Don't be fooled by a robot's impressive display of movement and dexterity during onstage demos in controlled environments. Humanoid exploits in the real world can disappoint.
Humanoid robots, one of the more visible examples of physical AI, are becoming more capable and gaining more attention. Demos of robots walking about and performing tasks are getting more impressive, polished and compelling. But what works on stage in controlled venues doesn't always translate to production environments, where reliability, cost and dexterity issues are still unresolved.
In some cases, robotic systems presented as autonomous have relied on human assistance or teleoperation, highlighting how much of the technology is under development. Tesla's Optimus, for example, received a lot of attention when it was first announced in 2021. Later reports made it clear that human involvement was still essential to its operation. This pattern is apparent elsewhere.
Capability vs. reality
The thing about demos is they show peak performance, not repeatability. Humanoids on display often perform predefined tasks under ideal conditions. And while that's exciting to watch and imagine how they can be deployed in business, humanoids aren't functionally ready for what's next.
The expectation gap is driven by a combination of cost, dexterity and precision limitations, with current humanoid robots not yet capable of large-scale, precise operations. "Businesses tend to believe that these robots are plug-and-play, much like a human employee, but they are far from that," said Lian Jye Su, chief analyst at Omdia, a division of Informa TechTarget. "They will need recalibration and reorientation of the actual workflow, which may come at a high cost for most enterprises."
The humanoid form factor itself can further widen the expectation gap, leading potential customers to project capabilities onto systems that aren't there. That's not accidental, said Bill Ray, VP analyst and chief of research at Gartner. Companies design robots to look human, partly because it taps into how people interpret movement and behavior.
"When we see a humanoid robot dancing or running, we subconsciously attribute other skills to it -- jumping, climbing, throwing -- even if it can't actually perform them," Ray explained. "The shape becomes a distraction and a limitation, … which sets unrealistic expectations in buyers." Because humanoid robots are often better at looking impressive than delivering real utility, he added, "[a]ctual use cases are minimal, as the robots are not nearly as functional as they appear."
Businesses subsequently need to shift focus to functionality instead of form, specifically whether a system can be applied across multiple tasks and improve productivity, rather than serve as a more expensive, less efficient substitute for human labor.
"At Gartner we tend to talk about polyfunctional robots … rather than the physical form factor," Ray said. "The way to evaluate such robots is to consider how they can improve the productivity of existing workers, rather than thinking of them replacing workers."
'Mostly works' isn't enough
The drop-off in performance between controlled settings and real-world environments remains one of the bigger hurdles to broader human robot deployment in businesses. A robot that can execute a predefined task repeatedly in ideal conditions is very different from one that can adapt in real time when something changes, fails or simply doesn't match the robot's training data.
"The main gap lies in complex task resolution," Su explained. "The moment they need to resolve tasks they are not well trained for, the system falls apart rapidly."
Businesses need to approach any robotic deployment, including humanoids, as an infrastructure readiness question, not just a robotics decision.
Samuel PasquierVice president of product management, Cisco Systems
Beyond the robot itself, other factors can affect performance. "Network reliability and latency are foundational for robotic automation in real-world environments," said Samuel Pasquier, vice president of product management for industrial IoT networking at Cisco. Systems must run consistently, or they're not usable. If the network isn't stable, problems surface quickly. In production environments, even short disruptions can delay decisions, interrupt operations or stop them altogether.
"Businesses need to approach any robotic deployment, including humanoids, as an infrastructure readiness question, not just a robotics decision," Pasquier said. "That means assessing network reliability, predictable performance, edge compute, bandwidth, mobility and security from the outset."
Robots still struggle with dexterity and adaptability
It's happened many times. A humanoid attempts to demonstrate performance on stage, then stumbles and falls. Although balance and the ability to walk have improved, handling objects is still a challenge.
While the overall form factor of a full-size humanoid is fairly mature, Su said, the bigger limitation is the robot's hands, or end effector, which lack the precision and flexibility needed to handle variability consistently. "I expect this market to be a significant one moving forward," he noted, "with lots of innovations around soft robotics and solid-state sensors."
Those advances are expected to improve how robots interact with the physical world, giving them better feedback and more adaptive control. But for now, the gap between controlled demonstrations and real-world handling remains a key constraint on broader adoption.
Why the math doesn't always work
The cost of humanoid robots isn't just in the hardware; it's also everything around it. In fact, the cost of the robot itself is becoming less of a barrier, with many vendors offering robots-as-a-service models. The challenge lies in integration, workflow redesign and the opportunity cost for the ideal multi-tasking system compared to simpler automation alternatives.
Only by applying the robot to multiple tasks can the additional cost be justified.
Bill RayVP analyst and chief of research, Gartner
Polyfunctional robots tend to be more expensive, meaning the ROI only works if they can be applied across multiple tasks rather than a single use case, according to Ray. "Only by applying the robot to multiple tasks can the additional cost be justified," he said.
That often places humanoids at a disadvantage compared to more specialized systems. A robotic arm designed for a single task or an automated storage and retrieval system in a warehouse can be deployed faster with less integration work and more predictable results. In contrast, humanoids require broader changes to workflows and infrastructure, which adds time, cost and complexity before they deliver value. In many cases, companies are finding it easier and more cost-effective to automate specific tasks than to introduce a general-purpose system that still needs significant adaptation.
Where humanoids are being used, deployments tend to be narrow and tightly controlled. AgilityRobotics' Digit has been deployed in factories and warehouses for specific material-handling tasks. Figure AI's humanoid robots have operated inside a BMW plant for more than two years. And UBTech's Walker S2 is being used and tested in automotive and aerospace environments. These deployments remain relatively contained in scope and visibility, closer to targeted pilots than broad, production-scale rollouts.
In addition to cost, ROI -- or the lack of it -- is another issue. Humanoid adoption is likely to be phased in, with larger manufacturers and logistics companies moving first where the ROI makes more sense, while others wait for the technology to mature, according to Su.
Where humanoids will make economic sense
While overall adoption of humanoid robots is likely to be gradual and selective, Ray sees deployments sooner in healthcare and environments offering plenty of room for humanoids to operate. "Places with space and automated doors," he said. "The dull, dirty and dangerous jobs provide the highest ROI."
Many analysts expect broader adoption to center on more flexible, multipurpose robotic systems rather than strictly humanoid design. Humanoid robots might continue to attract attention, Ray said, but polyfunctional systems are more likely to see meaningful adoption in the near term.
As these systems evolve, infrastructure will play a bigger role in how quickly deployments scale. Connectivity, edge compute and closer alignment between IT and operational technology will factor into how and where these systems are used.
Humanoid robots are improving quickly, but their use remains limited. Getting them to work reliably in places like factories and warehouses is taking longer. They're being asked to operate in environments far less predictable than demos offer.
Liz Hughes is an award-winning editor and writer covering AI and emerging technology and the former editor of AI Business and IoT World Today.
