Robotics vs. Reality Humanoid Robots: Why the Industry Needs Wheels Instead of Legs

Related Vendors

Taiwan Machine Tool & Accessory Builders' Association (TMBA)

SIMTOS (Seoul International Machine Tool Show)

The hype around humanoids is real. But what do they achieve in the industry when they're not performing flips and dancing in viral videos? We look behind the hype and at the current reality of humanoid robots.

At the Humanoid Days at the University of Erlangen, Dr. Maximilian Metzner, Global Lead Manufacturing Electronics at the Siemens site in Erlangen (Germnay), provided insight into the current state of humanoids and their industrial practicality.

The Siemens device factory in Erlangen is a classic brownfield environment: The factory was built in 1971 and has grown organically over decades. The approximately 1,000 employees in manufacturing produce items such as frequency converters for power electronics, which are used across industries—from simple electric motors to machine tools and battery production.

This broad spectrum is precisely the biggest hurdle for classical automation. There is no mass production in the sense of millions of identical components per year. Instead, an extreme high-mix/low-volume approach dominates: some products are built thousands of times a day, while others only 50 times a month. The physical nature also varies massively.

"At the end of the day, it's simply physics," explained Dr. Metzner at the Humanoid Days. "Even though we manufacture electronics, you can't create a purely software-based version of power electronics. Whether you draw five kilowatts or five megawatts of power fundamentally requires different circuit boards, components, and heat sinks."

To address demographic change and global cost pressures, the factory is heavily reliant on massive automation and digitalization. However, the solutions sought must be flexible, which is why Siemens relies on AI-driven systems.

In the Erlangen electronics plant, the wheel-driven humanoid robot HMND 01 Alpha from the English manufacturer "Humanoid" took on a clearly defined logistics task in a two-week practical test in spring 2026: it destacked material containers, transported them autonomously to the conveyor belt, and made them available there for employees. The so-called "Tote-to-Conveyor-Destacking" process is thus a real application case in factory logistics, but not yet an announced regular operation. According to Siemens, the robot achieved 60 container movements per hour and achieved a success rate of over 90 percent in autonomous gripping and placement processes.

The Appeal of the Humanoid: The "Drop-In Replacement"

But why is an established plant like Erlangen now focusing so intensively on humanoid or human-like systems? The answer lies in the very nature of the factory itself. Brownfield facilities were historically designed exclusively for humans and their ergonomics. The great advantage of the humanoid form factor is therefore its function as a so-called drop-in replacement. After all, they "hypothetically offer the greatest market potential of all time," as could be read on Metzner's slides. This refers to the replacement of human physical labor.

Additionally, the human-like physique significantly simplifies the training of artificial intelligence. Since the mechanics of the robotic arms resemble human anatomy, movement sequences can be directly transferred via teleoperation and used as learning data for the systems.

Perhaps the greatest economic appeal for plant operators lies in the drastic risk minimization. With fully automated special-purpose machines, there is often no Plan B in case of malfunctions. The humanoid approach, on the other hand, offers a simple fallback option, as Metzner emphasized:

"Classic automation is often a final decision: once a system is fully automated and robots are permanently installed, there is no going back to manual work the next day. A humanoid approach offers a built-in emergency concept here: if the robot has not yet mastered a new task reliably, we simply move it aside, and a human employee takes over the position again. No investment is lost, and operations continue."

Pragmatism Instead of Science Fiction: Why the Industry (Almost) Doesn't Need Legs

When humanoid robots are discussed in public, images of walking robots balancing on two legs over uneven terrain or climbing stairs dominate. For use in classic brownfields, however, these highly complex bipedal systems are currently neither mature nor economically viable.

Instead, Metzner advocates for hybrid systems: a mobile AMR base (Autonomous Mobile Robot) combined with a humanoid upper body (torso and arms). This form factor drastically reduces complexity and already enables robust, productive applications today. Human walking requires constant balancing calculations with every movement, which is not only extremely demanding in terms of software but also poses real regulatory challenges in the factory.

"You simply can't get bipedal systems certified and approved in an industrial environment today," the Siemens expert emphasized at the Humanoid Robot Days. "The corresponding standards are missing, the safety technology is incredibly complex, and it only limits you."

The relevance of a humanoid in the factory ultimately arises from the principle of exclusion, which Metzner summarizes as the "95-percent rule": if it's purely about material transport, a simple autonomous transport system is entirely sufficient. For highly specific, stationary tasks, a classic industrial robot is the better choice. Only when a task requires universal, mobile manipulation does the humanoid approach demonstrate its strengths. Even in over 95 percent of these applications, legs are simply not needed in the industry, Metzner adds: "Factory floors are rarely uneven and are generally designed for wheels."

Use Cases in Practice: From Cardboard Boxes to Cognition

But what exactly is this new type of robot supposed to achieve in the factory? A study by Fraunhofer IPA paints a surprisingly mundane picture: at the top of the industry's wish list are simple skills such as material transport (84 percent) and machine loading (79 percent).

