Glorious championship victory, robots outperform humans in the half marathon, but the most crucial competition is not in Yizhuang.

```

From 2 hours and 40 minutes to 50 minutes, the leap of a year is just the beginning.

This morning, in front of the finish line at Nanhaizi Park in Beijing Yizhuang, a humanoid robot named "Lightning" rushed through, with a net time of 50 minutes and 26 seconds. It completed 21.0975 kilometers, exactly the same distance as the human runners present today. When the first human runner reached the finish line, "Lightning" had been waiting there for almost 17 minutes.

Just one year ago, on this same track, the best-performing robot—Tiangong Ultra—took 2 hours, 40 minutes, and 42 seconds, with a guide accompanying it along the way.

One year, the same distance, nearly two hours shorter, and this time without anyone following.

What happened today

Today's champion actually experienced a bit of a twist before claiming victory. The robot to cross the line first was not "Lightning," but the remote-controlled team "Jueying Chitu," with a net time of 48 minutes and 19 seconds, nearly two minutes faster than "Lightning." However, the rules for this race stipulate that the completion time for remote teams must be multiplied by a weighted coefficient of 1.2—the reason is simple, remote control and full autonomy are different technical challenges. After weighting, the actual score for "Jueying Chitu" is about 57 minutes, placing it behind Glory's "Lightning."

This rule design is worth discussing. The organizing committee used a coefficient rather than separate tracks for the two robot types, meaning they compete on the same leaderboard, and remote teams must pass a higher threshold to surpass autonomous teams. In other words, the awards for this competition are actually designed for robots that can see the road and make decisions on their own.

Last year, almost all participating teams relied on remote or semi-autonomous solutions; this year, about 40% chose full autonomous navigation. This change in proportion is even more telling than the champion's performance itself.

The change in scale is also noteworthy: 21 robots grew to more than 300, 20 teams became over 100, 26 brands covered the main domestic humanoid robot manufacturers, and international teams from Germany, Brazil, Portugal, and France participated for the first time.

Two races, same route, but this time added the Nanhaizi Park ecological section, introducing more complex terrain, with difficulty not lower than last time. Under these conditions, the champion’s time dropped from 2 hours 40 minutes to 50 minutes, and it was completed with no human intervention. This is the result the industry achieved in one year.

But one point needs to be clarified: the track is a carefully designed known environment, routes are planned ahead, terrain is repeatedly tested, teams had ample time to fine-tune for specific scenarios. Being able to finish in 50 minutes here is real, but it measures capability under certain conditions, not a different type of challenge.

Robots in factories

During the same week as today’s competition, another batch of humanoid robots were operating somewhere completely different: on assembly lines.

Over the past half year, several car factories, battery plants, and parts enterprises in China have gradually introduced humanoid robots into production. According to public reports, CATL's Luoyang plant began trial use of humanoid robots, undertaking transport and assembly support for specific stations; BYD’s joint tests with Zhiyuan robots are underway; SAIC and GAC are deploying robots at their plants as well.

What do these factory robots do? Roughly: pick up a bolt, align it with a hole, tighten it, put it down, and pick up the next one. Repeat thousands of times daily. Workers walk around nearby, there may be oil stains or parts on the floor, parts batches may vary slightly in shape, sometimes anomalous parts appear and robots need to decide how to handle them.

This is a completely different challenge compared to running a marathon. The marathon route is fixed, obstacles are known, rules are prewritten, worst case is a fall then get up and continue. Factories are different—they require robots to work reliably a whole shift in a partly unpredictable environment, even if workers change routines, even if parts shift two centimeters, even if the floor is more slippery today than yesterday.

Current robots in trial use mostly focus on positions with relatively lax accuracy demands but high stability requirements: chemical handling, high-strength repetitive delivery, and assistance near high-temperature environments. These are tasks humans reluctant to do or harmful over long periods. Industry calls these scenarios "3D”—dangerous, dirty, and dull—where robots can gain a foothold first.

The prevailing view in the industry is that 2025 will be the concept validation year for robots in factories, 2026 a small-scale pilot year, and true large-scale deployment will begin at the earliest in 2027 or 2028. The 50-minute track record is a report card, but between it and full-scale factory launch still lies a long road.

Why China is the main stage

The 300+ robots participating today are almost a microcosm of China’s humanoid robot industry: Unitree, Zhiyuan, Tiangong, Fourier, CloudMinds, Glory... 26 brands nationwide. This scale of concentration itself tells something.

From parts to whole machines, China’s presence in this industry far exceeds conventional expectations. Reducers, servo motors, force sensors, linear actuators—these core components of the robot motion system, China’s involvement in global supply chains is about 63%. If the U.S. wants to build the same supply chain, costs are about 2.2 times China’s according to industry estimates. In 2025, global humanoid robot shipments will be about 17,000 units, and Chinese companies will contribute about 14,000 units, over 80% share.

This supply chain density formation is not directly related to humanoid robots themselves—it’s more a by-product of decades of accumulation in lithium batteries, consumer electronics, and new energy vehicles. Most core components for humanoid robots already have mature supplier systems from phones and EVs, now combined into a new product form. China also has another advantage: the world’s largest manufacturing base, meaning the most potential deployment scenarios, and the densest source of real operational data—which is crucial for training the robot's brain.

