BEIJING — Four months after hosting the world’s first humanoid robot marathon, China is raising the stakes.

This August, Beijing will host the inaugural World Humanoid Robot Sports Games, a multi-event competition featuring sprints, relays, gymnastics, and scenario-based challenges set in mock factories, hospitals, and hotels.

Events will take place at several Olympic venues, including the National Stadium and the National Speed Skating Hall. Organizers say the games are designed to showcase progress in robotics and attract global participation.

Under competition rules, humanoid robots must walk on two legs in athletic events. But in scenario-based contests — like those simulating factory work or hotel service — robots with wheels or other non-bipedal designs will also be allowed.

The announcement comes on the heels of April’s half-marathon in Beijing — the first long-distance endurance race ever attempted by humanoid robots. Of the dozens that entered, only six machines finished. Many collapsed within minutes. Some froze mid-stride, overheated, or lost parts. Others were wheeled off the course by support crews.

The event, meant as a showcase, turned into a reality check. Engineers and executives from the competing teams say the race helped expose how far the field still has to go — and what it will take to get there.

As teams prepare for a new round of challenges in August, Sixth Tone spoke with four startups that competed in April’s marathon: Noetix Robotics, High Torque Robotics, Cyan Robotics, and DroidUp.

They spoke about what held up, what broke down, and what they’re changing as the field moves rapidly forward.

Sixth Tone: Were you satisfied with the result of the marathon?

Tan Bangyu, algorithm engineer at Noetix Robotics: Yes — considering the circumstances, we were. We only had a month to prepare, while some teams had far more time and custom-built robots. Our goal was simply to finish. In the end, both of our entries crossed the line.

One of our robots placed second, finishing in 3 hours and 37 minutes. In our internal tests, we had completed the full 21 kilometers with a single robot in under three and a half hours, so we knew it was possible. But there were surprises — like a wiring issue we discovered just two days before the race. We stayed up all night fixing it.

It wasn’t perfect, but we learned a lot and proved the platform’s potential.

Sixth Tone: Why did your team choose to participate in the marathon?

Bai Zhaoyang, marketing director at Cyan Robotics: Running was always part of our existing roadmap. But the technical path we were developing wasn’t built for long-distance running, and we hadn’t optimized for that kind of stress. But we saw this as a rare chance to test our robot in extreme conditions, and be part of a high-profile event that could show where humanoid robotics really stands.

We treated it like a large-scale experiment. We knew the robot wouldn’t finish — it ran 1.8 kilometers before timing out — but that wasn’t the point. The real value was what we learned about performance under pressure. For example, that our motors could overheat above 80 degrees Celsius and trigger a protection mode. That’s something we might never have caught in standard lab tests.

Sixth Tone: How did the race change your understanding of what robots need to do?

Bai Zhaoyang, Cyan Robotics: It made us rethink what “working” really means for a humanoid robot. In the lab, a robot might walk smoothly and execute all its motions. But in a real, high-pressure environment, that’s not enough.

We realized just how much stress even simple movements can cause over time. Motors overheated, control systems drifted, and minor design flaws became serious risks. After the race, we made upgrades to both our hardware and algorithms. It also helped us better explain those limitations to clients. A polished video can create the illusion that robots are ready for anything. This race made it clear they’re not — yet.

Sixth Tone: A lot of people asked: Why make robots run at all?

Tan Bangyu, Noetix Robotics: Running is just one skill, not the end goal, but it’s a good way to test endurance, balance, and control under stress. It’s like learning to play table tennis: maybe that’s not what the robot was made for, but it still teaches important fundamentals.

It’s also a necessary step. When cars were first invented more than a century ago, people held races to push their limits. Most of those early cars broke down — axles snapped, wheels flew off — but the failures helped engineers figure out what needed to improve. We think humanoid robots are in that same early stage. You can’t skip the messy part.

Zhang Zihao, technical team lead at High Torque Robotics: Right now, we’re only testing the “cerebellum” — gait, balance, and movement. There’s no real brain yet. No cognition, no object recognition, no real-world autonomy. But that doesn’t make it pointless.

