Across the entire industry, fewer than 3,000 humanoid robots have actually been deployed in factories.

But according to several industry insiders, most of these robots are little more than “mascots” on the production line.

When visitors arrive, they run for two minutes. Once the audience leaves, they are pulled aside and left to gather dust in a corner. On Tesla’s latest earnings call, Elon Musk admitted bluntly: “Optimus is not doing anything genuinely useful in the factory.” Across the Pacific, UBTech, China’s first listed humanoid robot company, has also shown its hand.

Tan Min, UBTech’s chief brand officer, told reporters that Walker S2’s actual working efficiency in tasks such as palletizing and quality inspection is only 30% to 50% of a human worker’s.

Is putting humanoid robots into factories really a false proposition?

Mascots?

News about humanoid robots entering factories has been constant over the past two years.

But most of it has been advance hype around planned deployments. The robots that have actually landed are limited to these cases.

Automakers account for the largest share so far. According to incomplete statistics, BYD already has more than 150 robots on the job.

Xpeng has also placed more than 500 robots in its own factories for trial runs. Geely’s Zeekr has deployed dozens as well. Xiaomi announced two weeks ago that its robots had begun internships on the factory floor, though it did not disclose the number.

Then there are factory deliveries claimed by robot companies themselves. UBTech’s Walker S1/S2 robots are said to have been delivered in the hundreds.

Leju Robotics’ “Kuafu” has also delivered several hundred units to factories including FAW Hongqi. Galbot has deployed robots at CATL’s factories, though the number is unknown. Add all of them together, and the total is likely no more than 3,000 units.

How are these robots actually performing? “I visited several factories in person. Most robots are just using the sites for training, a small number are there to greet visitors, and very few are really on assembly lines,” Luo Yanqing said.

UBTech executive Tan Min was candid: Walker S2 reaches only 30% to 50% of human productivity in specific tasks such as palletizing and quality inspection. In other words, three robots together barely equal one skilled worker.

The head of one robot company put it plainly: “Even when humanoid robots enter production lines today, they cannot be treated as labor. They can only be counted as mascots on the line.” As soon as a leadership inspection or media tour ends, they are powered down and sent back to charging piles to gather dust.

Some in the industry jokingly call this tacit arrangement “tour-guide-style cooperation.” It is happening in China, and it is happening overseas too.

At the end of last year, official reports said Figure AI had entered a BMW factory for testing, with a polished scorecard claiming it had “participated in the production of 30,000 X3 vehicles and handled more than 90,000 parts.”

But when industry insiders dug into it, they found the robot’s job was simply placing metal plates onto fixtures.

That kind of rough handling work could already be done by traditional six-axis robotic arms 20 years ago.

Eleven months later, the battered Figure 02 was “retired” and sent back to headquarters. BMW’s official comment was restrained and telling: “This was an early concept test.” The same script played out at GXO Logistics.

Official marketing said Digit robots moved 100,000 totes. In reality, the robots were just higher-end porters, moving items from point A to point B. Even Musk, on Tesla’s recent earnings call, admitted: “Tesla’s robots are not doing anything genuinely useful in the factory.” That is the industry’s real state: deployed in factories?

It is mostly a marketing spectacle and PR exercise.

Entering Factories, or a False Proposition?

“Humanoid robots entering factories today is definitely a false proposition,” Luo Yanqing said.

Look at the current state of factories: at leading companies, robotic arms can account for 40% to 95% of operations, and many processes are completed entirely by robotic arms.

In factories, robotic arms are the undisputed incumbents. Compared with them, humanoid robots have no real advantage.

First, they are not fast enough. Today’s assembly lines are designed for human workers and traditional industrial robotic arms.

A human worker may take only two seconds to tighten a screw. A traditional robotic arm takes 0.5 seconds. What about a humanoid robot?

“Today’s humanoid robots move very slowly. Asking one to pick up a part, align it and screw it in could take 10 seconds or even longer,” Luo Yanqing said.

Second, they cannot lock onto position precisely enough. Humanoid robots move around, unlike robotic arms that are fixed in place.

“Once a robot moves, it is extremely difficult for it to return 100% to its original position. There will always be a deviation of a few millimeters,” Luo Yanqing said.

Third, their accuracy is not good enough. Factories have extremely demanding requirements for “consistency” and “stability.”

“For robots today, achieving 90% accuracy is already top-tier for the industry,” Luo Yanqing said.

But factories need 100%. A single error can jam an entire assembly line, which is a disaster for the factory.

Right now, the story most often told by capital and the media goes like this: robots can work 24 hours a day. They may cost 300,000 to 500,000 yuan today, but they are a once-and-for-all solution.

A worker costs about 80,000 yuan a year. A robot working around the clock can replace three people, so the economics are excellent. Luo Yanqing said this is pure fantasy.

First, robots cannot work 24 hours a day.

Tesla’s robot can operate continuously for about four hours in practice, while Figure 02 claims five hours.

But in real factory environments, they often last only one to two hours.

Take Agility’s Digit as an example. Its disclosed operating data shows a 2:1 ratio: among three robots, two work while one charges, rotating in shifts to maintain continuous operation.

“A single robot can truly work only about eight hours a day,” Luo Yanqing said.

Second, robots also incur costs for depreciation, maintenance and computing power support.

Luo Yanqing estimated that annual maintenance for one robot is about “150,000 to 200,000 yuan.”

So replacing an 80,000-yuan worker with a 500,000-yuan machine, while paying an extra 150,000 to 200,000 yuan a year in maintenance, clearly violates business logic.

Many factory workers online have complained that robots in a warehouse require a team to repair and monitor them, costing “20 times” as much as human labor.

Luo Yanqing also exposed an uncomfortable truth in the industry: to get humanoid robots into factories, many companies are “forcing demand into existence.” In factories, 80% of repetitive labor has long been handled well by traditional six-axis robotic arms, AGV carts and dedicated machines. They are not only cheaper, but far faster.

Insisting on using a bipedal robot to move boxes is like “asking a college student to push a millstone.”

A Two-Way Rush

If the robots cannot work well and are painfully expensive, why are factories still buying them in large numbers?

The answer has little to do with production efficiency and everything to do with capital-market storytelling. For factories, spending several million yuan at this stage to buy a few robots is not about buying capacity. It is about buying the label of “smart manufacturing.”

Tesla’s Optimus has still not been delivered at scale, but it has done one thing: it made “humanoid robots entering factories” look like a validated story.

BYD, Changan, Geely and other automakers have followed suit, none wanting to be left out. For robot companies, orders can help push up valuations or lift share prices. This is a story of mutual pursuit.

There have already been too many successful examples.

At the end of February 2024, UBTech released a video of its robot undergoing “practical training” at Nio’s Hefei factory. Its main tasks were inspecting car door locks and applying vehicle badges.

Soon after the video was released, UBTech saw a historic surge.

On March 6, 2024, its intraday gain topped 88%, and the share price jumped from an offering price of about HK$90 to a peak of HK$202.

UBTech later announced that its robots had entered production lines at Dongfeng Liuzhou Motor, FAW-Volkswagen and others.

Figure also rode factory entry to a comeback. In January 2024, Figure was still a startup.

But with an agreement to enter BMW’s factory, it secured $675 million in financing a month later, sending its valuation soaring to $2.6 billion.

“The money spent on buying robots will be paid for by capital and the stock market, and may come back hundreds or even thousands of times over,” Luo Yanqing said. It is like “The Emperor’s New Clothes”: everyone in the industry understands it, but no one says it out loud.

“I think it will take at least another 10 years before robots truly enter factories,” he said bluntly. The real humanoid robot era is still far from arriving...