The human-like form often raises false expectations about industrial value, Metzner emphasized. As an example, he cited the frequently demonstrated operation of handheld power tools by robotic hands. "Why should I design and program a highly complex humanoid hand just to operate a standard cordless screwdriver?" Metzner questioned these use cases. "There are already mature screwdriving systems that fit directly onto standard robot flanges."

To reflect the technological reality, Siemens divides the applications for humanoids into three complexity levels (tiers):

• Tier 1 (Low): Simple manipulation tasks with sufficient space, such as gripping and carrying standardized storage bins or boxes (Pick & Place). These use cases are already being addressed in proof-of-concepts (PoCs) in the Erlangen production facility.
• Tier 2 (Medium): Targeted actions in tighter environments (e.g., reaching into shelves) and greater dexterity to specifically remove individual parts from collection bins. According to Metzner, this level is not yet commercially available and is the subject of ongoing research.
• Tier 3 (High): Tasks requiring comprehensive, human-like dexterity and cognitive adaptation.

The high investment costs for humanoid hardware are ultimately justified only by their eponymous universality. "It makes no economic sense to place a universal robot at a machine for six hours straight to perform a single repetitive task," Metzner explained the principle of the 'monolith.' "Such a system pays off when it collects boxes for half an hour, then picks parts from a shelf, and subsequently operates the machine. The robot must be able to perform several different task chains in succession."

Software Proves to be a Bottleneck

Initial test runs with hybrid systems in Erlangen show that the pure hardware (chassis and arms) has often already reached a very good, industry-suitable level. Despite all the mechanical advancements, the actual problem lies elsewhere. While standard grippers often still prove to be a weak point in universal tasks, the real bottleneck in integration is the software.

Since mobile robots never stop at exactly the same point, fixed programming misses the mark. Everything must be sensor-supported and camera-based, calculated in real time. Metzner summarizes the challenges in the following six points:

• Early software stage: The software stack is in a very early iteration. Comprehensive application frameworks for programming on the robots are missing.
• Inherent complexity: The hardware combines a mobile base (or legs), a lift system, two seven-axis arms, and movable camera systems in one device.
• Synchronous skill execution: The system requires the absolutely synchronized interaction of navigation, object recognition, motion planning, and gripping process.
• Fully sensor-controlled: Since mobile robots never land in the exact same spot down to the millimeter, traditionally programmed waypoints are useless. Every grip must be necessarily calculated using sensor data.
• Unpredictable full-body movement: A humanoid body automatically compensates for movements. Metzner explained the problem as follows: "When you extend one arm, the other makes a compensatory movement for stabilization, even if it wasn't actively controlled."
• Complex collision avoidance: The system must dynamically predict its own changing form as well as all compensatory movements to avoid obstacles or humans.

"You have a mobile robot with two seven-axis arms and multiple camera systems that move relative to each other," summarized Dr. Metzner, adding, "That's roughly the nightmare of every roboticist."

USA vs. China: Two Poles in the Market

The Western Model (Closed Systems): Manufacturers from the USA (and partly Europe) typically offer a closed system that comes with pre-trained capabilities. Metzner compares this to the principle of a temporary worker: the robot arrives ready at the station, an operator sets it up, and it starts running. The drawback for industrial users: "To put it bluntly: they sell you nothing. You are offered a closed system where the task no longer even belongs to you because you are renting a service," Metzner described the "Robot-as-a-Service approach." Moreover, highly specific tasks in electronics manufacturing cannot be addressed by general closed data models. Not to mention that this offering simply doesn't exist yet due to a lack of skills.

The Asian model (Open but incomplete platforms): Manufacturers from China take the exact opposite approach: They already sell pure hardware platforms without pre-installed AI skills. This offers maximum freedom for in-house developments on paper. The problem here: the software is not yet fully developed, and the system is relatively complex.

Outlook: The Tipping Point for Humanoids

That humanoid systems, in whatever form, will eventually establish themselves in the industry is, for Metzner, not a question of "if" but rather a simple economic equation: on one side stands demographic change, which inevitably increases the cost and scarcity of human labor. On the other side are technological advancements and the decreasing unit costs of hardware. "If these robots become smarter through better models, then they can do more tasks and become universally applicable. That’s when the system pays off," predicted the engineer.

Humanoid robots have the potential to revolutionize automation in high-mix/low-volume manufacturing, as they fit perfectly into existing brownfield factories as a drop-in replacement. However, anyone expecting Hollywood-style androids on two legs will be disappointed. Industrial pragmatism leads through hybrid, mobile systems.

"In industry, it's not the coolest thing that prevails, but the most practical," sums up Dr. Metzner from Siemens. However, the real race for the factory of the future will not be won by building legs, but in software development. That is, in the battle for the best AI models that turn a piece of hardware into a true, flexible machine colleague.