Of course, this doesn’t mean there’s no competition. Boston Dynamics, Figure AI, 1X Technologies still have distinct technological advantages in high-end models, and Tesla Optimus’ mass production roadmap is one of the most watched variables in the industry. In the second half of this competition, hardware gaps may gradually narrow, and software—the reasoning and generalization capabilities of embodied intelligence models—will ultimately decide the final landscape.

This path, China has already walked once.

The script from ten years ago

In 2012, global annual sales of EVs were about 120,000 units, most of which were Tesla Model S. That year, almost nobody believed EVs could become mainstream in ten years: range anxiety, lack of charging infrastructure, and high costs seemed unsolvable.

By 2022, global EV sales exceeded 10 million units, China contributed about 6 million, with BYD surpassing Tesla in annual sales.

Looking back at the “unsolvable problems” of 2012, you’ll find they were ultimately resolved, not by a single technological breakthrough but by a mutually reinforcing flywheel: mass production decreased costs, lower costs drove demand, expanded demand prompted charging network construction, improved charging networks lowered consumer concerns. Once this flywheel starts, it’s hard to stop externally, and its acceleration is often faster than anyone predicted.

The cost curve for humanoid robots is now on a similar path. In 2024, the unit cost for top models is about $100,000 to $150,000; by 2025, some Chinese manufacturers have pushed this number down to $30,000–50,000; with scaling, by 2027–2028, it may enter the sub-$10,000 zone. This slope of the curve almost overlaps the lithium battery price reduction path of those years.

Rest of World describes the current position of the industry as the “Pre-iPhone moment”—the eve of smartphone birth: hardware is usable, software is catching up, costs aren’t at consumer-friendly levels, but all critical factors are moving at the same time.

Humanoid robots have an accelerator that didn’t exist in the EV era: large language models. Pre-2022 robots used rule trees and hard coding, switching tasks often meant redeveloping. Now, the robot brain can be a general reasoning model, capable of understanding natural language instructions and rapidly adapting to new scenarios, dramatically lowering “onboarding costs” and rapidly expanding general-purpose boundaries.

But this analogy has a boundary not to be ignored. EVs replaced fuel cars—they replaced another machine and didn’t directly affect anyone’s livelihood. The power system changed but driving itself didn’t disappear. Humanoid robots replace humans—factory workers earning wages. This resistance never appeared in EV history, but in the future of humanoid robots, it will inevitably become a serious issue needing careful handling, not an afterthought.

Unsolved obstacles

Today’s performance looks great, but a few things, however impressive, can’t be hidden.

One is the adaptation cost of deployment. Even for robots that finished in 50 minutes today, if moved to a new factory or production line, they need to rebuild environmental models, recalibrate safety boundaries, and replan motion paths. The cost and duration of this process is currently the biggest friction against large-scale promotion. How to make robots like new employees, able to get started in a few days, not needing months of tailored adjustment—this is the hardest core problem all manufacturers are tackling. Whoever solves it first gets the ticket to scale-up.

One is the length of stable operation. A marathon is once-off, finished and done. Factories operate three shifts a day, 300 days a year, robots can’t stop for fatigue. Current pilot data shows some factory robots can work continuously for four to six hours; whether they can reliably endure a full eight-hour shift without malfunction is still unproven.

One is the gap in laws and standards. If a fully autonomous humanoid robot causes injury or damage in the factory, whose responsibility is it? The robot manufacturer, factory management, or software team? Current industrial safety standards are designed for robotic arms and don’t cover bipedal, autonomous robots. The establishment of this standards system lags far behind the technology itself, and this gap will remain a concern for enterprises until real accidents happen.

One more rarely discussed issue: factory workers. Some pilot feedback shows workers initially find robot behavior significantly uncomfortable—it doesn’t move or pace like a person, and path choices make people unsure where it’s going next. This unpredictability can impact collaborative efficiency and even affect workers’ moods. Operational norms and psychological adaptation for human-robot coexistence require time—not something solved automatically by installing robots.

None of these issues are dead ends, but neither can all be fixed in the short term.

Conclusion

Today, "Lightning" finished 21 km, nearly 17 minutes faster than the first-place human runner, all without anyone telling it where to go.

Just one year ago, achieving this required a person to guide alongside.

This speed improvement is real. But the track is a designed environment, the 21 km route preplanned, terrain repeatedly tested, and the whole system serves just this event. Factory floors aren’t like this—every day is new there, parts positions change, floor conditions vary, and workers’ paths don’t follow rules. Robots need to reliably work a whole shift in this open uncertainty, at costs companies are truly willing to pay, with clear rules on whom to turn to if accidents happen.

This industry has delivered a stunning report card today.

But between the beautiful report card and where this industry really wants to go, there’s still a long way to go.

Risk warning and disclaimerThe market involves risks and investment requires caution. This article does not constitute personal investment advice and does not consider the special investment goals, financial status, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article apply to their specific circumstances. Invest accordingly, at your own risk. ```