On this page
Models and variants
When designing a robot that will walk among humans, work beside us, and perhaps even live with us, what should it look like? Should it be clearly mechanical, a...
When designing a robot that will walk among humans, work beside us, and perhaps even live with us, what should it look like? Should it be clearly mechanical, a tool that never pretends to be anything but a machine? Or should it mirror us so closely that the line between human and artificial becomes uncomfortable to draw?
In November 2025, this question exploded into public debate when Chinese EV maker XPENG unveiled its second-generation Iron humanoid robot at AI Day in Guangzhou. The reveal included something most robotics companies had carefully avoided: a robot with a distinctly feminine form.
The demonstration was so lifelike that CEO He Xiaopeng noted some colleagues asked during rehearsal, "is there a real human inside the iron?" The robot walked across the stage with what observers described as a gentle, model-like gait. Standing 1.73 meters tall and weighing around 70 kilograms, the Iron features 60 joints and nearly 200 degrees of freedom, allowing fluid and natural movements.
But it was not the technical specifications that went viral. Within hours, reactions poured in across Chinese social networks like Weibo and Xiaohongshu, as well as Western platforms. Users questioned: "Why sexualize a machine meant to work in a factory?" Some specialized Chinese media mocked this "unnecessarily gendered" design. Others saw it differently, viewing the aesthetic choices as a strategic decision about how humans will accept and interact with robots in service roles.
XPENG's CEO made clear the robot would not be used in factories or households initially. Instead, he explained, Iron would first serve as tour guides, sales assistants, and office building guides, beginning in XPENG facilities. The company plans to begin mass production by the end of 2026.
The controversy reveals something deeper than a design choice. It exposes a fundamental split emerging in humanoid robotics: machines designed to blend in versus machines designed to stand apart. And that split is driving the creation of distinct variants, each optimized for different roles in the world we are building.
Factory Robots: The Workhorses#
The most common humanoid robot you will encounter in the next five years will not be in your home. It will be in a warehouse, a factory, or a logistics center, moving boxes, inspecting parts, and assembling products. These robots prioritize function over form, strength over aesthetics, reliability over personality.
Tesla Optimus is being designed primarily for Tesla's own factories, where it will handle tasks like battery cell manipulation. The company has announced limited production planned for 2025, with broader deployment to follow. Optimus represents Elon Musk's bet that vertical integration in robotics can produce the same market advantage that Tesla achieved in electric vehicles.
Figure 02 is already deployed in real manufacturing. Figure AI's humanoid is being piloted in BMW facilities to handle manufacturing work, moving bins and boxes in the Spartanburg facility. The robot is designed to work in existing human environments without requiring factories to be rebuilt around its limitations.
Apptronik Apollo targets automotive and logistics applications. Apollo is already being tested by Mercedes-Benz for delivering parts and inspecting components in automotive plants. Unlike XPENG's approach, Apollo is expressionless and utilitarian, a machine that makes no attempt to appear human beyond its basic form factor.
These factory robots share common characteristics. They are built for durability and precision. They need to integrate seamlessly with existing machinery and survive long operational cycles in demanding environments. Their designs are converging on similar specifications: roughly human height and weight, enough dexterity to manipulate standard tools and objects, and the strength to handle payloads of several kilograms.
By 2050, if current trends continue, advances in materials science and computing power could enable these robots to tackle far more complex tasks. Imagine humanoids assembling rockets, constructing buildings, or performing intricate repair work in environments too dangerous for humans. But for now, they are learning to do what humans find repetitive, physically demanding, or simply undesirable.
The key insight: these robots do not need to look human. They need to fit into human spaces and use human tools. That is the extent of their anthropomorphism.
Service Robots: The Face of Automation#
If factory robots work behind closed doors, service robots will work in front of us. They will greet visitors, guide tours, answer questions, and represent brands. For these machines, appearance matters differently.
XPENG Iron represents the most aggressive bet on human-like design for service applications. The company revealed that its new robot features full-body synthetic skin intended to make it feel "warmer and more intimate," with customization options for body shape and hair style. The robot is powered by three of XPENG's self-developed Turing AI chips, which the company claims provide a maximum computing power of 2,250 TOPS (trillion operations per second).
XPENG's strategy diverges sharply from competitors. While many are targeting industrial or domestic applications, XPENG sees its robot succeeding as a receptionist, tour guide, or personal shopping assistant. The seventh-generation prototype already served as a tour guide at XPENG's headquarters in Guangzhou.
1X NEO takes a different approach to the home service market. Developed by Norwegian robotics company 1X, NEO is designed specifically for household environments. While it does not possess a traditional face, its design includes elements aimed at making it approachable and relatable. The robot is equipped with advanced AI, enabling it to perform tasks such as cleaning and assisting with household chores. Unlike XPENG's emphasis on aesthetic appeal, NEO prioritizes safety features and user-friendly operation for families.
The division between these approaches hints at a broader market segmentation emerging in service robotics. Some companies bet that people will accept and prefer robots that look distinctly mechanical. Others believe that human-like features will reduce resistance and increase adoption, particularly in contexts where the robot serves as a representative or companion rather than simply a tool.
What makes service robots distinct from factory robots is the requirement to interact with the general public, not just trained operators. They must navigate social cues, interpret emotional states, and respond appropriately to unexpected situations. This demands not just physical capability but social intelligence, a quality that remains far easier to describe than to build.
Companion Robots and Androids: Crossing the Uncanny Valley#
The most controversial category of humanoid robots are those designed not to work alongside us but to live with us, to provide companionship, emotional support, or even intimacy. These machines push furthest into human likeness, and they provoke the strongest reactions.
