laboratory, but they are entering the world cautiously, through pilot projects and controlled environments.
“The revolution is not about building a robot that looks like us,” says Dimitra Anastasiou. “It’s about building systems, both physical and digital, that understand us and work with us.”
The true test of humanoid robotics will not be whether robots can walk or speak, but whether they can collaborate safely, reliably and meaningfully with people. If that succeeds, the future workplace may be shaped not by replacement, but by partnership between human intelligence and machine cognition.
This research was partially supported by the French National Research Agency (ANR) and the Luxembourg National Research Fund (FNR) through the project AI4C2PS (INTER/ANR/22/17164924, 2023–2025), in cooperation with Université de Lorraine. More information about the project can be found at https://www.ai4c2ps.eu/.