Humanoid robotics has been living for years on two promises that rarely aligned: looking human and being useful in the real world. Recent announcements and demonstrations, with China as the main protagonist, suggest that this gap is beginning to close. Not because robots have stopped failing — they still fall, overheat, or run out of battery — but because the industry is addressing the details that distinguish a “walking machine” from a presence perceived as human: microexpressions, posture, surface temperature, and movements that no longer look mechanical to the naked eye.
In Shanghai, DroidUp introduced Moya, a humanoid described by the company as “totally biomimetic” and “embodied,” meaning designed not only to carry out commands but to interact socially with a repertoire of signals that humans interpret almost automatically. In demonstration videos, Moya maintains eye contact, nods, smiles, and participates in conversations with subtle gestures. The remarkable part isn’t that it “moves its face,” but the goal: to replicate microexpressions—those small changes in the eyes, cheeks, and corners of the mouth that often reveal attention, empathy, or doubt. These are millisecond-long non-verbal cues, and they are precisely what often doom many robots to the uncanny valley.
DroidUp has also quantified this realism. Moya stands 5 feet 5 inches tall (around 1.65 m) and weighs about 70 pounds (roughly 32 kg), with dimensions and proportions designed to avoid looking like an “industrial robot with a head,” instead resembling a body on a human scale. The company also states that the robot maintains a surface temperature between 89.6°F and 96.8°F (32°C to 36°C)—a detail that can significantly influence perception in customer-facing environments when someone approaches or interacts physically. Another shared detail is a “walking accuracy” of 92% relative to human gait, indicating that the emphasis is not just on avoiding falls but on moving naturally.
Nonetheless, two key elements remain in the shadows. First, DroidUp has not published a complete technical breakdown of hardware, sensors, actuators, or control systems. Second, information about the initial price and its conversion into Western currencies has circulated with some confusion: some reports suggest around 1.2 million Chinese yuan for a market release expected by late 2026, but the exact figure and, importantly, the reference currency, have not been officially confirmed. The implicit message, however, is clear: Moya is positioning itself as a premium system for institutions and businesses, not as a household gadget.
From social realism to survival: Unitree takes its G1 into the extreme cold
While Moya aims to appear “like someone,” Unitree continues to strengthen another major aspect of humanoid robotics: durability. The company demonstrated its G1 executing a long autonomous walk in extreme cold conditions — temperatures cited as low as -47.4°F (-44.1°C) — covering tens of thousands of steps over precise tracks in snow. The purpose of such tests isn’t aesthetic: in cold environments, batteries perform worse, lubricants thicken, some materials become brittle, and fine control of actuators becomes more difficult.
Beyond the demonstration, the G1 is known as a relatively compact humanoid platform tailored for research, education, and development. Its public specs list a weight of around 77 pounds (about 35 kg) and a base configuration of 23 degrees of freedom, with variants that extend capabilities. It’s the kind of platform aiming to become the “standard lab robot”: less polish, more iteration.
IRON and the reality scenario: when a stumble goes viral
The third example illustrates the clash between ambition and reality. XPeng, known for its electric vehicles, has showcased its humanoid IRON as part of a broader strategy in intelligent systems. Recent demonstrations depict a more “human-like” aesthetic — figures around 5 feet 10 inches (about 1.78 m) and roughly 154 pounds (around 70 kg) — with a complex movement architecture, including dozens of joints and dexterous hands. But it also experienced an incident that summarizes the sector’s current state: a public fall that went viral and overshadowed much of the messaging.
Such incidents matter for two reasons. The first is obvious: safety and reliability, especially if the robot operates near people. The second is strategic: mass adoption won’t be achieved with just one spectacular video but through thousands of hours of failure-free operation in real environments. Humanoid robotics is now entering a phase where social tolerance margins are shrinking: when a robot stops being “a demo” and starts to aim for coexistence with the public.
The silent key: joints and control for energy efficiency
If humanoids want to cease being expensive prototypes, there’s a common enemy that rarely makes headlines: energy. Walking, balancing, manipulating objects, and reacting to the environment all require a massive energy budget, and every watt matters when autonomy depends on a battery.
Two lines of advance are gaining prominence here. On one side, Harvard researchers have introduced joint designs inspired by the human knee based on “rolling contact,” with geometries optimized for improved strength and efficiency. In reported tests, this approach nearly corrects misalignments compared to conventional joints and enables significantly stronger grips using the same actuator. On the other side, Westwood Robotics is pushing an equally ambitious goal: enabling a humanoid to work while walking, not just walk and then manipulate. In other words, full-body coordination—maintaining balance and locomotion while performing tasks with the arms, a challenge many robots still face, often requiring them to “stand still” to avoid losing stability.
Therefore, progress is moving along two simultaneous paths: creating more believable robots in human interaction and developing more useful, energy-efficient robots for physical tasks. China is accelerating efforts and increasing demonstration volume, but the global race depends equally on both the “face” and the “knee.”
Frequently Asked Questions
What does it mean for a humanoid robot to be “biomimetic,” and why does it matter for social interaction?
In humanoid robotics, “biomimetic” typically refers to mimicking human traits—movement, proportions, gestures, and social signals. It matters because human communication relies on micro-signals (gaze, posture, minimal expressions) that influence trust and comfort.
What real-world use cases are humanoids like Moya being designed for?
They’re often targeted for environments involving prolonged interaction: customer service, education, healthcare assistance, reception, and guided tasks in offices or shops, where presence and non-verbal communication can be as important as the task itself.
Why is it so difficult for a humanoid to manipulate objects while walking?
Because the robot must maintain dynamic balance while its arm movements shift the center of mass and generate destabilizing forces. It requires full-body control, sensor fusion, and real-time planning.
What currently limits mass adoption of humanoid robots in businesses?
Primarily energy autonomy, reliability (especially near people), total operational and maintenance costs, and the ability to integrate these robots seamlessly into workflows without constant supervision.