Realbotix Aria, introduced at CES (Consumer Electronics Show) 2025, bills itself explicitly as a companion robot. Aria's lifelike movements are powered by 17 motors located from the neck up, which help the robot mimic mouth and eye movements. The robot can adjust her behavior based on the face she's wearing, thanks to RFID (radio-frequency identification) tags that recognize different facial attachments, allowing Aria to shift her personality to match the head chosen by the user.
Realbotix's CEO Andrew Kiguel stated, "My company was hoping to make robots indistinguishable from humans, which could also tackle the male loneliness epidemic. We are taking it to a different level that nobody else is really doing". Realbotix offers three versions of Aria: a $10,000 bust model with just the head and neck, a modular version for $150,000, and a full-standing model with a rolling base for $175,000.
When videos of Aria went viral, reactions ranged from fascination to revulsion. Some viewers were impressed by the technical achievement. Others found the concept disturbing, questioning the implications of designing robots specifically for emotional and romantic companionship.
Clone Robotics Protoclone takes a radically different approach to human-like design. Rather than prioritizing aesthetic appeal, Clone Robotics has focused on anatomical accuracy. The Protoclone V1 features over 200 degrees of freedom, 1,000 myofibers, and 500 sensors, integrating synthetic systems that mimic human skeletal, muscular, vascular, and nervous functions.
The prototype is designed to mimic the natural human gait using synthetic muscles beneath translucent skin. Currently powered by pneumatics, the prototype is set to transition to hydraulics as its development continues. The design includes synthetic organ systems for skeletal, muscular, vascular, and nervous functions, and even incorporates a cooling system that simulates the way humans sweat with real water.
Clone Robotics plans to release 279 units of its Alpha android series in 2025. The company positions the Protoclone as aimed at home use and everyday challenges, not just research. When Clone's co-founder demonstrated the robot on social media, he noted that "you can stab it with a fork and it will bleed out," a statement that simultaneously showcased the biological accuracy and unsettled observers.
The question these androids raise is not whether we can build machines that look human. Clearly, we can. The question is whether we should, and for what purposes. Companion robots blur the line between tool and entity, between product and relationship. They force us to confront what we value in human connection and whether a sufficiently convincing simulation can satisfy genuine human needs.
Specialized Variants: Robots for Extreme Environments#
Beyond factories, service contexts, and homes, a third category of humanoids is emerging: robots designed for environments where humans cannot easily go or where the work is too dangerous, too remote, or too demanding.
Agricultural Robots will be optimized for harsh outdoor conditions. Fields are humid, muddy, uneven, and subject to extreme temperatures. Agricultural humanoids will need rugged designs, advanced perception systems to navigate complex terrain, and specialized capabilities for tasks like planting, harvesting, and crop monitoring. Companies like John Deere, which already produces autonomous agricultural tractors, are well-positioned to develop humanoids that can operate in these challenging environments, reducing labor costs and environmental impact.
Space Exploration Robots represent perhaps the most ambitious application. Elon Musk has announced plans to send a Tesla Optimus to Mars in 2026 aboard a SpaceX Starship, laying the groundwork for robotic exploration of extraterrestrial environments. The humanoid form allows these robots to operate in environments designed for humans, such as pressurized rovers and habitat modules, while advanced AI enables them to adapt to unforeseen challenges far from Earth.
Apptronik's Apollo, inspired by NASA's Valkyrie, is also being considered for Mars missions and potential deployment on the International Space Station or lunar habitats by the 2030s. These robots will be designed for versatility, performing tasks like building habitats, conducting experiments, or maintaining equipment in conditions where human presence is difficult or impossible.
Construction Robots must handle heavy-duty tasks in unstructured environments. Companies like Persona AI are developing purpose-built humanoids with the dexterity and durability to work in shipyards, energy infrastructure, and construction sites. These robots need to operate in environments filled with obstacles, hazards, and constantly changing conditions. They must be strong enough to manipulate heavy materials yet precise enough to perform delicate assembly work.
The specialized variants reveal a pattern: as humanoids move into more extreme environments, their designs diverge further from general-purpose models. A robot built for Mars will prioritize radiation resistance and minimal maintenance requirements over speed or aesthetic appeal. A construction robot will emphasize strength and durability over the fine motor control needed for household tasks.
What the Variants Reveal#
The emergence of distinct humanoid variants tells us something important about the next decade of robotics. There will not be a single "humanoid robot." There will be families of machines, each optimized for different contexts, each making different tradeoffs between capability, cost, appearance, and function.
The factory robots will remain largely faceless, functional, and focused on strength and reliability. The service robots will experiment with degrees of human likeness, testing where aesthetic appeal aids adoption and where it triggers discomfort. The companion robots will push furthest into the uncanny valley, raising questions about intimacy, loneliness, and the boundaries of human relationships. The specialized robots will prioritize environmental adaptation over everything else, becoming tools for exploration and construction in places too hostile for human workers.
Within a decade, we will likely see further specialization. Home robots may split into cleaning-focused models and caregiving-focused models. Entertainment robots may emerge as a distinct category, designed for theme parks, exhibitions, and interactive experiences. Educational robots could be optimized for teaching and tutoring, with designs that put students at ease.
The most telling pattern is this: companies are not waiting to perfect a single general-purpose humanoid before deploying them. They are shipping specialized variants now, learning from real-world use, and iterating based on data from actual deployments. This pragmatic approach, led by companies in China and the United States, suggests that humanoid robots will arrive in waves, each variant suited to specific roles, rather than as a single revolutionary product that does everything.
We are not building one kind of robot. We are building an ecosystem of machines, each designed to fit into different parts of our world. Factory workhorses, service androids, deep-space explorers. They differ in form, in function, in ambition. But they share the same engineering problems: how to build a body that moves, a hand that grips, a brain that learns, and a battery that lasts. Those problems are what the rest of this section is about.