Humanoid robots have entered the public consciousness and remain at the center of discussions about the future of work in warehouses and factories alike—but they may not be the ultimate form factor for the warehouse automation initiatives of the future, despite the hype.
A number of prominent companies, like GXO and Amazon, have started testing humanoid robots as part of their operations. The robots, often designed with a biped form factor, typically take on the appearance of the human body and thus, can walk and move in ways that imitate humans.
Related Stories
Still, research and advisory firm Forrester projected that, by the end of 2025, less than 5 percent of the robots deployed in factories and warehouses will walk on two legs. That’s, in part, because the technology is nascent compared with other types of robots already deployed in warehouses at a larger scale. But the prognosis also hints at a greater trend—Forrester noted in its annual Smart Manufacturing & Mobility Predictions report that robots with lower centers of gravity, stronger arms and wheels rather than legs could be the way forward.
Humanoids present a number of challenges, experts noted—they have complex pieces liable to break; roboticists have yet to figure out issues with their battery life; they’re costly and more.
Experts said the artificial intelligence models and algorithms that train these robots receive could be better suited for autonomous mobile robots (AMRs) with different form factors; while standard, biped humanoid robots are impressive to look at, they lack some of the critical use cases they may need to perform at scale.
John Santagate, senior vice president of robotics at Infios, said that biped humanoid robots have seen a great deal of attention, but contends the real value could be in mobile humanoids or other robots on wheels, particularly where warehouse use cases are concerned.
“There’s been a lot of buzz…around biped humanoid robots, which are different than mobile humanoid robots. Bipeds are walking on two feet. The question I always ask someone is, ‘If you’ve got to get somewhere most quickly and most efficiently, are you using your two feet? Are you getting on a bicycle, skateboard, roller skates [or another mode of transportation] that has wheels?’” Santagate told Sourcing Journal.
Other experts agreed with Santagate’s assertion that biped humanoids may not be the most efficient way forward in warehouse robotics. While they acknowledged that, if some issues can be patched, humanoids have potential to disrupt warehouses, they also indicated that adoption at scale is far from a reality today for most adopters.
Jasper Platz, investor at growth fund G2 Venture Partners, said that, in general, companies want to use AMRs in unstructured environments, but today, many can only handle one-off tasks. Humanoids seem to be even more limited in what they can do at scale.
“Where we are on humanoids is definitely still in the demo pilot phase. From what we’ve seen, none of these systems have proven to be scalable commercially at scale,” he said.
Part of the reason for that is that because most companies say they don’t plan to replace human workers with humanoid robots—at least in the immediate near term—humanoids need to be able to work in high-mix, unstructured environments alongside, and sometimes directly with, human warehouse workers.
Jenny Shern, general manager at robotics company NexCobot, said that means that safety remains a major concern in these environments, particularly because human employees lack experience working alongside a robotic form factor that mirrors their own body. Many of the robots already working in warehouses are stationary robotic arms, mobile robots that operate via wheels or other kinds of stationary robots designed for just one function.
“Unlike simpler robots designed for repetitive tasks in controlled environments, humanoids must function in dynamic, human-centered spaces. That requires far more sophisticated AI for perception, behavior and even emotion, alongside advanced control systems,” Shern told Sourcing Journal.
Today, many humanoid robots struggle to handle multiple tasks simultaneously—and to reason or understand context in the ways humans can. Shern said that while humanoid robots have the potential to work on a handful of tasks in the relative near term, myriad questions remain on how to push this form factor to the next level.
“[These robots’] human-like form allows them to work with existing shelving, tools and interfaces, making tasks like stocking, item picking and basic customer assistance possible without major infrastructure changes,” she said. “We expect to see niche deployments within the next three to five years, particularly in settings with high labor turnover. However, widespread adoption will depend on improvements in battery life, safety certification and overall cost efficiency.”
Cost efficiency seems to be a major hold up, both for companies interested in piloting the technology and investors fueling the companies building humanoid technology. Santagate said that, as companies consider the type of automation they implement into their warehouses and operations, they need to better understand whether they’re bleeding unnecessary capital on complicated solutions, when they could be leveraging existing, simpler robotic options with similar underlying technology, powered by AI.
Because of the limited nature of many of today’s humanoids for practical, business-add applications, the cost per task likely remains high. Santagate said he feels it’s more likely that companies adopt biped humanoids in pilots and as holdovers for other technology while they work through their at-large strategy.
“You get to the challenge of the viability of [humanoids]. That robot costs three to four times, annually, the cost of a human operator. So is it economically viable, whether or not it can do the task?” Santagate said. “I think we will continue to see adopters applying it in stop gap measures. I’m less bullish on the [idea that], ‘Oh, there are going to be 500,000 humanoids deployed by 2030.’”
In order to scale up humanoids’ ability to work in the warehouse, Platz said, companies will need massive amounts of data. The issue for many developing the robots is that, for physical AI—in this case humanoid robots—there’s a lack of publicly available, free data to train the robots on. While digitally simulating situations to train robots’ brains on what to do in a similar moment in the real world has helped with this hold up, and will likely continue to do so, Platz said some of the functions warehouse operators would like to see robots taking care of can be tedious to digitally replicate without a real-world example for the robot to learn from.
“Simulating, for example, soft fabrics is really, really hard…because these materials or these garments behave in ways that are very hard to simulate,” he explained.
Romain Moulin, CEO of robotics company Exotec, said he believes that the software and digital simulation behind training humanoid models have shown a great deal of promise over recent years, and noted that many companies—including Exotec—have started to integrate the AI-powered algorithms into robots taking other form factors. Exotec makes Skypod robots, which traverse warehouse shelves easily to help with picking and inventory.
He said one of the most major issues preventing humanoid robots from coming to life en masse for warehouse use cases is that, beyond being technically complex under the hood, their hardware is also complicated. Humanoid robots, because they are designed to mirror parts of the human body, have more axes—or joints—on them than many other form factors, like robotic arms or other AMRs do. Moulin said the addition of extra axes drive up the initial cost of building the robot, make it more difficult to power for long periods of time and even make it more risky for the warehouse.
“The more axes you add to your machine, the more it breaks. Even if it’s super reliable, it’s just axis on axis on axis, so probabilistically it has more chances to break,” he said.
In general, roboticists often try to skinny down the physical complexity of the robots they create, typically in an attempt to cut costs. The idea there, Moulin explained, is that when a customer purchases or leases a robot from a company, they have to pay for the hardware each time, while the software embedded in it comes at a fixed price because it doesn’t require physical materials to be put into use.
That alone could stymie the future starpower of biped humanoids, he said—if warehouse operators adopt robots with other form factors because they boast a lower cost-per-task ratio than their human counterparts, humanoids have the potential to be disrupted before they make a meaningful dent in the market.
“Hardware has a price. Software has no price—meaning, if you do very complex software and you spend hours and hours and millions on R&D, on training AI models, then having 10 customers or having 100 customers is almost the same price,” Moulin told Sourcing Journal. “Investors invest, most of the time, in software, so they are used to that mindset. When they go to hardware, they completely forget [that] each time you deliver [to a] customer, you pay for it. If you don’t deliver the simplest solution in hardware…someone else will come and have a simpler solution.”
