Robot autonomy relies heavily on human interpretation and manual correction

One of the most significant insights from the study is that roboticists must engage both technically and physically with their systems. Programming is not a detached intellectual exercise; it involves bodily immersion. Engineers crouch beside robots, listen to motor sounds, adjust joint resistance by hand, and physically position limbs to demonstrate intended paths. This embodied involvement reveals that robot autonomy is deeply dependent on human presence and intervention.

Humanoid robots that appear in public demonstrations conceal the intricate, messy, and deeply human processes behind their creation. New research offers an unprecedented look inside robotics laboratories, exposing how humanoid robots are fabricated through iterative translations, careful calculations, and the continual negotiation between expectation and material reality.

A peer-reviewed article titled “The fabrication of humanoid robots: methods, media and artefacts,” published in AI & Society, examines these behind-the-scenes processes through ethnographic fieldwork in three German robotics labs. The study shifts focus away from human–robot interaction and instead documents how roboticists themselves work, think, and problem-solve while building human-shaped machines. The findings challenge widely held assumptions about robot autonomy and human likeness, showing that robots are, above all, products of extended human labour, cultural interpretation, and experimental trial and error.

Robot fabrication begins with observation and handcrafted abstractions

The study finds that fabricating a humanoid robot begins long before mechanical parts are assembled or software is executed. Robotics teams first translate real-world objects and motions into simplified representations they can work with. Through careful observation of familiar items such as doors, handles, and drawers, engineers break down physical interactions into core mechanical gestures like pushing, pulling, gripping, or rotating. These analogue objects anchor the design process and help roboticists imagine how a robot might move within human environments.

But this initial translation is highly reductive. A door, for instance, may function differently depending on its weight, hinge tension, or orientation. To manage this variability, researchers create handcrafted sketches, drawing abstract shapes to represent objects and hypothetical robot movements. These sketches reduce three-dimensional reality into two-dimensional lines, enabling engineers to isolate forces and directions. Although simple, these drawings become powerful tools, forming the conceptual bridge between physical spaces and mathematical formalisation.

The study highlights that handwriting and sketching are essential components of robot fabrication. Through these manual practices, roboticists externalise their thinking, test ideas, and negotiate shared understanding within teams. Sketches evolve rapidly as engineers revisit assumptions, correct interpretations, and refine what they believe a robot must do to perform a given task. This early stage, according to the researchers, is foundational to the entire fabrication process, shaping the artefacts and decisions that come later.

From this starting point, roboticists move to formalisation, converting sketches into mathematical expressions. Angles, vectors, torques, and joint rotations are calculated to describe how a humanoid robot could replicate the targeted movement. Here, abstraction intensifies. Real-world interactions are translated into formulas that appear systematic but rely heavily on assumptions about the precision and reliability of robot components. The study notes that these formulas often oversimplify the complexity of human-environment interactions, but they are necessary for programming to begin.

Programming, testing, and the constant confrontation with mechanical reality

Once sketches and formulas are in place, robot fabrication transitions to the digital environment. Engineers convert mathematical representations into program code, using simulation tools and scripting languages to instruct robot joints, motors, and sensors. However, the research shows that this translation is neither seamless nor predictable. What appears logically sound in mathematical form often behaves unexpectedly when converted into real motions on a humanoid robot.

Testing becomes a key phase in the fabrication cycle. Roboticists load new code into robot prototypes and observe how the machines respond. Movements that seemed fluid in theoretical models may become jerky or misaligned in practice. Robots may crash into objects, fail to grasp handles, slide off narrow edges, or become stuck mid-motion. The study captures moments in which robots repeatedly strike the same obstacle, fall into loops, or misinterpret the intended movement trajectory.

These failures are not anomalies, they are expected outcomes that shape the work culture inside robotics labs. Engineers spend long hours debugging, adjusting parameters, revisiting sketches, and recalculating forces. Testing often leads back to earlier phases, forming a recursive loop of correction and refinement. Humanoid robot fabrication thus emerges as a cyclical rather than linear process, constantly shifting between abstraction and physicality.

One of the most significant insights from the study is that roboticists must engage both technically and physically with their systems. Programming is not a detached intellectual exercise; it involves bodily immersion. Engineers crouch beside robots, listen to motor sounds, adjust joint resistance by hand, and physically position limbs to demonstrate intended paths. This embodied involvement reveals that robot autonomy is deeply dependent on human presence and intervention.

Even advanced humanoid robots exhibit fragility and unpredictability. Their movements are influenced by minute variations in friction, weight distribution, battery levels, and sensor calibration. These physical constraints impose limits on what mathematical models and digital simulations can predict. As a result, roboticists must continually negotiate between their imagined designs and what the robot hardware can actually perform.

Cultural assumptions and human-likeness shape every stage of robot creation

The study argues that humanoid robots are not neutral technological objects but are shaped by human cultural expectations. When designing human-like machines, roboticists embed ideas about what constitutes human behaviour, gesture, and physical ability. These assumptions influence everything from sketching styles to programming strategies to the aesthetic choices made in robot design.

Humanoid robots are therefore cultural artefacts as much as they are engineering creations. Engineers do not merely reproduce mechanical motions but attempt to simulate actions that appear recognisably human. Decisions about limb proportions, degrees of freedom, or grasping motions are intertwined with social interpretations of human normality. The fabrication process becomes a blend of technical reasoning and cultural imagination.

The research also highlights the social organisation within robotics labs. Robot fabrication is a team-based effort involving diverse roles, engineers, programmers, designers, and students, each contributing specialised knowledge. Collaboration is structured through meetings, shared workspaces, and informal exchanges. These interactions shape how problems are understood, how priorities are set, and how solutions emerge.

Failure, too, has a social dimension. Robotic experiments often fail in dramatic or humorous ways, prompting laughter, tension, or collective troubleshooting. These emotional responses help teams navigate the stresses of iterative work and the constant uncertainty surrounding robot performance. In this sense, the study underscores that building humanoid robots is not only a technical endeavour but also a lived social practice.

Another key finding concerns the limitations of representational tools. Sketches, formulas, and simulations each simplify reality differently, but none fully capture the complexities of physical interaction. Humanoid robots reveal these limitations through their repeated misalignment with planned movements. These persistent gaps demonstrate why robot fabrication requires ongoing human oversight, adaptation, and interpretation.

Humanoid robot fabrication cannot be understood through technical descriptions alone. It must be examined as a sociomaterial process involving diverse media, recurrent translation, and embodied labour. Robots do not emerge fully formed; they are continually shaped by the constraints of their materials, the tools used to model them, and the cultural assumptions embedded in their design.

A new perspective on robotics 

The study challenges widespread public narratives that position humanoid robots as autonomous machines approaching human-level capability. Instead, it reveals how robots depend on extensive human input at every stage of their creation. The fabrication process is a dynamic interplay between abstraction and practice, imagination and friction, design and revision.

The authors argue that attention must shift from end-stage human-robot interaction to the ongoing labour that makes robot interaction possible. By examining the methods, media, and artefacts that shape robot fabrication, the study offers a fuller understanding of how humanoid robots come into being and why their behaviour remains constrained by the tools and assumptions used in their creation.

The findings also point toward the need for deeper investigation into the institutional structures surrounding robotics. Funding priorities, academic norms, publication incentives, and industry expectations all influence the direction of humanoid robot research. The study suggests that future work should map these broader dynamics to better understand how robotics as a field evolves.