
Humanoid robots have long fascinated us from a cultural perspective, from Robby the Robot in 1956's sci-fi classic, Forbidden Planet, to C-3PO in Star Wars. As technology has advanced over the decades, the idea of creating a life-size robot bearing the shape of its human creators has gone from the stuff of science fiction, to a reality on the floor of modern factories and warehouses. Yet in many ways, we have yet to reach a stage where humanoid robots have proven that they can do those jobs as quickly, reliably and cost-effectively as the machines already in use today.
"From where I sit, the factory floor still runs on conventional automation," says Dustin Bobrowske, owner of surplus automated equipment buyer and reseller VB Industrial Supply. "That equipment is proven, and when something fails, a maintenance tech can swap a part and have the line back up the same day. That's the bar a humanoid robot has to clear."
According to a January analysis from Gartner, less than 100 companies are expected to progress humanoid robot proofs-of-concept beyond the experimentation phase through 2028, with fewer than 20 aiming to use the technology for supply chain and manufacturing purposes. Instead, the majority of humanoid robot deployments over the next couple years will likely be contained within tightly controlled environments.
Read More: Is This the ‘Year of the Humanoid Robot’?
Today, most humanoid robots currently in use are limited to relatively simple tasks, like moving warehouse totes between two points. A pilot program from BMW also had robots retrieve sheet metal parts for vehicles and position them for welding, while Tesla has spent years developing its "Optimus" robot for repetitive factory tasks.
However, the technology is still severely limited in a number of ways. First and foremost, humanoid robots lack the dexterity, intelligence and adaptability needed for complicated environments that involve things like SKU picking or assembly lines. Humanoids also still cost several times more than task-specific factory robots, despite delivering lower throughput and uptime, Gartner notes.
"If a job is repetitive and doesn’t change much, a robot arm or a co-bot in a cell is going to be quicker, more accurate and easier to look after than a humanoid," says Pete Wood, director for engineering software provider DesignSpark. "Factories have been built around those machines for decades."
A humanoid has far more to account for from a technical perspective as well, Wood adds, given that it has to cope with balancing legs, arms and dozens of motors that allow it to walk, lift objects and manipulate tools. Conversely, a specialized robotic arm on an assembly line can be designed around a narrow set of precise movements, simplifying both its day-to-day operation and longer term maintenance needs.
Perhaps it's no surprise, then, that the market remains small. The Industrial Federation of Robotics estimates that just 7,000 full-size humanoid robots were sold worldwide in 2025, compared to more than 600,000 industrial robots that were installed in factories. And globally, widespread adoption continues to face challenges stemming from safety standards, high maintenance costs, "and the lack of a strong business case, particularly in industrial settings," the IFR said in a September market analysis.
Across the larger robotics industry, few seem to agree on what the best possible use case for humanoid robots would even be. NASA's humanoid "Robonaut" is designed to measure air movement, clean handrails, test switches, and assist with microgravity tasks aboard the International Space Station. China is currently looking to develop humanoids that can perform menial tasks in everyday homes. And German defense conglomerate Rheinmetall is working toward humanoid robots that would be used in combat.
“I think in some areas we are probably putting the cart before the horse,” says Rick Balzano, VP of AI services, safety and robotics with AI training and safety services provider TaskUs. Even so, he adds, a diverse swathe of early deployments could also provide the data that developers need to improve the robots' performance, and train them to work safely alongside people.
"Let's gamble on this," he posits. "Let's collect the data. Let's figure out how we can train these robots to ultimately perform, and then most importantly, interact cohesively with humans."
Gathering enough data remains the biggest gap to close for humanoid robots, Balzano says. Simulations might offer an efficient way to train robots, but eventually developers need real-world experience to uncover situations those training environments fail to capture. Even teaching a robot to pick up a water bottle or an egg requires extensive data to develop the dexterity needed to handle each object.
Ultimately, making humanoid robots viable in a variety of environments — much less a single factory or a warehouse — will require answering a handful of crucial questions. What actually is the optimal use case for a robot with two legs, two arms and human-like hands? And can it perform that job safely and reliably alongside people, while delivering enough value to justify choosing it over existing automation?
For all the appeal of building machines in our own image, manufacturers might just have to simply judge them by how well they keep production moving.
"It's not whether it can do a task in a demo — it's whether a plant can keep it running for 10 years with parts it can actually get," says Bobrowske.


