Look at the first DARPA Grand Challenge in 2004 (a U.S. government-sponsored competition for self-driving vehicles). Every car failed. Not a single one finished. But that moment pushed the field forward. Today, autonomous driving is a real industry. You have to start with what doesn’t work before you can build what does.

Bai Zhaoyang, Cyan Robotics: The biggest value was in managing public expectations. A lot of people only see robots in polished promotional videos. But this race showed what robots look like under pressure, in the real world. That helps close the gap between perception and reality. That’s essential if the industry is going to grow without overpromising.

Sixth Tone: What’s the biggest technical challenge your team faces now?

Tan Bangyu, Noetix Robotics: The biggest problem is battery life. That’s a huge limitation. We’re working on energy efficiency and longer-lasting batteries, but it’s not something that can be fixed overnight.

The second challenge is the supply chain. Most teams are still working at prototype scale. We’re taking steps towards mass production this year. However, improving industrial chains and reducing costs remain key challenges.

And finally, there’s public awareness. If people don’t trust or accept humanoid robots, none of this technology will scale. That’s why events like the marathon matter, they help build understanding, not just hype.

Sixth Tone: How did the race affect your company’s visibility or product interest?

Tan Bangyu, Noetix Robotics: By noon after the marathon, our boss and colleagues in charge of product sales were overwhelmed with messages. Many were reaching out for potential collaborations. Some headhunters got in touch to help with recruitment, too. It’s been a hectic time.

Bai Zhaoyang, Cyan Robotics: We didn’t finish the race, but the exposure still helped. Media interest increased, and we started getting inquiries about partnerships and applications. For a startup like us, just being part of the event brought more attention than we expected and gave us a way to explain our technology to a broader audience.

Sixth Tone: What’s next for your robots? Where are they actually going to be used?

Tan Bangyu, Noetix Robotics: Right now, our N2 robots are mainly used for education and research. We sell them to schools and labs for secondary development. But we’re working on an upgraded model called E1. It’s taller, more capable, and designed for real-world tasks.

We want to eventually deploy it in homes for light chores or elderly assistance, and also in public-facing roles like reception or interaction in retail settings. That doesn’t require marathon endurance, but it does require stability, autonomy, and safety. We’re building toward that.

Bai Zhaoyang, Cyan Robotics: Our strength lies in our fully humanoid design — from upper limb manipulation, motion control, to interaction capabilities — all are built to mimic human functions comprehensively. We’ll continue upgrading the robot across more technical dimensions and adding new features.

Since it’s a general-purpose humanoid robot, we haven’t restricted its application scenarios. As for commercialization, for now we’re currently seeing the most demand in fields like education and research, as well as exhibition, presentation, and guided tours.

Sixth Tone: What do you see as the most important trends shaping the future of humanoid robotics?

Tan Bangyu, Noetix Robotics: We’re optimistic, but realistic. The robots aren’t ready yet for full independence — but neither were cars or smartphones in their first versions. What matters is steady improvement. We’re seeing that already, and the pace is accelerating.

Zhang Zihao, High Torque Robotics: Three things stand out. First, miniaturization. Smaller robots are safer, more flexible, and more energy efficient — and we saw several strong small-scale performers in the marathon.

Second, brain development. Right now, we’re good at balance and motion, but there’s still no real perception or cognition. Teaching robots to understand the physical world — not just move through it — is the next frontier.

Third, integration. We’re getting better at moving from simulation to real-world deployment, but that bridge still has gaps. And as production scales, costs will drop, which will help expand access and experimentation.

Li Qingdu, founder of DroidUp: This competition was just the beginning. Our technical path was completely new, and we weren’t sure it would hold up — but we proved it could. I think years from now, people will look back at this marathon as the moment when humanoid robots started being tested in real-world conditions. That’s a turning point.

Additional reporting: Xie Annan; editor: Apurva.

(Header image: Noetix Robotics-N2 crosses the finish line during the humanoid robot half marathon in Beijing, April 19, 2025. Pedro Pardo/AFP via VCG)